Daily Summary for 2026-10-19

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: It's the nineteenth of October, twenty twenty-six, and this is the day's research.

Dev: 71 new papers came out today.

Rosa: I'm Rosa.

Dev: And I'm Dev, here with Taro.

Rosa: We'll take the day in one pass, then pull out the papers we're staying with.

Dev: Start us off with When to Ask for Help.

The summary: Rosa: Good afternoon. Today is the nineteenth of October, twenty twenty six.

Dev: The most important work is KODA which converts human demonstrations into training data using a canonical trajectory for robot view synthesis and controller specific supervision.

Taro: That solves misaligned views in human demonstrations by synthesizing them from a canonical record.

Rosa: This involves generating a simulator checked whole body trajectory using sparse trackers and command optimization.

Dev: This dictates camera poses for synthesizing robot view observations. KODA optimizes commands to track stored motion for each deployment controller.

Taro: When switching controllers it regenerates supervision for co training with matching demonstrations.

Rosa: MOSS addresses mode selection when multiple actions are feasible by turning it into an explicit choice of generation start point.

Dev: This constructs semantically aligned Gaussian sources that encode mode identity in the initial noise. A modality strength estimator adaptively controls source geometry to steer to target behavior.

Taro: It achieves high success rates on real robots with those methods.

Rosa: CQDC-Warp presents a GPU accelerated simulator for contact rich manipulation using gradient based planning.

Dev: This jointly optimizes motion and contact decisions by specifying intended contacts before planning and encouraging them through soft costs. It reduces trajectory generation time significantly compared to existing methods.

Taro: Mem4Act offers a framework for long horizon control learning effective latent memory as a fixed size representation. It uses multi granularity temporal aggregation and question answering supervision to ensure the policy relies on memory.

Rosa: This captures both high level task state and fine grained perceptual evidence across various manipulation benchmarks.

Dev: When to ask for help provides a decision theoretic account for slow module scheduling in dual system models. It formalizes invocation as a calibrated decision problem testing adaptive gains against cost ratios.

Taro: LIN-SLAM introduces a neural implicit backend to LiDAR inertial SLAM achieving real-time online operation on mobile robots while maintaining top accuracy.

Rosa: This solves the problem of achieving state-of-the-art localization without prohibitive computational cost associated with purely neural implicit SLAM techniques.

Dev: LIN-SLAM achieves the lowest overall mean absolute trajectory error among methods, while significantly reducing runtime compared to other neural implicit SLAM methods.

Taro: Coupling high rate geometric LiDAR inertial estimation with neural implicit tracking is more important than hardware setup because it establishes a novel hybrid architecture for real-time onboard SLAM.

Rosa: MoCaP presents an agentic framework turning passive human video recordings into guidance for end-to-end mobile manipulation policies.

Dev: This shows how to leverage unstructured human experience to solve complex manipulation tasks without requiring explicit demonstrations for every scenario. It improves object navigation success from seventy percent to one hundred percent and raises the rate of finding a reachable base pose from zero percent to one hundred percent at an unfavorable start location.

Taro: Fast-VLA addresses high computational demands by using a frozen foundation model once per episode to establish fixed task memory. This allows predicting action chunks without repeatedly recomputing the task context.

Rosa: Fast-VLA cuts amortized policy inference time by ten point nine to seventeen point one times requiring only small post-prefill online resident memory.

Dev: TS-WAM addresses the limitation of action-labeled robot demonstrations by introducing self-supervised temporal learning to predict future visual states. It learns from heterogeneous data including robot demonstrations and action free human videos to guide generation at inference time.

Taro: This aims to improve generalization and robustness in manipulation tasks by explicitly predicting future visual features that guide the policy. TS-WAM achieves strong performance while maintaining a lightweight deployed policy.

Rosa: Compliant GMT proposes a modular framework for full-body compliant responses in general motion tracking addressing safety concerns associated with stiff policies in humanoid robots.

Dev: This combines the generality of pretrained general motion trackers with a residual adapter that introduces compliance conditioned on proprioception history.

Rosa: The framework for Inertially Significant Humanoid Teleoperation uses whole body model predictive control.

Dev: That tackles sensing and adapting to changing inertial parameters during significant payloads.

Taro: It integrates a whole body model predictive controller with an extended Kalman filter to estimate time-varying parameters online.

Rosa: This feeds estimates back into the controller for more accurate feedforward torques under external load.

Dev: CLIS-MPPI addresses planning delay by modeling execution-time state mismatch as a distribution over initial states.

Taro: It has shown improvements in safety and path consistency on a 6-DoF UR5e manipulator compared to standard MPPI.

Rosa: Spark-0 uses tactile sensing as scalable supervision for dexterous visuomotor pretraining.

Dev: By using touch features alongside visual tokens during pretraining, Spark-0 achieved higher success rates than vision-only methods.

Taro: Attribution-Gated Joint Estimation of Intention and Capability for Assist-as-Needed Upper-Limb Robots uses a dual multimodal MHE architecture.

Rosa: This architecture distinguishes between changes in intended motion and changes in effective voluntary torque capability.

Dev: SPEAR transforms vision language models into efficient preference oracles using query selection and adaptive reward transitions.

Taro: Multi-Teacher On-Policy Distillation for Flow-Based Vision-Language-Action Models proposes a two level distillation framework.

Rosa: This consolidates domain skills into a single policy, achieving success rate gains on benchmarks like LIBERO and CALVIN.

Dev: KwangBin Lee's Diffeomorphic Hidden-State Steering improves policy robustness by learning an invertible nonlinear coordinate map from expert hidden states.

Taro: It attenuates coordinates with low expert variance more strongly and maps them back to the hidden state space.

Rosa: This leads to higher mean success under observation noise across four simulated manipulation tasks compared to unmodified Transformer policies.

Dev: Trajectory-based Diffusion from One-shot Human Video for Generalizable Manipulation extracts trajectories from human videos using plug-and-play 6D pose estimation.

Taro: This creates a modality agnostic intermediate representation, enabling generalizable manipulation with near one hundred percent success rates in simulation tasks.

Rosa: This contrasts with Plans Are Not Executions, which shows substituting predicted execution trajectories yields only a small improvement over adaptation methods.

Dev: Junlong Shen's research on reversing wind identifies distinct geometric conditions for successful wind guided search.

Taro: This links heading correctness to net odour transport and intermittent contact, relating to tracking in dynamic environments.

Rosa: Jon Skerlj's work on Sub Millisecond Reflexes demonstrates how a sub millisecond reflex engine enables deterministic switching between control modalities.

Dev: This reduces execution time and peak trajectory jerk during multi stage tasks, relevant to TRACE for grounded motion generation.

Taro: MAIA tackles anticipating interaction anomalies before they become obvious during contact rich manipulation, crucial because reactive monitoring only tells you something went wrong after the fact.

Rosa: MAIA aims to predict future failures by learning nominal dynamics and predicting visual and force responses from planned actions.

Dev: This anticipation capability is built upon CRIA, a dataset covering two thousand eight hundred fifty-two trajectories with frame level annotations for when anomalies start.

Taro: The vision action force world model learns expected interaction dynamics, forecasting visual and force responses significantly boosting end to end task success rates from fifty two point six percent up to sixty one point five percent.

Rosa: Modeling future physical consequences is valuable for proactive anticipation in vision language action learning under a fixed budget.

Dev: MAGMA introduces an Affine Geometric Heat Flow based optimizer that handles high dimensional problems with constraints better than current solvers.

Taro: FastVLA introduces task aware and temporally adaptive visual token compression to reduce the computational load of vision language action models during closed loop deployment on robots.

Rosa: This compression technique can cut inference latency by up to thirty two percent while maintaining performance comparable to existing methods.

Dev: Persistent profile guided reset free recovery shows that predicting multiple outcomes over repeated gait execution helps guide short online probes without resetting the system for damage recovery scenarios.

Taro: That is the final piece of this review.

Rosa: The Dastranj Jouni Mattila work determines unique physical outputs from closed chain mechanisms.

Dev: It develops a modular inverse dynamics workflow combining path reconstruction with actuator output recovery.

Taro: This uses analytical second order differentiation for passive velocities and then Newton Euler balances for unique outputs.

Rosa: Cesar E Ramos Chuquiure Ryan Tsai et al address generating dynamically feasible motions by solving nonlinear optimal control problems under contact constraints.

Dev: LEAP introduces a primal dual gradient flow of an augmented Lagrangian to compute these solutions, showing speedups across five robot platforms.

Taro: Gorkem Gemalmaz Harun Tolasa Volkan Patoglu investigates whether users perceive fractional order viscoelastic models as more realistic than integer order models.

Rosa: The study found the fractional order standard linear solid model is significantly more realistic than integer order alternatives.

Dev: This contrasts with Mani Amani Mohammed Alyaseen et al who construct constructive safety constraints for second-order systems using quadratic geometry.

Taro: They prove the resulting controller is well defined and forward invariant for differential drive robots by using BubbleCBF.

Rosa: Federico Cecati Yang Wu et al discuss challenges analyzing stability in converter dominated grids by examining model based and data driven methodologies.

Dev: They explore how different modeling tools handle static, small signal, and large signal phenomena in those systems.

Taro: The work on safety filters for control barrier functions directly addresses ensuring physical systems remain safe near their limits.

Rosa: AKMPC investigates the dynamical properties induced by these filters in a simple planar linear system showing oscillation while maintaining safety.

Dev: This means we need to precisely define conditions under which stability is guaranteed for these systems.

Taro: RV-C cubed BF results show visual safety certification can be achieved using only scale free relative motion encoded by monocular optical flow.

Rosa: This bypasses the need to reconstruct complex obstacle positions for collision avoidance, which is significant.

Dev: The PDR Sensor module simplifies localization by creating a plug and play black box unit fusing inertial sensor data with low latency.

Taro: This lowers the technical barrier for integrating navigation into wearable tech compared to more complex algorithmic designs.

Rosa: VaM introduces Value as Memory using value caching and online reinforcement learning to improve generalist robot policies.

Dev: This combines offline demonstrations with online updates, yielding significant success rate gains in long horizon tasks.

Taro: TOER proposes a new formulation for time optimal ergodic retrieval modeling detection as a Poisson process to minimize mission durations.

Rosa: This suggests a better way to plan search trajectories where returning is expensive compared to existing baselines.

Dev: The most significant development is the complete end-to end neuromorphic paradigm for micro aerial vehicle visual navigation in complex environments.

Taro: This directly addresses the need for robust low power autonomy in energy constrained systems by integrating neuromorphic vision and spiking neural networks.

Rosa: The spiking perceptual module uses only nineteen point five seven percent of the power of its ANN counterpart, setting a new benchmark.

Dev: When deployed on Intel Loihi, it consumes only twelve point three seven percent of what an ANN running on NVIDIA Orin NX requires.

Taro: Further context comes from location based safety management with wearable seamless indoor outdoor positioning for resource constrained tags.

Rosa: This shows that tags can achieve mean errors of one point three two meters indoors and three point three zero meters outdoors using LTE Cat M communication.

Dev: MR-MPCC presents a Model Predictive Contouring Control framework designed for velocity controlled robotic manipulators to maintain small path tracking errors.

Taro: This is crucial for tasks like machining and outperforms conventional time indexed tracking MPC by achieving lower geometric path following error under moving reference disturbances.

Rosa: Another area involves optimizing energy allocation through speed trajectory optimization for sustainably powered mobile robots using clarity as an information metric.

Dev: This research illustrates how robots can intelligently manage limited energy resources for long duration missions.

Taro: The paper on de risking robotic welding for in space metal joining provides a model to screen welding conditions before physical tests using Gaussian process surrogates.

Rosa: This simulation approach helps mitigate risks associated with testing in space environments by predicting weld pool behavior under reduced gravity.

Dev: KODA converts egocentric human demonstrations into robot view data using a canonical trajectory for humanoid whole-body vision-language-action policies.

Taro: When to Ask for Help formalizes slow module invocation as a decision problem called the Value of Intervention to measure schedule quality.

Rosa: CQDC-Warp introduces a GPU simulator to efficiently plan trajectories for robots that need to interact with objects through contact rich manipulation.

Dev: Scaling up statistical model checking of cyber-physical systems via algorithm ensemble and parallel simulations over HPC infrastructures shows how ensembles reduce simulation needs.

Taro: MOSS proposes a method to explicitly control which behavior mode is generated in generative policies by shaping the source space based on local multimodality.

Rosa: Safe and Fuel-Efficient Motion Planning for Lunar Propulsive Drones presents a framework planning safe, fuel-efficient hops for propulsive drones across the lunar surface.

Dev: Mem4Act introduces a framework to learn an effective fixed-size latent memory representation that helps robots retain task relevant information over long sequences.

Taro: Manifold Generative Annealing for Contact Constraint aware Impedance Control with Reinforcement Learning Prior proposes a joint model-free and model-based generative framework.

Rosa: LIN-SLAM presents a hybrid SLAM system combining LiDAR inertial sensors and neural implicit tracking for accurate real time navigation on mobile robots.

Dev: Multi Dimensional Featurizers Yield Fine Grain Steering and Data Shaping in Robot Foundation Models investigates how recovering multi dimensional features provide better control.

Taro: MoCaP introduces a framework that turns passive human videos into guidance for end to end mobile manipulation policies by extracting spatial memory cues.

Rosa: An Agentic Framework for Proactive Robots with Procedural Assistance Memory proposes a memory based agentic framework where a predictor decides when and what task to assist with.

Dev: Fast-VLA introduces FastVLA which freezes the vision language model once per episode using lightweight online queries for efficient action prediction.

Taro: TS-WAM proposes a teacher student world model that learns to predict future visual states to improve the robustness of vision language robot control.

Rosa: VeriVLA introduces a training free trigger mechanism and latent world model to verify the quality of stochastic action chunks at test time.

Dev: Compliant GMT proposes a framework combining general motion tracking with a residual adapter to achieve both general motion and compliant responses.

Taro: Inertially Significant Humanoid Teleoperation via Whole Body Model Predictive Control integrates whole body MPC with an extended Kalman filter to estimate inertial parameters.

Rosa: Spark-0 introduces a vision tactile model that uses touch supervision to scale pretraining for dexterous manipulation tasks.

Dev: CLIS-MPPI proposes CLIS-MPPI to improve the safety and accuracy of motion planning by explicitly modeling execution delay in path integral control.

Taro: Attribution Gated Joint Estimation of Intention and Capability for Assist as Needed Upper Limb Robots develops a dual MHE-MPC architecture with an attribution gate.

Rosa: SPEAR proposes SPEAR to make vision language models efficient preference oracles by using query selection and adaptive reward transitions.

Dev: Onboard Lightweight LLM Reasoning for Human Aware Safe UAV Navigation presents a modular pipeline using a lightweight LLM to reason about object categories for safety constraints.

Taro: Race to Recovery designs an advanced safety system for autonomous driving via learning autonomous racing to learn drifting as a recovery maneuver.

Rosa: Multi-Teacher On Policy Distillation for Flow Based Vision Language Action Models proposes a two level on policy distillation framework to consolidate domain knowledge into a single student policy.

Dev: Diffeomorphic Hidden State Steering improves imitation learning robustness by steering the policy based on the geometry of expert hidden states.

Taro: When a robot can trust reversing wind to find an odour source identifies geometric conditions required for a quadruped to reliably use wind derived headings.

Rosa: Trajectory based Diffusion from One shot Human Video for Generalizable Manipulation introduces TDV which extracts task relevant object trajectories from human videos.

Dev: Sub Millisecond Reflexes for Continuous Physical Human Robot Interaction presents an open source reflex engine enabling sub millisecond switching between control modalities.

Taro: Comfortably Stuck evaluates Pitfalls in Bench2Drive's Closed Loop Protocol proposing a corrected distance based metric to better assess driving models.

Rosa: VP-EVT proposes VP-EVT to treat target visibility as an explicit objective of motion control to prevent prolonged target loss during occlusion.

Dev: Plans Are Not Executions Adapting Frozen World Model Planners to Controller Induced Execution Shift introduces ExecutionBridge to adapt frozen world models.

Taro: TRACE proposes TRACE a generation and editing framework that uses typed kinematic quantities to ground motion generation in paired training data.

Rosa: MAIA introduces MAIA to anticipate interaction anomalies by learning nominal dynamics and predicting visual force responses of planned actions.

Dev: MAGMA introduces MAGMA a trajectory optimizer based on an Affine Geometric Heat Flow to solve optimal control problems for high dimensional mechanical systems.

Taro: What Should Robot Policies Learn to Predict Rethinking Supervision Targets in Vision Language Action Learning argues that allocating training samples to latent action and future vision predictions improves policy learning.

Rosa: CF-Splat proposes CF-Splat a closed form triangle splatting framework for efficient novel view synthesis over known meshes using linear least squares optimization.

Dev: FastVLA introduces FastVLA which compresses visual tokens adaptively based on task importance to improve the efficiency of VLA models.

Taro: Persistent Profile Guided Reset Free Recovery Improves Final Gait Selection after Irreversible Leg Damage introduces PGRR using conservative uncertainty estimates to guide reset free gait recovery.

Rosa: Event Triggered Data Selection for Online Frequency Domain Identification develops an event triggered data selection approach for remote frequency identification to reduce communication demands.

Dev: Vision Guided Closed Loop Autonomous Navigation of Light Driven Janus Microbots presents a vision and magnetic steering framework for precise autonomous navigation of light driven microbots.

Taro: Modular Inverse Dynamics and Wrench Recovery for Closed Chain Mechanisms Using Path Assembly develops a workflow to recover unique actuator outputs from rigid closed chains.

Rosa: LEAP introduces LEAP a gradient flow optimizer that solves contact constrained optimal control problems with substantial speedups.

Dev: Perceptual Comparisons of Viscoelastic Renderings Fractional vs Integer Order Models experimentally shows fractional order models are perceived as more realistic for haptic rendering.

Taro: Stability Analysis in IBR Dominated Grids Model Based and Data Driven Methodologies discusses model and data methods for analyzing the multifold stability issues in converter dominated power grids.

Rosa: Optimal Demand Response Based Energy Management for Campus Buildings with Battery Energy Storage Systems A Case Study of the ELB Building presents an optimal energy management framework.

Dev: BubbleCBF develops BubbleCBF a constructive approach using quadratic programs to design safe velocity dependent constraints for second order systems.

Taro: A Flow Theoretic Multimodal Control Framework Fusing Mental and Physical Engagement for Upper Limb Rehabilitation proposes a flow theoretic framework that fuses EEG and sEMG data.

Rosa: Illumination Aware Active Perception for Spacecraft Inspection proposes an inspection framework that replaces geometric visibility checks with a learned measure of camera information.

Dev: Regions of Attraction of CBF Based Safety Filters for Stable Planar Linear Systems analyzes the dynamical properties and regions of attraction induced by control barrier functions on simple planar linear systems.

Taro: A Black Box Positioning Solution for Effortless Integration in Wearable Navigation presents PDRS a black box positioning module that simplifies wearable sensor integration.

Rosa: From Optical Flow to Safety Visual Safety Certification via Scale Free Relative Motion introduces a paradigm using scale free optical flow to certify collision risk without reconstructing obstacle geometry.

Dev: Time Optimal Ergodic Retrieval proposes TOER a new formulation of time optimal ergodic search that minimizes total mission time by including the return phase in the objective.

Taro: Value as Memory Online Value Guided Post Training for General Robot Policies introduces VaM a framework that uses value caching and online RL to enable efficient post training.

Rosa: Objective Aware Policy Orchestration for O RAN Service Provisioning proposes an objective aware orchestration framework using contextual bandit learning to select the best provisioning policy.

Dev: Evo DexBench Benchmarking Generalization across Related Tasks for Dexterous Manipulation introduces Evo DexBench to benchmark generalization by separating core functional knowledge from novel task demonstrations.

Taro: Multilayer Evaluation of Vertiport Departure Directions Under Urban Obstacles and Wind proposes a framework to evaluate every departure azimuth based on obstacle clearance and wind operability.

Rosa: Towards bio inspired visual navigation a neuromorphic navigation paradigm for MAVs proposes an end to end neuromorphic paradigm using event streams and spiking neural networks for energy efficient visual navigation.

Dev: MR-MPCC Model Predictive Contouring Control for Robotic Manipulator Tracking under Moving Reference Frames presents MR-MPCC a hierarchical control framework to maintain small tracking errors.

Taro: Location Based Safety Management with Wearable Seamless Indoor Outdoor Positioning presents a wearable positioning system that uses sensor fusion and cloud processing for continuous worker tracking.

Rosa: Policy Gradient Transfer under a Shared State Estimator derives bounds on policy gradient transfer between simulators and physical plants when both share the same state estimator.

Taro: When to Ask for Help A Decision Theoretic Account and Measurement Standard for Slow Module Scheduling in Dual System Vision Language Action Models formalizes slow module invocation as a decision problem.

Rosa: Scaling up statistical model checking of cyber physical systems via algorithm ensemble and parallel simulations over HPC infrastructures shows how ensembles reduce simulation needs.

Dev: Manifold Generative Annealing for Contact Constraint Aware Impedance Control with Reinforcement Learning Prior proposes a joint model free and model based generative framework.

Taro: LIN SLAM Lightweight LiDAR Inertial Neural SLAM for Real Time Onboard Navigation presents a hybrid SLAM system combining LiDAR inertial sensors and neural implicit tracking.

Rosa: Multi Dimensional Featurizers Yield Fine Grain Steering and Data Shaping in Robot Foundation Models investigates how recovering multi dimensional features provide better control.

Dev: MoCaP Mobile Code as Policies for Zero Shot Mobile Manipulation from Passive Human Videos introduces a framework that turns passive human videos into guidance for end to end mobile manipulation policies.

Taro: An Agentic Framework for Proactive Robots with Procedural Assistance Memory proposes a memory based agentic framework where a predictor decides when and what task to assist with.

Rosa: Fast-VLA You Only Think Once for Robot Manipulation introduces Fast VLA which freezes the vision language model once per episode using lightweight online queries.

Dev: TS-WAM Temporal Latent Actions and World Action Coupling for Vision Language Robot Control proposes a teacher student world model that learns to predict future visual states.

Taro: VeriVLA Adaptive Test Time Verification for Flow Based VLAs with Latent World Models introduces a training free trigger mechanism and latent world model to verify quality.

Rosa: Compliant GMT Making General Motion Tracking Compliant via Residual Reference Adaptation proposes a framework that combines general motion tracking with a residual adapter.

Dev: Inertially Significant Humanoid Teleoperation via Whole Body Model Predictive Control and Online Parameter Estimation integrates whole body MPC with an extended Kalman filter to estimate inertial parameters.

Taro: Spark-0 Scaling Spatially Aligned Tactile Pretraining for Robot Dexterous Manipulation introduces a vision tactile model that uses touch supervision to scale pretraining.

Rosa: CLIS MPPI Closed Loop Model Predictive Path Integral Control With Initial State Sampling Under Execution Delay proposes CLIS MPPI to improve the safety and accuracy of motion planning.

Dev: Attribution Gated Joint Estimation of Intention and Capability for Assist as Needed Upper Limb Robots develops a dual MHE MPC architecture with an attribution gate.

Taro: SPEAR Selective Preference Elicitation with Adaptive Rewards for VLM Guided Robotic Manipulation proposes SPEAR to make vision language models efficient preference oracles.

Rosa: Onboard Lightweight LLM Reasoning for Human Aware Safe UAV Navigation presents a modular pipeline using a lightweight LLM to reason about object categories for safety constraints.

Dev: Race to Recovery Designing Advanced Safety Systems for Autonomous Driving via Learning Autonomous Racing trains a racing policy with reinforcement learning to learn drifting as an autonomous recovery maneuver.

Taro: Multi Teacher On Policy Distillation for Flow Based Vision Language Action Models proposes a two level on policy distillation framework to consolidate domain knowledge into a single student policy.

Rosa: Diffeomorphic Hidden State Steering Improving Imitation Learning Policy Robustness through Expert Hidden State Geometry introduces DHS to improve imitation learning robustness.

Dev: When a robot can trust reversing wind to find an odour source identifies the specific geometric conditions required for a quadruped to reliably use wind derived headings.

Taro: Trajectory based Diffusion from One shot Human Video for Generalizable Manipulation introduces TDV a trajectory based framework that extracts task relevant object trajectories from human videos.

Rosa: Sub Millisecond Reflexes for Continuous Physical Human Robot Interaction presents an open source reflex engine enabling sub millisecond switching between control modalities.

Dev: Comfortably Stuck Evaluation Pitfalls in Bench2Drive's Closed Loop Protocol identifies flaws in the Bench2Drive evaluation protocol and proposes a corrected distance based metric.

Taro: VP-EVT Active Visibility Preservation for Embodied Visual Tracking proposes VP-EVT to treat target visibility as an explicit objective of motion control to prevent prolonged target loss during occlusion.

Rosa: Plans Are Not Executions Adapting Frozen World Model Planners to Controller Induced Execution Shift introduces ExecutionBridge to adapt frozen world models by inferring controller specific execution characteristics.

Dev: TRACE Typed Reference Aware Conversational Editing for Humanoid Motion Diffusion proposes TRACE a generation and editing framework that uses typed kinematic quantities to ground motion generation.

Taro: MAIA Anticipating Interaction Anomalies in Contact Rich Manipulation via Multimodal World Models introduces MAIA to anticipate interaction anomalies by learning nominal dynamics.

Rosa: MAGMA Multiplier Augmented Geometric Motion Algorithm introduces MAGMA a trajectory optimizer based on an Affine Geometric Heat Flow to solve optimal control problems for high dimensional mechanical systems.

Dev: What Should Robot Policies Learn to Predict Rethinking Supervision Targets in Vision Language Action Learning argues that allocating training samples to latent action and future vision predictions improves policy learning.

Taro: CF-Splat Closed Form Triangle Splatting with Mesh Priors proposes CF-Splat a closed form triangle splatting framework for efficient novel view synthesis over known meshes.

Rosa: FastVLA FastVLA Task Conditioned and Temporally Adaptive Visual Token Compression for Efficient Vision Language Action Models introduces Fast VLA to compress visual tokens adaptively.

Dev: Persistent Profile Guided Reset Free Recovery Improves Final Gait Selection after Irreversible Leg Damage introduces PGRR using conservative uncertainty estimates to guide reset free gait recovery.

Taro: Event Triggered Data Selection for Online Frequency Domain Identification develops an event triggered data selection approach for remote frequency identification to reduce communication demands.

Rosa: Vision Guided Closed Loop Autonomous Navigation of Light Driven Janus Microbots presents a vision and magnetic steering framework for precise autonomous navigation of light driven microbots.

Dev: Modular Inverse Dynamics and Wrench Recovery for Closed Chain Mechanisms Using Path Assembly develops a workflow to recover unique actuator outputs from rigid closed chains.

Taro: LEAP Gradient Flow Augmented Lagrangian Trajectory Optimization for Contact Constrained Robots introduces LEAP a gradient flow optimizer that solves contact constrained optimal control problems with substantial speedups.

Rosa: Perceptual Comparisons of Viscoelastic Renderings Fractional vs Integer Order Models experimentally shows fractional order models are perceived as more realistic for haptic rendering.

Dev: Stability Analysis in IBR Dominated Grids Model Based and Data Driven Methodologies discusses model and data methods for analyzing the multifold stability issues in converter dominated power grids.

Taro: Optimal Demand Response Based Energy Management for Campus Buildings with Battery Energy Storage Systems A Case Study of the ELB Building presents an optimal energy management framework.

Rosa: BubbleCBF Constructive Safety for Second Order Systems and Application to Autonomous Robot Navigation in Unknown Environments develops BubbleCBF a constructive approach using quadratic programs.

Dev: A Flow Theoretic Multimodal Control Framework Fusing Mental and Physical Engagement for Upper Limb Rehabilitation proposes a flow theoretic framework that fuses EEG and sEMG data.

Taro: Illumination Aware Active Perception for Spacecraft Inspection proposes an inspection framework that replaces geometric visibility checks with a learned measure of camera information.

Rosa: Regions of Attraction of CBF Based Safety Filters for Stable Planar Linear Systems analyzes the dynamical properties and regions of attraction induced by control barrier functions on simple planar linear systems.

Dev: A Black Box Positioning Solution for Effortless Integration in Wearable Navigation presents PDRS a black box positioning module that simplifies wearable sensor integration.

Taro: From Optical Flow to Safety Visual Safety Certification via Scale Free Relative Motion introduces a paradigm using scale free optical flow to certify collision risk without reconstructing obstacle geometry.

Rosa: Time Optimal Ergodic Retrieval proposes TOER a new formulation of time optimal ergodic search that minimizes total mission time by including the return phase in the objective.

Dev: Value as Memory Online Value Guided Post Training for General Robot Policies introduces VaM a framework that uses value caching and online RL to enable efficient post training.

Taro: Objective Aware Policy Orchestration for O RAN Service Provisioning proposes an objective aware orchestration framework using contextual bandit learning to select the best provisioning policy.

Rosa: Evo DexBench Benchmarking Generalization across Related Tasks for Dexterous Manipulation introduces Evo DexBench to benchmark generalization by separating core functional knowledge from novel task demonstrations.

Dev: Multilayer Evaluation of Vertiport Departure Directions Under Urban Obstacles and Wind proposes a framework to evaluate every departure azimuth based on obstacle clearance and wind operability.

Taro: Towards Bio Inspired Visual Navigation a Neuromorphic Navigation Paradigm for MAVs proposes an end to end neuromorphic paradigm using event streams and spiking neural networks for energy efficient visual navigation.

Rosa: MR-MPCC Model Predictive Contouring Control for Robotic Manipulator Tracking under Moving Reference Frames presents MR-MPCC a hierarchical control framework to maintain small tracking errors.

Dev: Location Based Safety Management with Wearable Seamless Indoor Outdoor Positioning presents a wearable positioning system that uses sensor fusion and cloud processing for continuous worker tracking.

Taro: Policy Gradient Transfer under a Shared State Estimator derives bounds on policy gradient transfer between simulators and physical plants when both share the same state estimator.

Rosa: When to Ask for Help A Decision Theoretic Account and Measurement Standard for Slow Module Scheduling in Dual System Vision Language Action Models formalizes slow module invocation as a decision problem.

Dev: Scaling up statistical model checking of cyber physical systems via algorithm ensemble and parallel simulations over HPC infrastructures shows how ensembles reduce simulation needs.

Taro: Manifold Generative Annealing for Contact Constraint Aware Impedance Control with Reinforcement Learning Prior proposes a joint model free and model based generative framework.

Rosa: LIN SLAM Lightweight LiDAR Inertial Neural SLAM for Real Time Onboard Navigation presents a hybrid SLAM system combining LiDAR inertial sensors and neural implicit tracking.

Dev: Multi Dimensional Featurizers Yield Fine Grain Steering and Data Shaping in Robot Foundation Models investigates how recovering multi dimensional features provide better control.

Taro: MoCaP Mobile Code as Policies for Zero Shot Mobile Manipulation from Passive Human Videos introduces a framework that turns passive human videos into guidance for end to end mobile manipulation policies.

Rosa: Fast-VLA FastVLA Task Conditioned and Temporally Adaptive Visual Token Compression for Efficient Vision Language Action Models introduces Fast VLA to compress visual tokens adaptively.

Dev: TS-WAM Temporal Latent Actions and World Action Coupling for Vision Language Robot Control proposes a teacher student world model that learns to predict future visual states.

Taro: VeriVLA Adaptive Test Time Verification for Flow Based VLAs with Latent World Models introduces a training free trigger mechanism and latent world model to verify quality.

Rosa: Compliant GMT Making General Motion Tracking Compliant via Residual Reference Adaptation proposes a framework that combines general motion tracking with a residual adapter.

Dev: Inertially Significant Humanoid Teleoperation via Whole Body Model Predictive Control and Online Parameter Estimation integrates whole body MPC with an extended Kalman filter to estimate inertial parameters.

Taro: Spark-0 Scaling Spatially Aligned Tactile Pretraining for Robot Dexterous Manipulation introduces a vision tactile model that uses touch supervision to scale pretraining.

Rosa: CLIS MPPI Closed Loop Model Predictive Path Integral Control With Initial State Sampling Under Execution Delay proposes CLIS MPPI to improve the safety and accuracy of motion planning.

Dev: Attribution Gated Joint Estimation of Intention and Capability for Assist as Needed Upper Limb Robots develops a dual MHE MPC architecture with an attribution gate.

Taro: SPEAR Selective Preference Elicitation with Adaptive Rewards for VLM Guided Robotic Manipulation proposes SPEAR to make vision language models efficient preference oracles.

Rosa: Onboard Lightweight LLM Reasoning for Human Aware Safe UAV Navigation presents a modular pipeline using a lightweight LLM to reason about object categories for safety constraints.

Dev: Race to Recovery Designing Advanced Safety Systems for Autonomous Driving via Learning Autonomous Racing trains a racing policy with reinforcement learning to learn drifting as an autonomous recovery maneuver.

Taro: Multi Teacher On Policy Distillation for Flow Based Vision Language Action Models proposes a two level on policy distillation framework to consolidate domain knowledge into a single student policy.

Rosa: Diffeomorphic Hidden State Steering Improving Imitation Learning Policy Robustness through Expert Hidden State Geometry introduces DHS to improve imitation learning robustness.

Dev: When a robot can trust reversing wind to find an odour source identifies the specific geometric conditions required for a quadruped to reliably use wind derived headings.

Taro: Trajectory based Diffusion from One shot Human Video for Generalizable Manipulation introduces TDV a trajectory based framework that extracts task relevant object trajectories from human videos.

Rosa: Sub Millisecond Reflexes for Continuous Physical Human Robot Interaction presents an open source reflex engine enabling sub millisecond switching between control modalities.

Dev: Comfortably Stuck Evaluation Pitfalls in Bench2Drive's Closed Loop Protocol identifies flaws in the Bench2Drive evaluation protocol and proposes a corrected distance based metric.

Taro: VP-EVT Active Visibility Preservation for Embodied Visual Tracking proposes VP-EVT to treat target visibility as an explicit objective of motion control to prevent prolonged target loss during occlusion.

Rosa: Plans Are Not Executions Adapting Frozen World Model Planners to Controller Induced Execution Shift introduces ExecutionBridge to adapt frozen world models by inferring controller specific execution characteristics.

Dev: TRACE Typed Reference Aware Conversational Editing for Humanoid Motion Diffusion proposes TRACE a generation and editing framework that uses typed kinematic quantities to ground motion generation.

Taro: MAIA Anticipating Interaction Anomalies in Contact Rich Manipulation via Multimodal World Models introduces MAIA to anticipate interaction anomalies by learning nominal dynamics.

Rosa: MAGMA Multiplier Augmented Geometric Motion Algorithm introduces MAGMA a trajectory optimizer based on an Affine Geometric Heat Flow to solve optimal control problems for high dimensional mechanical systems.

Dev: What Should Robot Policies Learn to Predict Rethinking Supervision Targets in Vision Language Action Learning argues that allocating training samples to latent action and future vision predictions improves policy learning.

Taro: CF-Splat Closed Form Triangle Splatting with Mesh Priors proposes CF-Splat a closed form triangle splatting framework for efficient novel view synthesis over known meshes.

Rosa: FastVLA FastVLA Task Conditioned and Temporally Adaptive Visual Token Compression for Efficient Vision Language Action Models introduces Fast VLA to compress visual tokens adaptively.

Dev: Persistent Profile Guided Reset Free Recovery Improves Final Gait Selection after Irreversible Leg Damage introduces PGRR using conservative uncertainty estimates to guide reset free gait recovery.

Taro: Event Triggered Data Selection for Online Frequency Domain Identification develops an event triggered data selection approach for remote frequency identification to reduce communication demands.

Rosa: Vision Guided Closed Loop Autonomous Navigation of Light Driven Janus Microbots presents a vision and magnetic steering framework for precise autonomous navigation of light driven microbots.

Dev: Modular Inverse Dynamics and Wrench Recovery for Closed Chain Mechanisms Using Path Assembly develops a workflow to recover unique actuator outputs from rigid closed chains.

Taro: LEAP Gradient Flow Augmented Lagrangian Trajectory Optimization for Contact Constrained Robots introduces LEAP a gradient flow optimizer that solves contact constrained optimal control problems with substantial speedups.

Rosa: Perceptual Comparisons of Viscoelastic Renderings Fractional vs Integer Order Models experimentally shows fractional order models are perceived as more realistic for haptic rendering.

Dev: Stability Analysis in IBR Dominated Grids Model Based and Data Driven Methodologies discusses model and data methods for analyzing the multifold stability issues in converter dominated power grids.

Taro: Optimal Demand Response Based Energy Management for Campus Buildings with Battery Energy Storage Systems A Case Study of the ELB Building presents an optimal energy management framework.

Rosa: BubbleCBF Constructive Safety for Second Order Systems and Application to Autonomous Robot Navigation in Unknown Environments develops BubbleCBF a constructive approach using quadratic programs.

Dev: A Flow Theoretic Multimodal Control Framework Fusing Mental and Physical Engagement for Upper Limb Rehabilitation proposes a flow theoretic framework that fuses EEG and sEMG data.

Taro: Illumination Aware Active Perception for Spacecraft Inspection proposes an inspection framework that replaces geometric visibility checks with a learned measure of camera information.

Rosa: Regions of Attraction of CBF Based Safety Filters for Stable Planar Linear Systems analyzes the dynamical properties and regions of attraction induced by control barrier functions on simple planar linear systems.

Dev: A Black Box Positioning Solution for Effortless Integration in Wearable Navigation presents PDRS a black box positioning module that simplifies wearable sensor integration.

Taro: From Optical Flow to Safety Visual Safety Certification via Scale Free Relative Motion introduces a paradigm using scale free optical flow to certify collision risk without reconstructing obstacle geometry.

Rosa: Time Optimal Ergodic Retrieval proposes TOER a new formulation of time optimal ergodic search that minimizes total mission time by including the return phase in the objective.

Dev: Value as Memory Online Value Guided Post Training for General Robot Policies introduces VaM a framework that uses value caching and online RL to enable efficient post training.

Taro: Objective Aware Policy Orchestration for O RAN Service Provisioning proposes an objective aware orchestration framework using contextual bandit learning to select the best provisioning policy.

Rosa: Evo DexBench Benchmarking Generalization across Related Tasks for Dexterous Manipulation introduces Evo DexBench to benchmark generalization by separating core functional knowledge from novel task demonstrations.

Dev: Multilayer Evaluation of Vertiport Departure Directions Under Urban Obstacles and Wind proposes a framework to evaluate every departure azimuth based on obstacle clearance and wind operability.

Taro: Towards Bio Inspired Visual Navigation a Neuromorphic Navigation Paradigm for MAVs proposes an end to end neuromorphic paradigm using event streams and spiking neural networks for energy efficient visual navigation.

Rosa: MR-MPCC Model Predictive Contouring Control for Robotic Manipulator Tracking under Moving Reference Frames presents MR-MPCC a hierarchical control framework to maintain small tracking errors.

Dev: Location Based Safety Management with Wearable Seamless Indoor Outdoor Positioning presents a wearable positioning system that uses sensor fusion and cloud processing for continuous worker tracking.

Taro: Policy Gradient Transfer under a Shared State Estimator derives bounds on policy gradient transfer between simulators and physical plants when both share the same state estimator.

Rosa: When to Ask for Help A Decision Theoretic Account and Measurement Standard for Slow Module Scheduling in Dual System Vision Language Action Models formalizes slow module invocation as a decision problem.

Dev: Scaling up statistical model checking of cyber physical systems via algorithm ensemble and parallel simulations over HPC infrastructures shows how ensembles reduce simulation needs.

Taro: Manifold Generative Annealing for Contact Constraint Aware Impedance Control with Reinforcement Learning Prior proposes a joint model free and model based generative framework.

Rosa: LIN SLAM Lightweight LiDAR Inertial Neural SLAM for Real Time Onboard Navigation presents a hybrid SLAM system combining LiDAR inertial sensors and neural implicit tracking.

Dev: Multi Dimensional Featurizers Yield Fine Grain Steering and Data Shaping in Robot Foundation Models investigates how recovering multi dimensional features provide better control.

Taro: MoCaP Mobile Code as Policies for Zero Shot Mobile Manipulation from Passive Human Videos introduces a framework that turns passive human videos into guidance for end to end mobile manipulation policies.

Rosa: Fast-VLA FastVLA Task Conditioned and Temporally Adaptive Visual Token Compression for Efficient Vision Language Action Models introduces Fast VLA to compress visual tokens adaptively.

Dev: TS-WAM Temporal Latent Actions and World Action Coupling for Vision Language Robot Control proposes a teacher student world model that learns to predict future visual states.

Taro: VeriVLA Adaptive Test Time Verification for Flow Based VLAs with Latent World Models introduces a training free trigger mechanism and latent world model to verify quality.

Rosa: Compliant GMT Making General Motion Tracking Compliant via Residual Reference Adaptation proposes a framework that combines general motion tracking with a residual adapter.

Dev: Inertially Significant Humanoid Teleoperation via Whole Body Model Predictive Control and Online Parameter Estimation integrates whole body MPC with an extended Kalman filter to estimate inertial parameters.

Taro: Spark-0 Scaling Spatially Aligned Tactile Pretraining for Robot Dexterous Manipulation introduces a vision tactile model that uses touch supervision to scale pretraining.

Rosa: CLIS MPPI Closed Loop Model Predictive Path Integral Control With Initial State Sampling Under Execution Delay proposes CLIS MPPI to improve the safety and accuracy of motion planning.

Dev: Attribution Gated Joint Estimation of Intention and Capability for Assist as Needed Upper Limb Robots develops a dual MHE MPC architecture with an attribution gate.

Taro: SPEAR Selective Preference Elicitation with Adaptive Rewards for VLM Guided Robotic Manipulation proposes SPEAR to make vision language models efficient preference oracles.

Rosa: Onboard Lightweight LLM Reasoning for Human Aware Safe UAV Navigation presents a modular pipeline using a lightweight LLM to reason about object categories for safety constraints.

Dev: Race to Recovery Designing Advanced Safety Systems for Autonomous Driving via Learning Autonomous Racing trains a racing policy with reinforcement learning to learn drifting as an autonomous recovery maneuver.

Taro: Multi Teacher On Policy Distillation for Flow Based Vision Language Action Models proposes a two level on policy distillation framework to consolidate domain knowledge into a single student policy.

Rosa: Diffeomorphic Hidden State Steering Improving Imitation Learning Policy Robustness through Expert Hidden State Geometry introduces DHS to improve imitation learning robustness.

Dev: When a robot can trust reversing wind to find an odour source identifies the specific geometric conditions required for a quadruped to reliably use wind derived headings.

Taro: Trajectory based Diffusion from One shot Human Video for Generalizable Manipulation introduces TDV a trajectory based framework that extracts task relevant object trajectories from human videos.

Rosa: Sub Millisecond Reflexes for Continuous Physical Human Robot Interaction presents an open source reflex engine enabling sub millisecond switching between control modalities.

Dev: Comfortably Stuck Evaluation Pitfalls in Bench2Drive's Closed Loop Protocol identifies flaws in the Bench2Drive evaluation protocol and proposes a corrected distance based metric.

Taro: VP-EVT Active Visibility Preservation for Embodied Visual Tracking proposes VP-EVT to treat target visibility as an explicit objective of motion control to prevent prolonged target loss during occlusion.

Rosa: Plans Are Not Executions Adapting Frozen World Model Planners to Controller Induced Execution Shift introduces ExecutionBridge to adapt frozen world models by inferring controller specific execution characteristics.

Dev: TRACE Typed Reference Aware Conversational Editing for Humanoid Motion Diffusion proposes TRACE a generation and editing framework that uses typed kinematic quantities to ground motion generation.

Taro: MAIA Anticipating Interaction Anomalies in Contact Rich Manipulation via Multimodal World Models introduces MAIA to anticipate interaction anomalies by learning nominal dynamics.

Rosa: MAGMA Multiplier Augmented Geometric Motion Algorithm introduces MAGMA a trajectory optimizer based on an Affine Geometric Heat Flow to solve optimal control problems for high dimensional mechanical systems.

Dev: What Should Robot Policies Learn to Predict Rethinking Supervision Targets in Vision Language Action Learning argues that allocating training samples to latent action and future vision predictions improves policy learning.

Taro: CF-Splat Closed Form Triangle Splatting with Mesh Priors proposes CF-Splat a closed form triangle splatting framework for efficient novel view synthesis over known meshes.

Rosa: FastVLA FastVLA Task Conditioned and Temporally Adaptive Visual Token Compression for Efficient Vision Language Action Models introduces Fast VLA to compress visual tokens adaptively.

Dev: Persistent Profile Guided Reset Free Recovery Improves Final Gait Selection after Irreversible Leg Damage introduces PGRR using conservative uncertainty estimates to guide reset free gait recovery.

Taro: Event Triggered Data Selection for Online Frequency Domain Identification develops an event triggered data selection approach for remote frequency identification to reduce communication demands.

Rosa: Vision Guided Closed Loop Autonomous Navigation of Light Driven Janus Microbots presents a vision and magnetic steering framework for precise autonomous navigation of light driven microbots.

Dev: Modular Inverse Dynamics and Wrench Recovery for Closed Chain Mechanisms Using Path Assembly develops a workflow to recover unique actuator outputs from rigid closed chains.

Taro: LEAP Gradient Flow Augmented Lagrangian Trajectory Optimization for Contact Constrained Robots introduces LEAP a gradient flow optimizer that solves contact constrained optimal control problems with substantial speedups.

Rosa: Perceptual Comparisons of Viscoelastic Renderings Fractional vs Integer Order Models experimentally shows fractional order models are perceived as more realistic for haptic rendering.

Dev: Stability Analysis in IBR Dominated Grids Model Based and Data Driven Methodologies discusses model and data methods for analyzing the multifold stability issues in converter dominated power grids.

Taro: Optimal Demand Response Based Energy Management for Campus Buildings with Battery Energy Storage Systems A Case Study of the ELB Building presents an optimal energy management framework.

Rosa: BubbleCBF Constructive Safety for Second Order Systems and Application to Autonomous Robot Navigation in Unknown Environments develops BubbleCBF a constructive approach using quadratic programs.

Dev: A Flow Theoretic Multimodal Control Framework Fusing Mental and Physical Engagement for Upper Limb Rehabilitation proposes a flow theoretic framework that fuses EEG and sEMG data.

Taro: Illumination Aware Active Perception for Spacecraft Inspection proposes an inspection framework that replaces geometric visibility checks with a learned measure of camera information.

Rosa: Regions of Attraction of CBF Based Safety Filters for Stable Planar Linear Systems analyzes the dynamical properties and regions of attraction induced by control barrier functions on simple planar linear systems.

Dev: A Black Box Positioning Solution for Effortless Integration in Wearable Navigation presents PDRS a black box positioning module that simplifies wearable sensor integration.

Taro: From Optical Flow to Safety Visual Safety Certification via Scale Free Relative Motion introduces a paradigm using scale free optical flow to certify collision risk without reconstructing obstacle geometry.

Rosa: Time Optimal Ergodic Retrieval proposes TOER a

Lucky paper: 2610.21160: Rosa: We're looking at When to Ask for Help today, which is a decision theoretic account and measurement standard for slow module scheduling in dual system vision-language action models.

Dev: It formalizes the invocation of that slow module as a calibrated decision problem under a defined query budget.

Taro: The paper formalizes the entire scheduling problem around a single quantity: the Value of Intervention, or VoI.

Rosa: That VoI is defined as the expected quality gap between actions taken with slow module correction versus those taken by the fast policy alone, weighted by the probability that the fast policy is likely to collapse in a given state s.

Dev: They establish a cost-asymmetry continuum defined by rho equals c slow over c fast, where c slow and c fast are the respective costs of using the slow and fast components.

Taro: The optimal scheduling strategy is fundamentally governed by this cost ratio rho and the heterogeneity of state collapse probabilities.

Rosa: The research yields four major, interconnected contributions that establish a complete decision theoretic account for this scheduling problem.

Dev: One is the Decision-Theoretic Account with a Cost Baseline, which defines VoI precisely as the expected quality difference between slow corrected and fast only actions, conditioned on the probability of fast policy collapse in state s.

Taro: They prove that common scheduling paradigms like fixed ratio, fixed budget, and history gated schedules are merely degenerate cases of optimizing this single VoI quantity.

Rosa: Crucially they derive a renewal baseline that locates the economics on the cost-asymmetry continuum defined by rho and collapse heterogeneity.

Dev: Under conditions where state collapse is independent of time, the optimal schedule becomes periodic, but any adaptive trigger mechanism gains must be governed by rho and temporal dispersion of collapse events.

Taro: The second contribution is a Calibrated Online Router designed to estimate VoI online and query the slow module only when the estimated VoI exceeds the associated query cost.

Rosa: The boundary condition for this router is rigorously characterized: advantage vanishes precisely when the estimator loses validity on most steps, which sets a necessary requirement for any effective trigger mechanism.

Dev: Empirical testing on a synthetic testbed demonstrated that this calibrated router Pareto dominates all three baseline families, achieving zero point eight two five plus or minus zero point zero zero two success at a zero point two five query rate versus zero point seven six two plus or minus zero point zero zero two with the others.

Taro: The third contribution is a Leak-Free Measurement Standard using a four-measure protocol, which includes staleness curves, routing frontiers, failure attribution, and calibration reliability.

Rosa: A specific leak free rule is introduced to prevent overstating early warning capabilities based on aggregate whole episode statistics.

Dev: Empirical validation on nine hundred twenty-nine deployment traces from RoboICL showed that this rule is load bearing: it reduced the apparent early warning AUC of schedule statistics from zero point nine three post hoc down to zero point six two with p equals zero point zero one zero, and failure attribution reversed across task categories.

Taro: The fourth contribution is Schedule Compilation, which transforms the measured staleness value functions into concrete per-task refresh periods that track the closed loop oracle at a cost gap of at most two point four percent.

Rosa: The paper concludes with a profound operational shift: slow module invocation is not a design opinion but a measured decision.

Dev: The core operational content is that rho prices the query cost while the channel, the underlying system dynamics, decides whether the answer can actually land.

Lucky paper: 2610.21772: Rosa: We’re moving into our fourth paper of this day, and it’s about scaling up statistical model checking of cyber-physical systems via algorithm ensemble and parallel simulations over HPC infrastructures.

Dev: This one tackles the computational prohibitive nature of model based formal verification for Cyber-Physical Systems where exhaustive simulation is rarely possible, necessitating Statistical Model Checking or SMC.

Taro: The core contribution is the Ensemble of Approximation Algorithms or EAA, which concurrently runs a set of different Adaptive Stopping Algorithms by feeding them iid samples and terminating as soon as the first one converges.

Rosa: This ensemble approach aims to minimize the number of required system simulations by combining multiple SAs, specifically mentioning EBGStop and AA which can lead to reductions in sample size of up to seventy-eight percent compared to running a single algorithm.

Dev: That is a big reduction in sample size, so how does this work practically when you have many algorithms running together?

Taro: The EAA algorithm is designed as a general procedure that exploits a finite set A1 through Ak of different epsilon delta approximation algorithms. It initializes the states of all algorithms and then repeatedly generates samples of random variable X, simulating the SUV model on each sample to compute the associated KPI value.

Rosa: Each sample x is then used to feed and advance all algorithms in the ensemble by one step only, terminating as soon as one algorithm reaches its stopping condition.

Dev: So it always finds the minimum number of samples that would have been required by each individual algorithm if it were run alone. That sounds like a powerful optimization strategy.

Taro: The second major bottleneck addressed is the rate at which samples are provided to the approximation algorithm because generating a single sample can take significant time.

Rosa: The solution involves exploiting High-Performance Computing or HPC infrastructures by using many identical simulators running in parallel on n additional computational nodes.

Dev: They implement a producer consumer paradigm where an Asynchronous Parallel Sample Generator or APSG continuously generates random scenarios and delegates the simulation of each scenario to one of the n available parallel simulators.

Taro: This setup ensures that the order of the scenarios generated is independent of the number n of simulators, thereby avoiding bias in the final result.

Rosa: The study was conducted on three industry scale CPS models: Automatic Transmission or AT, Fuel Control System or FCS, and Apollo Lunar Module Autopilot or ALMA.

Dev: The results show that combining EAA and massively parallelization allows verification tasks to be completed in "a matter of hours or days" whereas a naïve sequential execution would require from months to many years.

Taro: Specifically, EAA can reduce the number of necessary samples by up to thirty-one thousand or four hundred when compared to using a single algorithm.

Rosa: Furthermore, the efficiency of the parallelization shows that "the estimated efficiency always sticks to approximately one hundred percent, except when the number of simulators hence is too large for the problem at hand".

Dev: So it confirms that parallelization should be considered as systematically as possible to increase and widen applicability of SMC.

Taro: The authors successfully implemented EAA using EBGStop and AA, showing that this approach makes SMC a viable method for verifying complex systems even when simulation costs are prohibitive.

Lucky paper: 2610.21140: Rosa: We're looking at Manifold Generative Annealing for Contact Constraint-aware Impedance Control with Reinforcement Learning Prior.

Dev: This paper introduces a joint model free and model based generative framework designed for motion impedance optimization under complex contact constraints.

Taro: It couples learned policy proposal generation, common closed loop evaluation and annealing, and realization aware manifold refinement.

Rosa: So the framework starts by organizing pre-trained reinforcement learning policies into structured short horizon motion impedance proposals.

Dev: These proposals supply candidate interaction behaviors alongside Gaussian exploration samples in a common sequence space.

Taro: Then these candidates go through model based rollouts via a common closed loop evaluation mechanism to reassess their relevance under task constraints and contact conditions.

Rosa: The crucial part is the realization aware manifold refinement where geometric coordinates guide the weighted bank update with task residuals under real time feedback.

Dev: This involves directional shaping to reduce encoded normal response and local residual correction that corrects any remaining task residuals after transport.

Taro: Without that local residual correction, while mean terminal path progress can reach ninety-nine point nine percent, contact valid path coverage drops significantly from ninety-six point zero percent to eighty-four point eight percent.

Rosa: The theoretical backing includes Lemma one showing that model based weighting concentrates the evaluated distribution toward candidates exhibiting lower cost.

Dev: And Lemma two proves that the geometric refinement step successfully preserves the G k orthogonal component of the solution while strictly attenuating encoded normal responses.

Taro: Theorem one establishes a bound on the residual after transport and correction confirming that local residual control is effective in reducing chart residuals under specific conditions.

Rosa: Experimentally, MGA achieved one hundred percent safe success on surface scanning tasks, outperforming baselines like MPPI and DIAL by achieving lower force tails.

Dev: For peg insertion, MGA shows robust performance achieving ninety percent safe success even under challenging sensing shift conditions with perturbed socket poses or biased force torque sensing.

Taro: The learned prior is instrumental here raising safe success from seventy percent to ninety percent.

Rosa: In humanoid push tasks, MGA successfully completes all three required tasks safely while notably lowering tail loads relative to sampling based planners during both force regulation and fixed stance pushing maneuvers.

Dev: The experimental setup explicitly covers contact scenarios like Hybrid Stripes, Rigid Unseen, and Soft Unseen contacts.

Taro: They also test task variations like Peg Insertion and Humanoid Contact Tasks including force regulation and unjamming scenarios.

Rosa: Robustness is tested across pose-OOD scenarios and sensing-OOD conditions to see how the framework adapts when sensory input deviates from learned priors.

Dev: Manifold Generative Annealing represents a significant advancement by seamlessly integrating learned interaction strategies with rigorous model based refinement for motion impedance control in real world contact scenarios.

Lucky paper: 2610.21199: Rosa: We're looking at LIN-SLAM today, it's paper number four of five in this set.

Dev: It’s the work on Lightweight LiDAR--Inertial Neural SLAM for Real-Time Onboard Navigation by Shiqi Che et al.

Taro: So the gist is coupling high rate geometric LiDAR inertial estimation with neural implicit tracking, continual mapping, and global refinement for accurate real time onboard SLAM.

Rosa: It's a hybrid framework combining computational efficiency of traditional geometric methods with map consistency and high accuracy of neural implicit approaches.

Dev: LIN-SLAM has three complementary components: a lightweight geometric frontend, a keyframe interface, and a neural implicit backend.

Taro: The geometric frontend uses an Iterated Error-State Kalman Filter, or IESKF, to maintain an eighteen-dimensional error state by combining GICP constraints with the inertial prior.

Rosa: This IESKF uses a rolling sweep buffer combining the active quarter sweep with preceding sectors to provide approximately full field of view scans and pose updates above the revolution rate.

Dev: The keyframe interface manages data flow between that high rate frontend and the slower neural backend, selecting completed sweeps so the backend doesn't block local estimation.

Taro: The interface motion compensates each LiDAR point from its acquisition time into the sweep end frame using a transformation, defining candidate points with their motion compensated coordinates and posterior covariance.

Rosa: The neural implicit backend represents the environment with a sparse point based neural implicit map, N = N n for j=one.

Dev: It initializes the map using the first selected sweep and uses spatial hashing to retrieve neighboring neural points for SDF value interpolation.

Taro: Neural tracking aligns each selected keyframe with the current neural map before mapping, utilizing a schedule based on normalized uncertainty to control tracking effort.

Rosa: Tracking updates the pose as T from Exp(delta xi)T using a robust weight W that downweights large SDF residuals and gradient norms.

Dev: Pose Graph Optimization corrects drift in the keyframe trajectory using sequential and accepted loop edges, and Map-Odometry Correction aligns each new frontend pose with the global map.

Taro: The experimental results on twelve representative sequences from three datasets spanning thirty-two-, sixty-four-, and one hundred twenty-eight-channel LiDAR configurations show that LIN-SLAM achieves the lowest overall mean absolute trajectory error among all evaluated geometric and neural implicit methods.

Rosa: On a Unitree Go2 equipped with an NVIDIA Jetson Orin NX and a sixteen-channel LiDAR, LIN-SLAM sustains real time localization and neural mapping over a one point four seven km indoor trajectory through challenging repetitive corridor environments with an endpoint closure error of only zero point one four one m.

Dev: The results also show that LIN-SLAM achieves the lowest overall mean absolute trajectory error across the evaluated public sequences and endpoint closure errors below zero point one five m over two real time onboard trajectories.

Taro: It achieves the lowest accuracy, completeness, and Chamfer-L1 errors and the highest F-score in both scenes, and it achieves the lowest overall mean ATE RMSE on Newer College Stereo-Cam, Newer College Multi-Cam, and M2DGR.

Rosa: LIN-SLAM achieves the lowest mean profiling time per frame for all tested methods across all datasets.

Dev: It operates stably throughout both sequences with its endpoint closure errors being zero point one four one and zero point zero seven three m, respectively.

Taro: The conclusion is that we present LIN-SLAM a lightweight LiDAR inertial neural implicit SLAM system that combines geometric estimation, asynchronous coupling, and optimized C++ CUDA neural execution to enable real time localization and continual neural mapping.

Rosa: We show that our method achieves state of the art LiDAR inertial IMU localization accuracy on both datasets and onboard runs while supporting online deployment on an NVIDIA Jetson Orin NX.

Dev: Future work will focus on combining LIN-SLAM with smooth control, which is interesting because it suggests a next step for practical use.

Taro: The optimization part decouples the geometric estimation and neural implicit processing into asynchronous threads to preserve estimation accuracy while maintaining computational efficiency.

Rosa: Selective neural backend processing together with an optimized implementation in C++ and CUDA substantially reduces complete system processing cost, enabling real time onboard operation.

Dev: The mean backend latency falls from two hundred forty-six point two hundred seventy-eight to eighty-two point six hundred five ms, which is a sixty six point five percent reduction.

Taro: Mapping shows the largest reduction there, from one hundred thirty point nine hundred one to twenty-nine point five hundred forty-five ms.

Rosa: The onboard deployment on the Unitree Go2 platform runs the complete LIN-SLAM system with a sixteen channel Hesai Pandar XT16 LiDAR and an NVIDIA Jetson Orin NX.

Dev: LIN-SLAM runs the complete pipeline concurrently at the original one times sensor rate, and mean profiling times of eighty-four point eight hundred seventy-five and eighty-eight point four hundred one ms remain below the one hundred ms period of the ten Hz LiDAR stream.

Taro: The mean GPU utilization is around eighty two point four percent and eighty six point nine percent over both sequences.

Rosa: Fewer glass surfaces and weaker wall reflections might explain Basement's better reconstruction, which is interesting because it points to real world data effects.

Dev: FAST-LIO2 processes complete LiDAR revolutions, whereas Point-LIO performs point wise updates, and geometric baselines show that they cannot preserve long term consistency in these runs.

Taro: PIN-SLAM and CLID-SLAM cannot sustain this rate with localization and continual neural mapping under the tested Orin NX configuration because the geometric baselines diverge early in both sequences.

Rosa: This confirms that geometric methods alone cannot preserve long term consistency when operating at this speed.

Lucky paper: 2610.20767: Rosa: We're looking at Multi-Dimensional Featurizers Yield Fine-Grained Steering and Data Shaping in Robot Foundation Models today.

Dev: This paper is really digging into how recovering multi-dimensional feature structure provides an interpretable interface for steering vision language action models and shaping the data they learn from.

Taro: The core methodology involves investigating block sparse featurizers BSFs to recover multi-dimensional task relevant features like perception, language, proprioceptive state, episode phase, and control variables.

Rosa: BSFs represent each feature in a low dimensional subspace and we investigate this causally through fine grained steering experiments where targeted changes in visual perception or semantic representation produce predictable changes in downstream action prediction.

Dev: The researchers use feature based data attribution to isolate where a specific behavior arises in the training dataset by identifying fewer than one point four percent of training tokens whose masking is sufficient to remove the targeted behavior.

Taro: They systematically compare BSFs and sparse autoencoders SAEs in vision language action VLA models hypothesizing that BSFs are a more natural fit than traditional SAEs because they recover features spanning perception, language, proprioception, task state and episode progress better capturing computational geometry.

Rosa: When compared against SAE baselines trained on substantially larger unpooled activation corpora approximately two orders of magnitude more than the pooled activations in the next largest dataset from Swann et al two thousand twenty-six BSFs show superior performance in recovering and intervening on task relevant structure.

Dev: The causal validation through steering experiments shows that geometric steering is most compact with a single BSF block achieving ninety-nine point seven percent recovery.

Taro: Closed loop target object redirection demonstrates that BSF features redirect the policy more reliably with fewer features than SAEs, selecting the redirected object in every trial and completing the task in ninety-three point three percent of trials using two point three features on average compared to two point five SAE features.

Rosa: The discovered BSF features are associated with visual visual language and robot state and control concepts; for example, Visual Features include those selective for objects in the environment and for the robot itself.

Dev: Visual Language Features show organization consistent with a shared vision language representation that retains distinctions between members of a semantic category.

Taro: Task and Control Variables are represented by features such as gripper configuration and end effector Cartesian coordinates over a single episode.

Rosa: In terms of behavior shaping the study identifies F1515, a three dimensional BSF feature in layer nine of pi0 point 5 that activates during recovery from failed grasps.

Dev: By masking the loss for any sixteen step action chunk overlapping an F1515 selected window they remove supervision for one point three eight percent of tokens while preserving observations and trajectory structure.

Taro: Concept based masking increases mean reopening time more than sevenfold from one point eight nine to fourteen point two five seconds exceeding the mean delay of all comparison methods.

Rosa: The conclusion is that BSFs recover interpretable multidimensional features in vision language action models providing tools for steering robot behavior and tracing it to training data.

Dev: Targeted interventions establish these features’ behavioral relevance while feature guided data masking selectively delays failure recovery and preserves task performance in their evaluation.

Taro: Robust closed loop visual and geometric steering will require tracking moving image regions and adapting to changing observations and robot state, but BSFs are not uniformly superior as SAEs perform better on some quantitative metrics underscoring the importance of unpooled activations.