From Sketch Prior to Trajectories: A Mission-Oriented Coordinated Navigation Framework for Indoor UAV Swarm

arXiv:2607.11386 · cs.RO · Submitted 2026-07-13 · Read on arXiv

Listen

Transcript

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

Rosa: Today's paper: "From Sketch Prior to Trajectories".

Dev: UAV swarm for applications, such as indoor inspection, security patrol, and logistics delivery, are often mission-oriented rather than exploration-oriented.

Rosa: First, who's behind it and why it matters.

Title and authors: Rosa: So we're diving into "From Sketch Prior to Trajectories: A Mission-Oriented Coordinated Navigation Framework for Indoor UAV Swarm." This paper tackles a really practical problem where UAVs need to follow a specific sequence in indoor environments, like inspecting a building. It focuses on using pre-existing structural information instead of building massive maps from scratch.

Dev: I agree, Rosa. It sounds like they're addressing the latency and complexity that comes with full metric mapping when you only care about visiting specific areas in order. The idea of leveraging sketch map priors to reduce the need for dense global metric maps is something that really interests me from a control standpoint because it suggests a way to keep things computationally light.

Taro: From an autonomy angle, I'm curious how this handles unexpected events when the environment doesn't match the prior. If there are dynamic obstacles or unexpected blockages, how does this framework manage those deviations from the planned mission sequence?

Rosa: That's a big question, Taro. The paper suggests that onboard observations are used for topological alignment and updating traversability, which implies it can adapt to local changes in real time. It builds a representation with static structural constraints and dynamic traversability layers to handle those things.

Dev: Exactly, Rosa. And from an engineering perspective, I want to know how fast this adaptation happens. If we have a high-frequency loop rate requirement, will the topological alignment process introduce significant latency that could cause issues? We need something robust for real-time decision making.

Taro: That robustness under misbehavior is crucial; if the environment misbehaves, we need to know how the system reacts to maintain mission progress without getting stuck or failing its sequence requirements.

Rosa: The framework's core idea is that it uses region-level topological alignment and fuses that with online observations to create a mission-oriented traversability representation. This representation has a static structural layer, a dynamic traversability layer, and a region-state layer to track mission progress.

Dev: That layered approach sounds like it could be quite efficient computationally compared to maintaining one massive map for the entire building. I wonder how tightly coupled those layers are during the path planning phase; we need predictable behavior when generating those guide paths.

Taro: The way they define region states, like unvisited, active, or completed regions, is important because it gives the UAV team a clear understanding of where they are in the overall mission sequence G = (g one g two g M).

Title and authors: Rosa: Right. And they then use that state information to drive the 2D guided path planning layer to generate mission-oriented guide paths based on region connectivity and dynamic traversability. It's a nice way to ensure the path respects the required sequence while avoiding known blocked areas.

Dev: I'm thinking about that transition between those two layers, from 2D guidance to three dee trajectory optimization. How does the system translate those mission-oriented guide paths into dynamically feasible and collision-free trajectories? That part needs solid control theory underpinning.

Taro: When we look at the objective function they set up, minimizing T f + lambda X subject to constraints like maintaining a minimum safety distance d safe between UAVs and ensuring the mission sequence is followed in time, it shows they've formalized the coordination aspect really well.

Rosa: It’s definitely focused on generating a set of safe and efficient trajectories T = tau one tau two tau N that satisfy all those structural constraints and dynamic obstacle avoidance requirements. The whole point is to generate feasible solutions incrementally rather than trying to solve the whole mission at once.

Dev: Receding-horizon optimization is a smart way to manage the complexity, especially in a dynamic setting where things can change faster than we can compute a single global solution. I'm concerned about the computational load during that receding horizon step, though; we need tight control loops for that to work well in practice.

Taro: And considering their validation, they test this framework in both communication-available and communication-loss conditions, which speaks directly to the robustness needed when external synchronization breaks down. That's where real autonomy gets tested.

Rosa: Indeed, the experiments in structured multi-room environments show how well it scales up to layered indoor structures as well. It really demonstrates that this method works beyond just a single room setup because of how it handles regional transitions via planar connectivity.

Dev: So, to summarize what we've discussed about "From Sketch Prior to Trajectories: A Mission-Oriented Coordinated Navigation Framework for Indoor UAV Swarm," we're looking at how it uses sketch map priors for structural constraints and fuses them with online perception to build a mission-oriented traversability representation. This then feeds into a layered 2D-three dee navigation system that plans guide paths based on region connectivity and optimizes them into safe, collision-free trajectories.

Taro: And the implication there is that this could significantly lower the barrier for deploying complex coordinated missions in real indoor settings without needing extensive prior mapping data. It shifts the burden from creating perfect maps to intelligently aligning existing structural priors with real-world observations.

Title and authors: Rosa: I think what stands out most is how compact they make the navigation interface; they don't need a shared dense global metric map, which simplifies deployment immensely and reduces data sharing overhead among the swarm members.

Dev: From my side, the emphasis on generating trajectories in a receding-horizon manner suggests that it’s designed to be practical for real-world execution rather than just an abstract theoretical exercise. We'll need to keep an eye on how fast those local updates are processing within that horizon.

Taro: And the scalability demonstrated in multi-floor simulations is key; if it works across different floors, it has much broader applicability in large facilities, not just a single test room.

Rosa: So, as we wrap up this discussion on "From Sketch Prior to Trajectories: A Mission-Oriented Coordinated Navigation Framework for Indoor UAV Swarm," the main implication is providing a way to achieve mission-oriented coordination using lightweight priors instead of relying on heavy global mapping infrastructure.

Dev: It seems like a very solid framework for handling the dynamic, constrained nature of indoor coordination where you need high fidelity but also computational efficiency. We'll be looking forward to seeing how they address latency in the next iteration.

Taro: I just want to emphasize that its ability to handle mission progress state updates across regions is what makes it truly mission-oriented rather than just being a good pathfinding algorithm. It tracks the sequence, which is vital for complex logistics.

Rosa: That's a great point about the mission state tracking. Overall, this paper shows how to effectively bridge the gap between static structural knowledge and dynamic real-world operation using this layered 2D-three dee approach. We've covered a lot of ground on "From Sketch Prior to Trajectories: A Mission-Oriented Coordinated Navigation Framework for Indoor UAV Swarm."

Dev: It's been really informative hearing how they structure the problem from a control loop perspective, even though the overall optimization is complex. The focus on dynamic traversability updates is where I see the most immediate engineering challenges and opportunities.

Taro: I think it’s exciting because it moves towards systems that can be deployed faster in real operational scenarios, not just simulated environments. The validation under communication loss really shows its resilience for true autonomy.

Rosa: Well, that's all we have time for today on this paper; we'll keep an eye out for any follow-up work they might do on refining the online observation fusion process in the next few months.

The paper's summary: Rosa: So, to recap, this paper introduces a framework that uses pre-existing structural layouts as a lightweight guide to navigate indoor spaces for multiple UAVs on missions, rather than relying on dense maps.

Dev: That's right; they’re essentially taking those sketch maps and aligning them with what the UAV sees in real time at the region level to build a usable navigation representation.

Taro: I'm thinking about how this handles things when the environment deviates from the initial map, Rosa. What happens if there are unexpected obstacles or if a room layout is slightly different than planned?

Rosa: Well, they create this layered representation with static structural constraints and dynamic traversability layers that account for those real-time observations, which should allow it to adapt locally without needing a complete rebuild of the global map.

Dev: From my end, I'm focused on the loop rate; if the topological alignment process takes too long, we lose our timing for trajectory generation and coordination, and we need to know how fast this adaptation actually runs in practice.

Taro: Exactly, because if it can't handle unexpected changes gracefully while still hitting its mission sequence goals—the region-order requirements you mentioned—then it’s not truly autonomous enough for complex operations.

Rosa: The real strength they show is that by treating the sketch map as just a structural prior and fusing it with onboard perception, you get a representation that captures both the persistent structure and the current dynamic traversability of each area.

Dev: That sounds computationally efficient, which is great because building massive metric maps for every operation would be too slow for real-time control loops.

Taro: And I'm really interested in how this impacts deployment outside of a controlled lab setting; Rosa, can you tell us if this framework has been tested long enough in the real world to prove its reliability over extended operational periods?

Rosa: The paper mentions both simulation and real-world experiments under both communication-available and communication-loss conditions, including multi-floor simulations, which suggests they've looked at scenarios beyond just a short lab test.

Dev: That’s what I’m really watching; the robustness in those communication loss scenarios is where most of the control system headaches happen; how well does it rely on local region states when external synchronization drops?

Taro: It seems to build mission progress tracking directly into the representation, which means even if communication is lost, each UAV knows exactly which part of the sequence it needs to focus on next, maintaining that goal orientation.

Rosa: So, this moves us closer to a future where UAV swarms can execute complex tasks—like detailed inspections or delivery sequences—in areas where perfect pre-deployment mapping isn't feasible.

Dev: The implication for me is that if the latency is low enough during those trajectory optimizations, we could get much faster response times when obstacles appear mid-flight.

Taro: And the big picture here is that this approach makes complex coordinated missions accessible to a wider range of real-world indoor applications without needing massive, resource-heavy global map infrastructure upfront.

Rosa: It’s exciting because it shifts the focus from building perfect maps to intelligently aligning existing structural knowledge with what the AI sees in real time, which is a much more practical approach for field robotics.

Dev: Before we move on to the results section, I'd like to know about those specific failure modes they tested; did they find any instances where the dynamic traversability layer failed catastrophically under high-density obstacle scenarios?

The paper's improvements: Rosa: So, to sum up the improvements they propose, they are really focusing on making this system more practical for real deployment by addressing several key areas of functionality and efficiency.

Dev: I'm looking at point number five about adaptive velocity constraints; that suggests an AI that can adjust speed based on immediate risks like obstacle proximity and how close neighboring UAVs are, which should help manage safety dynamically.

Taro: That adaptability is important because it means the system doesn't just stick to a pre-set speed when things get crowded or suddenly blocked by something unexpected.

Rosa: And point number six addresses communication loss; they want the system to keep making progress using only local information and region states, which is crucial for true autonomy in environments with unreliable connectivity.

Dev: That reliance on local state seems like a solid way to handle those intermittent links; it means the core navigation doesn't have to halt waiting for a signal from an external source.

Taro: I think this robustness under communication loss is what makes it really viable for mission-oriented tasks, as it ensures the swarm can continue executing its sequence even when the network hiccups.

Rosa: Plus, they emphasize computational efficiency in point number seven by treating the sketch map as a lightweight structural prior instead of requiring dense global maps for every single operation.

Dev: That's a big win for me from an engineering standpoint; if we can keep the computational load low by using that prior instead of a massive metric map, we can run those trajectory optimizations at higher frequencies.

Taro: And it connects back to my earlier point about real-world viability; if the system is computationally light and resilient, then its application outside a perfectly simulated environment becomes much more realistic.

Rosa: The implication is that we can deploy these sophisticated coordinated navigation systems in complex indoor settings without needing to build and distribute incredibly detailed global metric maps beforehand.

Dev: It really boils down to making the system efficient enough that it doesn't just look good on paper but actually runs reliably under the tight constraints of real-time control loops.

Taro: I'm also interested in how this efficiency scales; if it’s light on resources, could we potentially deploy this type of coordination framework across much larger swarms or even across multi-floor structures more effectively?

Rosa: That scalability was already touched upon with the multi-floor simulation results, suggesting they believe the framework is robust enough to handle those layered indoor structures successfully.

Dev: If we can keep that computational efficiency up while maintaining high safety standards, this could significantly impact how we design autonomous systems for logistics and inspection in complex facilities.

Taro: It's promising because it tackles the core challenge of coordination—getting multiple agents to move safely through a structured environment following a set path—without the massive data overhead usually associated with metric-map based approaches.

Conclusion: Rosa: So, to wrap things up on "From Sketch Prior to Trajectories: A Mission-Oriented Coordinated Navigation Framework for Indoor UAV Swarm," this paper essentially shows a way to use simple structural layouts as a guide for multiple UAVs in indoor missions, rather than needing massive maps.

Dev: That’s the core idea—using that lightweight sketch map prior to create a mission-oriented traversability representation that guides the 2D and three dee navigation layers.

Taro: And from an autonomy standpoint, this is exciting because it provides persistent structural constraints that help the AI understand its environment better even when things get messy.

Rosa: I think the implication here is that we can deploy these sophisticated coordinated missions in real-world indoor settings without needing to build and distribute incredibly detailed global metric maps beforehand.

Dev: It really does shift the burden from perfect mapping to intelligent alignment with what the AI observes on the ground, which makes it much more feasible for practical applications.

Taro: And when considering its resilience under communication loss, this approach suggests a level of autonomy that’s actually useful in environments where connectivity isn't guaranteed.

Rosa: Exactly; it’s about building systems that can maintain mission progress by relying on local region states rather than continuous external synchronization.

Dev: I just want to reiterate my concern about the latency; if those topological alignment updates aren't fast enough, the entire coordinated trajectory generation process could fall behind real-time requirements.

Taro: That’s a fair point, Dev; we need to see how they handle those dynamic changes in traversability without introducing delays that compromise the sequence adherence.

Rosa: Overall, it’s a very practical method for field robotics because it’s designed to be efficient and scalable across different indoor structures.

Dev: It seems like a really solid framework for handling the dynamic constraints of coordinated movement where you need high fidelity but also computational efficiency.

Taro: I think the ability to scale this to multi-floor environments is what makes this paper so relevant for larger facilities, not just single rooms.

Rosa: Well, that’s all we have time for on "From Sketch Prior to Trajectories: A Mission-Oriented Coordinated Navigation Framework for Indoor UAV Swarm"; it really shows how to bridge the gap between static structure and dynamic operation.

Dev: It's been a solid look at how they structured the problem from a control loop perspective, especially concerning those receding horizon optimizations.

Taro: I think its ability to track mission progress state updates across regions is what makes it truly mission-oriented rather than just being a good pathfinding algorithm.

Nanyang Technological University

cs.RO

Submitted: 2026-07-13

Updated: 2026-09-29

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 82/100

The gist: UAV swarm for applications, such as indoor inspection, security patrol, and logistics delivery, are often mission-oriented rather than exploration-oriented.

Key concepts

Sketch Map Priors: These are pre-existing structural layouts used as a lightweight guide for navigation instead of building massive metric maps. This approach reduces the need for extensive prior mapping data when navigating indoor spaces, making deployment simpler and reducing data sharing overhead among swarm members.
Layered Representation: The framework creates a representation with static structural constraints, dynamic traversability layers based on real-time observations, and a region-state layer to track mission progress. This allows the system to adapt locally to unexpected changes without rebuilding the entire global map.
Mission Progress Tracking: The system tracks mission progress by defining region states such as unvisited, active, or completed regions. This state information drives path planning, ensuring UAVs follow the required sequence of tasks even when communication is lost.
Receding-Horizon Optimization: This optimization method is used to generate trajectories incrementally rather than solving the entire mission at once. It helps manage computational complexity during real-time execution and allows for fast response times when obstacles appear mid-flight.

Terminology

Alignment — 18× in this paper · explained in 1 episode
Structural Constraints — 13× in this paper · explained in 2 episodes
Robustness — 5× in this paper · explained in 5 episodes
Core — 4× in this paper · explained in 2 episodes
Fidelity — 3× in this paper · explained in 2 episodes
Adaptation — 2× in this paper · explained in 1 episode
Objective Function — 2× in this paper · explained in 2 episodes
Abstraction — 1× in this paper · explained in 1 episode
Adaptability — 1× in this paper · explained in 1 episode
Benchmark — 1× in this paper · explained in 2 episodes
Decoupling — 1× in this paper · explained in 2 episodes
Failure Modes — 1× in this paper · explained in 3 episodes
Generalization — 1× in this paper · explained in 13 episodes
Information Gain — 1× in this paper · explained in 2 episodes
Local Optimization — 1× in this paper · explained in 2 episodes
Point Clouds — 1× in this paper · explained in 2 episodes
Reinforcement Learning — 1× in this paper · explained in 4 episodes
Task Planning — 1× in this paper · explained in 2 episodes

Summary

UAV swarm for applications, such as indoor inspection, security patrol, and logistics delivery, are often mission-oriented rather than exploration-oriented. In these tasks, UAVs are required to visit task-relevant regions in a prescribed sequence, and such region-level mission information can often be obtained from pre-deployment sketch-map priors, such as floor plans, CAD layouts, or evacuation diagrams. Although these tasks are executed in three-dimensional space, UAVs usually fly within a specific altitude layer or a nearly fixed altitude range on each floor, making mission-level region transitions mainly governed by planar connectivity. Based on these observations, this paper proposes a mission-oriented coordinated navigation framework that exploits sketch-map priors for multi-UAV indoor operations. Onboard observations are used to perform topological alignment, and the aligned prior is fused with online observations to construct a mission-oriented traversability representation. A layered 2D–3D coordinated navigation framework is further developed, where 2D guided path planning generates mission-oriented guide paths and guide-driven 3D trajectory optimization produces dynamically feasible and collision-free trajectories. Simulation and realworld experiments validate the effectiveness of the proposed framework in structured multi-room indoor environments and further demonstrate its coordinated navigation capability under both communication-available and communication-loss conditions. Multi-floor simulation results show the scalability of the system to layered indoor structures.

Instead of constructing and sharing a dense global metric map, the proposed framework treats the sketch map as a lightweight structural prior and aligns it with online observations at the region level. The aligned prior provides persistent structural constraints and mission-region relationships, while onboard perception updates dynamic traversability, neighboring UAV occupancy, and local 3D collision information. These elements are unified into a mission-oriented traversability representation consisting of a static structural layer, a dynamic traversability layer, and a region-state layer.

Based on this representation, the navigation framework follows a layered 2D–3D design. The 2D guided path planning layer generates mission-oriented guide paths according to region connectivity, mission progress, and dynamic traversability, while the guide-driven 3D trajectory optimization layer converts the guide into dynamically feasible and collision-free trajectories.

The main contributions are summarized as follows:

"We propose a mission-oriented traversability representation for multi-UAV indoor operations with sketch-map priors. By integrating region-level topological alignment, static structural constraints, dynamic traversability, and region-state information, the representation provides a compact navigation interface without requiring a shared dense global metric map."

"We develop a layered 2D–3D coordinated navigation framework that exploits the 2.5D structure of building operations. The 2D guided path planning layer performs mission-level guidance over planar region connectivity and dynamic traversability, while the guide-driven 3D trajectory optimization layer generates dynamically feasible and collision-free trajectories."

"We validate the proposed framework through simulation and real-world multi-UAV experiments in structured multi-room environments under both communicationavailable and communication-loss conditions. Additional multi-floor simulation demonstrates the scalability of the proposed representation and navigation framework to layered indoor structures."

The problem is framed as generating a set of safe and efficient trajectories T = τ1, τ2,..., τN that enables the UAV team to complete the given mission sequence while satisfying structural constraints, dynamic obstacle avoidance, interUAV safety, and region-order requirements. The objective is formulated as:

min

T,Tf

Tf + λ

i=1

Z Tf

∥p˙ i(t)∥ dt

s.t. pi(t) ∈ F(t), ∀i,

∥pi(t) − pj (t)∥ ≥ dsafe, ∀i ̸= j,

s(gk) = completed, ∀gk ∈ G,

t(gn) < t(gn+1), n = 1,..., M − 1.

In practice, this is not solved as a global multi-agent optimal control problem over the entire mission horizon. Instead, the proposed framework generates feasible solutions in a receding-horizon manner through a missionoriented traversability representation and a layered 2D–3D coordinated navigation framework. The 2D guide generation provides region-level efficient routes, while the 3D trajectory optimizer refines them into dynamically feasible and collision-free UAV trajectories.

The proposed framework consists of three tightly coupled components: sketch-map utilization and topological alignment, mission-oriented traversability representation, and 2D–3D coordinated navigation. Before deployment, the sketch map provides a lightweight structural prior, including region topology, entrances, exits, traversable connections, and the mission sequence. During operation, each UAV uses onboard perception to observe local structural features and dynamic obstacles.

Improvements for AI systems

Here are specific, actionable improvements for current AI systems based on the proposed framework:

I can develop a multi-UAV indoor navigation system capable of executing complex, mission-oriented sequences (e.g., inspection routes, logistics deliveries) in GNSS-denied environments without requiring a pre-built dense global metric map.
The improved AI system can perform coordinated navigation across multiple rooms and floors by leveraging a lightweight sketch-map prior for persistent structural constraints and aligning them with real-time onboard perception (topological alignment).
The system will generate dynamically feasible, collision-free trajectories by utilizing a layered 2D-3D framework:
The 2D layer will generate mission-aware guide paths based on region connectivity and dynamic traversability, while the 3D layer refines these into safe flight paths using local occupancy maps.
The system will incorporate adaptive velocity constraints by analyzing forward dynamic risk (based on obstacle proximity and neighboring UAV presence), adjusting speeds to maintain safety without unnecessary vertical movement or inefficient detours.
The improved AI can operate robustly under communication-loss conditions, maintaining mission progress by relying on local region states and onboard observations rather than continuous external synchronization.
The system will be highly efficient in terms of computational resources by treating the sketch map as a structural prior rather than requiring the construction and sharing of dense global metric maps for every operation.

Abstract

UAV swarm for applications, such as indoor inspection, security patrol, and logistics delivery, are often mission-oriented rather than exploration-oriented. In these tasks, UAVs are required to visit task-relevant regions in a prescribed sequence, and such region-level mission information can often be obtained from pre-deployment sketch-map priors, such as floor plans, CAD layouts, or evacuation diagrams. Although these tasks are executed in three-dimensional space, UAVs usually fly within a specific altitude layer or a nearly fixed altitude range on each floor, making mission-level region transitions mainly governed by planar connectivity. Based on these observations, this paper proposes a mission-oriented coordinated navigation framework that exploits sketch-map priors for multi-UAV indoor operations. Onboard observations are used to perform topological alignment, and the aligned prior is fused with online observations to construct a mission-oriented traversability representation. A layered 2D--3D coordinated navigation framework is further developed, where 2D guided path planning generates mission-oriented guide paths and guide-driven 3D trajectory optimization produces dynamically feasible and collision-free trajectories. Simulation and real-world experiments validate the effectiveness of the proposed framework in structured multi-room indoor environments and further demonstrate its coordinated navigation capability under both communication-available and communication-loss conditions. Multi-floor simulation results show the scalability of the system to layered indoor structures.