Masked Generative Motion Planning with Geometry-Guided Token Search

The gist The Masked Generative Motion Planning (MGMP) introduces a method that extends learned trajectory priors from efficient parallel generation to structural repair, enabling route-level restructuring beyond local trajectory deformation

How it works

MGMP reframes motion planning by formulating it as masked generation over discrete trajectory tokens, reusing the same token space for test-time structural repair. A trajectory tokenizer maps expert trajectories into discrete codes and reconstructs continuous trajectories, while a Masked Generative Transformer (MGT) predicts masked codes conditioned on the start state, goal state, and environment.

Geometry-Guided Token Search (GGTS)

To make the search effective, Geometry-Guided Token Search (GGTS) uses scene geometry to guide two decisions: where to edit and what to fill. Geometric attribution determines where to edit by selecting the token positions most responsible for the current violation while preserving unaffected parts of the trajectory. To decide what to fill, the MGT proposes alternative motion tokens within the selected repair region, which are first screened by a geometric score and then verified through exact trajectory evaluation before an improving edit is finalized. This targeted, coarse-to-exact search avoids unguided perturbations and exhaustive codebook scans, inducing large, coordinated changes in motion structure.

Test-Time Refinement Process

The refinement process involves iterative rounds where each round makes two decisions: which positions to re-mask and what codes to put there. The first MGT forward pass defines the token distribution pθ(zi c) at each position, and the Gauss-Southwell rule is used to select regions requiring refinement based on the magnitude of the gradient. Following this, candidate proposals are screened using a first-order approximation of the geometric objective J to rank replacements, and shortlisted candidates are verified using hard-token trajectories against the exact objective J. The best replacement is committed only if it improves the current solution.

Evaluation and Generalization

MGMP is validated on diverse tasks and environments, showing generalization across unseen layouts, additional obstacles and unseen geometries, single- and dual-arm planning, and real-world Baxter tasks. Diagnostic benchmarks like Ring Maze require substantial route changes to escape local planning basins, where MGMP achieves 96% success. The method generalizes to unseen obstacle layouts and geometries by utilizing DeepSDF for differentiable collision costs during refinement.

Performance Metrics

MGMP achieves 96% success on Ring Maze and 82% repair success on Controlled Route Invalidation, exceeding the strongest external baselines by 23 and 25 percentage points, respectively. In real-world experiments with a Baxter robot, MGMP performs consistently well across shelf placing, button pressing, and drawing tasks. The method demonstrates that GGTS provides most of the basin-escape capability while the final polishing stage resolves residual local infeasibilities.

Ablation Study Insights

The ablation study isolates the main design choices in GGTS, showing that geometry-only remasking provides a higher success rate than confidence-only remasking for where to edit. Furthermore, the analysis indicates that discrete search performs the main structural repair while continuous optimization resolves residual local infeasibilities. The use of k=2 for candidate verification and a 2x2 refinement schedule achieved the best overall performance in several settings.

Conclusion

MGMP reframes generative motion planning by keeping the learned trajectory prior active beyond initial generation, while GGTS turns prior-supported motion alternatives into a geometry-guided, targeted search space. Rather than relying on local continuous refinement to discover new routes implicitly, MGMP directly restructures trajectories through discrete token replacement, using continuous optimization only for residual correction. Controlled basin-escape and route-invalidation experiments show that discrete search provides the main repair capability while broader evaluations demonstrate generalization across diverse planning settings. More broadly, MGMP shows how learned trajectory priors can serve not only as generators, but also as structured proposal spaces for geometry-guided test-time adaptation.

--- Page 1 ---

MASKED GENERATIVE MOTION PLANNING WITH GEOMETRY-GUIDED TOKEN SEARCH Lipeng Zhuang Yingdong Ru Shiyu Fan Edmond S. L. Ho Gerardo Aragon Camarasa Paul Henderson University of Glasgow, Glasgow, United Kingdom ABSTRACT Generative motion planners typically use learned trajectory priors for initial generation, while leaving test-time repair to local continuous refinement. We introduce Masked Generative Motion Planning (MGMP), which extends the learned prior from efficient parallel generation to structural repair. A masked generative transformer generates discrete trajectory candidates in parallel, and Geometry-Guided Token Search (GGTS) uses scene geometry to target where to edit and which priorsupported alternatives to evaluate. This turns refinement into an efficient search over discrete motion alternatives, enabling route-level restructuring beyond local trajectory deformation. MGMP achieves 96% success on Ring Maze and 82% repair success on Controlled Route Invalidation on Kuka, exceeding the strongest external baselines by 23 and 25 percentage points, respectively. It further generalizes to unseen layouts, additional obstacles, unseen geometries, single- and dual-arm planning, and real-world Baxter tasks. 1 INTRODUCTION Motion planning in robotics aims to find a feasible trajectory between a start and goal configuration while satisfying geometric constraints (LaValle, 2006). Trajectory-optimization methods directly refine trajectories against task-specific objectives (Ratliff et al., 2009; Schulman et al., 2014), but their performance is typically sensitive to initialization in cluttered and non-convex environments (Pan et al., 2014). Generative planners mitigate this sensitivity by learning multimodal trajectory priors and producing diverse candidate motions that can subsequently be refined to satisfy collision constraints. Generative motion planning therefore involves two coupled inference problems: efficient initial trajectory generation and effective test-time geometric refinement. Existing planners make trade-offs on both fronts. For initial generation, diffusion-based methods rely on iterative denoising in a high-dimensional continuous space (Janner et al., 2022; Carvalho et al., 2023), leading to high inference cost; fewer denoising steps improve efficiency but can reduce sample quality. For test-time refinement, generated trajectories are typically refined through continuous guidance or trajectory optimization (Carvalho et al., 2023; Luo et al., 2024; Yang et al., 2025). Weak guidance may leave collisions unresolved, while stronger guidance can move trajectories away from the learned prior (Yang et al., 2025). Such continuous refinement is effective when feasibility can be recovered through local adjustments; however, in harder cases such as near-enclosing obstacles, reaching a feasible alternative may require many iterations, remain sensitive to initialization and guidance strength, and still fail to escape unfavorable local basins. We therefore ask: How can a learned trajectory prior directly expose qualitatively different routes as candidates for structural repair, rather than expecting continuous refinement to reach them? We introduce Masked Generative Motion Planning (MGMP), which addresses this question by formulating motion planning as masked generation over discrete trajectory tokens, reusing the same token space for test-time structural repair. A trajectory tokenizer learns compact motion tokens from feasible trajectories. For initial generation, a Masked Generative Transformer (MGT) (Chang et al., 2022; 2023) predicts token sequences in parallel, producing high-quality trajectory candidates.

--- Page 2 ---

Figure 1: Overview of MGMP. We replace continuous trajectory generation and refinement with discrete masked generation and Geometry-Guided Token Search, enabling efficient planning and structural route repair. We evaluate it across challenging simulated and real-world settings, including basin escape, unseen geometries, additional obstacles, dual-arm planning, and Baxter tasks. in a single forward pass and reducing the need for costly resampling and post-hoc repair during online planning. Importantly, the same representation allows us to cast repair as a guided search over prior-supported motion alternatives. To make this search effective, we introduce Geometry-Guided Token Search (GGTS), which uses scene geometry to guide two decisions: where to edit and what to fill. Geometric attribution determines where to edit by selecting the token positions most responsible for the current violation while preserving unaffected parts of the trajectory. To decide what to fill, the MGT proposes alternative motion tokens within the selected repair region, which are first screened by a geometric score and then verified through exact trajectory evaluation before an improving edit is finalized. This targeted, coarse-to-exact search avoids unguided perturbations and exhaustive codebook scans; instead, a small number of targeted discrete replacements induce large, coordinated changes in motion structure, enabling exploration of alternative routes beyond the current planning basin.

--- Page 3 ---

Figure 2: Overview of GGTS. We use geometric attribution to select edit positions, first-order scores to screen token replacements, and sequential verification to accept improving candidates, followed by optional continuous polish.

1. textbfMasked Generative Motion Planning (MGMP) for Structural Repair: Enables route-level restructuring beyond local deformation by formulating motion planning as masked generation over discrete trajectory tokens. This allows the system to achieve route-level restructuring beyond local trajectory deformation by leveraging a shared token space for both initial generation and test-time repair.

2. textbfGeometry-Guided Token Search (GGTS) for Targeted Repair: Turns refinement into an efficient search over discrete motion alternatives, enabling route-level restructuring. It uses scene geometry to determine where to edit and what to fill, moving beyond unguided perturbations by employing a process where it selects token positions based on the gradient of the geometric objective J.

3. textbfEnhanced Robustness via Multi-Stage Refinement: The system separates structural repair from initial generation quality by using discrete token replacement provides structural repair, while continuous optimization performs only local correction. This allows for basin escape through discrete search, as evidenced by achieving 96% success on Ring Maze and repairing trajectories even when the initial generation is poor.

4. textbfGuaranteed Repair Isolation: Controlled Route Invalidation allows researchers to isolate repair capability by initializing all methods from the same trajectory and testing whether structural repair capability from initial-generation quality. This proves that MGMP can achieve 82% repair success on Controlled Route Invalidation, demonstrating its ability to fix trajectories that require substantial route changes.

5. textbfReal-World Task Generalization: MGMP demonstrates generalization to complex, unseen scenarios, including unseen layouts, additional obstacles, unseen geometries and real-world Baxter tasks. This is achieved by using techniques like DeepSDF for collision costs during refinement and adapting the MGT architecture via AdaLN conditioning for different robot arms (single-, dual-arm planning).