Masked Generative Motion Planning with Geometry-Guided Token Search

2026-10-07

AI Summary

This study addresses the limitation of local continuous refinement in generative motion planning, which struggles to achieve path-level reconstruction. To overcome this, we propose the MGMP framework, which leverages a masked generative Transformer to produce discrete trajectory candidates in parallel, effectively reformulating continuous refinement as an efficient discrete search problem. Furthermore, a geometry-guided token search mechanism is introduced to facilitate global path reconstruction. Experimental evaluations demonstrate that the proposed framework achieves a 96% success rate on the Ring Maze task and an 82% repair rate for Kuka robotic arm manipulation, significantly outperforming existing baseline methods. Notably, the approach successfully generalizes to real-world physical scenarios, highlighting its practical applicability and robustness in complex motion planning tasks.

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 prior-supported 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.

Problem

• motion planning
• generative model
• trajectory repair
• structural refinement
• robotics

Innovation

• Masked Generative Motion Planning
• Geometry-Guided Token Search
• Generative Transformer
• Discrete Trajectory Candidates
• Structural Repair