Research Analyzer
← Back ICRA 2026

Global Tensor Motion Planning

An Thai Le, Kay Pompetzki, Joao Andre Mueller Carvalho, Joe Watson, Julen Urain, Armin Biess, Georgia Chalvatzaki, Jan Peters

PDF

AI summary

Key figure (auto-extracted from paper)
GTMP enables highly efficient, vectorized batch motion planning on GPUs/TPUs by using a fixed-layer tensor discretization, achieving up to 10^5x speedup over traditional planners while maintaining probabilistic completeness and path diversity.
Motion Planning Batch Planning GPU Acceleration Tensor Discretization Robot Learning Probabilistic Completeness

Problem

Traditional sampling-based motion planners rely on incremental graph structures that are difficult to vectorize, creating a bottleneck for large-scale robot learning tasks that require generating thousands of diverse motion plans quickly.

Approach

GTMP replaces incremental graphs with a fixed-layer random multipartite graph represented as tensors, enabling fully vectorized sampling, collision checking, and value iteration via JAX on modern accelerators.

Key results

  • Tensor-based fixed-layer discretization for vectorized planning
  • Up to 10^5× faster batch planning than CPU-based OMPL baselines
  • Competitive path smoothness and diversity without gradient-based optimization
  • Theoretical proof of probabilistic completeness for the tensor graph

Why it matters

It provides a scalable, hardware-efficient foundation for generating massive, diverse motion datasets required for modern robot learning and real-time control applications.

Abstract

Batch planning is increasingly necessary to quickly produce diverse and quality motion plans for downstream learning applications, such as distillation and imitation learning. This paper presents Global Tensor Motion Planning (GTMP)— a sampling-based motion planning algorithm comprising only tensor operations. We introduce a novel discretization structure represented as a random multipartite graph, enabling efficient vectorized sampling, collision checking, and search. We pro- vide a theoretical investigation showing that GTMP exhibits probabilistic completeness while supporting modern GPU/TPU. Additionally, by incorporating smooth structures into the mul- tipartite graph, GTMP directly plans smooth splines without requiring gradient-based optimization. Experiments on lidar- scanned occupancy maps and the MotionBenchMaker dataset demonstrate GTMP’s computation efficiency in batch planning compared to baselines, underscoring GTMP’s potential as a robust, scalable planner for diverse applications and large-scale robot learning tasks.

Index terms

Motion and Path Planning Manipulation Planning

Related papers