PATACON: GPU-Parallel Tangent-Bundle RRT-Connect for Constrained Motion Planning
In this work, we introduce PATACON, a GPU-parallel bidirectional planner that explores tangent space to exploit the local manifold geometry induced by known constraint functions.
ProofPaper ↗
Key points
- Fast whole-body motion planning enables continuous replanning for humanoid robots in dynamic environments while satisfying collision avoidance and kinematic constraints.
- However, constrained motion planning incurs substantial computational cost due to repeated constraint projection in addition to collision checking and nearest-neighbor search.
- Experiments on dual-arm closed-chain and humanoid whole-body constrained planning tasks show that PATACON outperforms the prior work as configuration-space dimension and constraint codimension grow, achieving up to an 11.25 times speedup.
- These results suggest that GPU-parallel tangent-space exploration brings constrained whole-body motion planning closer to real-time performance.
Sources (1)
- [1]PATACON: GPU-Parallel Tangent-Bundle RRT-Connect for Constrained Motion PlanningarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 08:13 AM
In this work, we introduce PATACON, a GPU-parallel bidirectional planner that explores tangent space to exploit the local manifold geometry induced by known constraint functions.
Fast whole-body motion planning enables continuous replanning for humanoid robots in dynamic environments while satisfying collision avoidance and kinematic constraints.
Extractive summary: sentences quoted from the sources.