Event
PUN neural operator enables faster planning under noise
1 reports · 1 sources
Michael Chen, Hong Chul Nam, Xi Deng, and Anima Anandkumar propose the Planning Under Noise (PUN) neural operator, which learns a noise-aware value field to handle motion planning with control execution noise in unseen geometric environments. The method changes the obstacle boundary condition to a reflective Neumann condition and sets a Dirichlet condition at the goal, and extends the Monte Carlo solver Walk on Stars to high dimensions, learning the mapping from geometry to value field from noisy, partially converged labels, avoiding expensive convergence computations for each training sample. In evaluations at four noise levels, PUN matches or outperforms all compared planners, achieving 95–100% success rate with the lowest runtime on the 4D replanning benchmark; at the highest noise level in 4D, its success rate improves by 10.7–60.6% relative to other methods, and it runs 3.29–32.15 times faster.
Report timeline
- Planning Under Noise (PUN): A Neural Operator Planning Method Under Noise
PUN (Planning Under Noise) is a neural operator that learns noise-aware value fields to enable motion planning with control execution noise in unseen geometric environments. It changes obstacle boundary conditions to reflecting Neumann conditions and sets a Dirichlet condition at the goal, and extends the Monte Carlo solver Walk on Stars (WoSt) to higher dimensions, learning the mapping from geometry to value field from noisy, partially converged labels.
arXiv cs.RO ·