Reproduce “Robust Predictive Motion Planning by Learning Obstacle Uncertainty” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jian Zhou, Yulong Gao, Ola Johansson, Björn Olofsson, Erik Frisk
An approach to robust predictive motion planning for autonomous systems is presented, which integrates online learning of an obstacle's intended control set with a robust Model Predictive Control (MPC) framework. The method achieved a 100% collision-free rate in simulations, maintained a safety distance five times greater than deterministic methods, and demonstrated real-time applicability in hardware experiments (average 64ms execution time), without requiring prior knowledge of obstacle uncertainty distributions.
Key insights
- Online learning of an obstacle's intended control set through efficient linear programming or recursive updates can effectively reduce prediction conservativeness compared to worst-case assumptions.
- Approximating the intended control set with parameterized polytopes and reforming the optimization into a Linear Program (LP) allows for real-time computation and adaptation to changing obstacle behaviors.
- Combining less conservative obstacle predictions with a robust MPC framework enables motion planning that is both safe (achieving high collision-free rates) and more feasible/efficient than traditional worst-case robust methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.