Case Studies for Computing Density of Reachable States for Safe Autonomous Motion Planning

Density of the reachable states can help understand risk safety-critical systems, especially in situations when worst-case reachability is too conservative. Recent work provides a data-driven approach to compute density distribution autonomous systems’ forward online. In this paper, we study use such combination with model predictive control for verifiable safe path planning under uncertainties. We first learned collision If exceeds acceptable threshold, our method will plan new around previous trajectory, below threshold. Our well-suited handle systems uncertainties and complicated dynamics as does not need an analytical form estimate state arbitrary initial design two challenging scenarios (autonomous driving hovercraft control) motion environments obstacles system show that estimation reach similar accuracy Monte-Carlo-based while using only 0.01X training samples. By leveraging estimated risk, algorithm achieves highest success rate goal reaching enforcing safety above 0.99.