Eikonal Model Predictive Path Integral for Risk-Aware Mobile Robot Navigation

In this paper, a stochastic-based path-planning optimization technique called Eikonal-MPPI is presented, which combines the model predictive path integral (MPPI) and the eikonal equation to achieve a high successful rate of risk traversal navigation and maintain computational efficiency. In this approach, the cost function of the MPPI is preprocessed by the eikonal equation based on the risk transform function to establish a velocity field. After the preprocessing, the resulting Eikonal cost map is injected into the MPPI, gains higher success in producing optimal and safe trajectories amidst complex environmental configurations. Compared to other MPPI-based path planners such as the baseline MPPI, Log-MPPI, Cluster-MPPI, and BiCMPPI, the proposed approach reached the highest success rate of 95.5 % while keeping the processing time low on BARN dataset, and low control effort on the elevation risk map. It is predicted that the proposed approach will be suitable for risk-sensitive autonomous robotic applications.