DO-MPPI: Improved Model Predictive Path Integral Path Planning with Dynamic Obstacle Evaluation for Mobile Robots

Abstract

To address the problems of delayed obstacle avoidance response and insufficient dynamic safety margin of traditional local path planning algorithms in complex dynamic environments, this paper proposes an improved Model Predictive Path Integral (MPPI) path planning algorithm fused with a dynamic obstacle evaluation mechanism, named DO-MPPI. The method introduces a Kalman filter-based obstacle motion state estimation and temporal position prediction module into the MPPI sampling framework, constructs a spatio-temporal joint collision detection model, and automatically eliminates sampled trajectories with collision risk through a high-weight collision penalty, thereby achieving proactive avoidance of dynamic obstacles while retaining the global optimization ability of random sampling. Simulation experiments are carried out in purely static, low-speed same-direction, and high-speed crossing dynamic obstacle scenarios, and physical experiments are conducted on a self-developed four-wheel differentially driven firefighting inspection robot platform. The results show that, compared with traditional MPC, standard MPPI, and DWA algorithms, the proposed DO-MPPI algorithm significantly reduces the collision rate, improves the arrival success rate, maintains a larger minimum safety distance, and shortens the obstacle avoidance response time, demonstrating superior proactive obstacle avoidance performance in complex dynamic environments.