Extended Euclidean Distance-Based Model Predictive Control for Safety-Critical Dynamic Obstacle Avoidance

This article proposes a dynamic obstacle avoidance framework for autonomous mobile robots (AMRs), which employs the extended Euclidean distance (EE) method in both the global path search and local trajectory optimization layers. An EE-based safety metric (EESM) is formulated, incorporating the velocity information of dynamic obstacles, to assess the potential collision risk with obstacles. The framework starts with a dynamic perception module that processes point cloud data to parameterize obstacles and predict their trajectories. In the global planning layer, a path-searching algorithm combines the EESM with kinematic constraints to generate a collision-free global path. An extended control barrier function (ECBF) was developed for the local planning layer, which was then combined with model predictive control (MPC) to formulate an ECBF-based model predictive control (MPC-ECBF) planner, ensuring real-time safe obstacle avoidance. Extensive simulations and real-world experiments have been implemented to validate the proposed framework, demonstrating its improved success rate and smoother trajectories, therefore validating its effectiveness and safety in autonomous navigations.