From Offline to Online: A Perception-Based Local Planner for Dynamic Obstacle Avoidance

Abstract

The deployment of robots within realistic environments necessitates robots to be capable of replanning their loco-manipulation trajectories on the fly to avoid unexpected interactions that may occur due to the uncertainty that is present in such dynamic and varying workspaces. This work introduces a novel method for the online local replanning of precomputed global trajectories for redoundant robots. The local nature of the problem leads to a sparse system that a hyper-graph can encode more intuitively. Using a graph, we can also store the entire global trajectory, preventing the local planner from getting stuck in local minima, and vertices and edges can be dynamically added or removed to ignore those constraints that do not interfere with the local problem, further increasing the computational efficiency. This process is accompanied by a control layer that iteratively takes the online refined solution and safely moves the robot. The method is validated both in simulation and experimentally on the wheeled-legged quadrupedal robot CENTAURO, demonstrating its effectiveness in replanning online the loco-manipulation trajectories of the robot under the occurrence of obstacles that intervene with the initially planned trajectories.