Evolutionary Gait Reconfiguration in Damaged Legged Robots

To assess the algorithm's effectiveness, each best solution corresponding to a damage scenario is implemented on the damaged robot in the lab environment. We compare the predicted simulation result with real-world data obtained from the robot's IMU and the motion-tracker system. Robot orientation was observed using both IMU and motion-tracker data, while translational motion was evaluated exclusively using motion-tracker recordings due to the high noise typically associated with IMU-based linear acceleration measurements. The summary of the straight motion recovery performance (averaged over 10 best solutions) is depicted in Table II. In addition to that, Figures 4 and 5 (damage: Legs 1 & 6 missing), 7 and 8 (damage: Legs 3 & 4 missing), 10 and 11 (damage: Leg 1 missing), 13 and 14 (damage: Leg 4 missing), illustrate the motion of the main body's orientation and CoM position when a best solution is implemented on the damaged robot.