The transferability of evolved hexapod locomotion controllers from simulation to real hardware

The creation of control programs for legged robots has received much attention in the robotics community. Evolutionary Robotics (ER) offers one method of developing such controllers. The goal of this work was to use ER to evolve simple open-loop locomotion controllers for a real-world hexapod (six-legged) robot. This evolution was performed in a physics engine-based simulation, as is typically done in the ER process. Since our ultimate goal was to produce controllers which could perform adequately on the real-world robot, the current study explored different techniques which could potentially improve the transferability of the evolved controllers from simulation to the real-world robot. These techniques included optimization of parameters of the simulator by making use of real-world data, the addition of noise to the simulator predictions during controller evolution and explicitly taking into account the torque limits of the robot's motors in the simulator. The results obtained indicated that the transferability of these evolved controllers from simulation to the real-world robot was greatly aided by the simulator optimization, noise injection and incorporation of torque limits in the simulator.