This benchmark aims at developing a computer program that controls a humanoid robot to walk as far as possible given a limited battery life. The programming language is Python and the humanoid robot model is a ROBOTIS OP2 robot.
The environment is a scale model based on the robot size.
The benchmark metric is the distance traveled by the robot.
The distance is the the difference between the initial and final position of the robot on the world's x-axis that is parallel to the street.
If the robot falls down or tries to go on all fours, the benchmark finishes immediately.
If you look at the Python program controlling the ROBOTIS OP2 robot, you will see that this program uses the RobotisOp2GaitManager to make the robot walk:
# At the beginning, start walking on the spot. # After 45 timesteps, begin taking steps forward. while robot.step(timestep) != -1: if looptimes == 45: gaitManager.setXAmplitude(gaitAmplitude) gaitManager.step(basicTimeStep) looptimes += 1
The parameters of the gait manager are customizable and it is possible to find a setup that produces better performance. For example, tuning the gait amplitude or disabling the balance functionality makes the robot cover a bigger distance.