Robot learned to stand up via RL without human reference

View on Reddit

A robot trained in MuJoCo using reinforcement learning learns to stand up from a lying position, with a reward function based on torso height and end pose, without using human motion references. The project demonstrates simple reward engineering for standing behavior.