Vertical Stabilization of Bipedal Walking Drone PAVO with Proximal Policy Optimization

While autonomous mobile robots have gained more popularity over the last few years and many traditional problems (such as navigation and locomotion) seem to have been solved with an adequate level of accuracy, walking and biped robots still face many challenges. We present an approach to the stabilization of mobile robots with two legs, which move via quasi-dynamic movement. A popular machine learning method, Proximal Policy Optimization (PPO), was used to learn how to stabilize the robot in a vertical position. This method is popular for solving complex problem domains with high-dimensional state-action spaces and continuous states and actions, which are common in areas involving walking robots with a high number of degrees of freedom. The method was tested on a custom-designed biped walking robot, PAVO.