3-Dimensional A* for Collision Free Walking Gait

Deep learning models have become a hot topic in many fields of study, including walking robotics. However, due to the complexities in dynamics in these systems, training and troubleshooting large networks can be an expensive and exhausting task. Therefore, developing a collision free algorithm for runtime foot path generation can be very valuable in the world of walking robots. This paper demonstrates the effectiveness of utilizing a unique blend of control theory techniques for balance and stability and artificial intelligence for dynamic foot path generation. An adaptive model predictive control (AMPC) controller is used for the balance and stability, while A* is utilized to dynamically generate foot paths. With each step, the algorithm generates a smooth foot path which can avoid or step onto obstacles, depending on their severity. This approach solves the problem of training large models by directly generating an optimal path with natural movements and no a priori training. The result is a stable and robust bipedal walking robot with the capability of stepping onto or over obstacles in new and complex environments.