IHMC Atlas and Valkyrie Autonomous Footstep Planning Over Rough Terrain

This video demonstrates the autonomous footstep planning developed by IHMC. Robots in this video are the Atlas humanoid robot (DRC version) and the NASA Valkyrie. The operator specifies a goal location in the world, which is modeled as planar regions using the robot's perception sensors. The planner then automatically computes the necessary steps to reach the goal using a Weighted A* algorithm. The algorithm does not reject footholds that have a certain amount of support, but instead modifies them after the plan is found to try and increase that support area. Currently, narrow terrain has a success rate of about 50%, rough terrain is about 90%, whereas flat ground is near 100%. We plan on increasing planner speed and the ability to plan through mazes and to unseen goals by including a body-path planner as the first step. Control, Perception, and Planning algorithms by IHMC Robotics (http://robot.ihmc.us). Video and paper submitted to IEEE Humanoids 2019 conference, and posted to arXiv at http://arxiv.org/abs/1907.08673. Atlas robot built by Boston Dynamics. Valkyrie robot build by NASA Johnson Space Center.