Florida Researchers Teach Humanoid Robots to Autonomously Traverse Rocky Terrain
The Institute for Human and Machine Cognition develops software that allows bipedal robots like Atlas and Valkyrie to independently plan footsteps across piles of cinder blocks. The system calculates the best path forward and achieves a 90% success rate on rough surfaces.
Florida's Institute for Human and Machine Cognition (IHMC) takes on the difficult challenge of enabling bipedal robots to walk across rough terrain. The research organization demonstrates its autonomous footstep planning program on Boston Dynamics' Atlas robot and the NASA-developed Valkyrie robot, showing impressive navigation over large piles of cinder blocks and narrow pathways.
The software works by establishing a starting point and a destination, mapping out every possible path on a footstep-by-footstep basis, and calculating the cost of each option to determine the best route. Modern hardware processes these complex calculations relatively quickly, allowing the robots to adapt to sudden environmental changes or path blockages in real time.
Despite the impressive capabilities, the system still faces some limitations, particularly with narrow paths where it currently achieves a 50% success rate. However, IHMC sees much better results on rough terrain with a 90% success rate, and the robots maintain a nearly perfect track record when walking on standard flat ground.