This project, developing a coarse/fine approach to walking on unstructured terrain based on integrating 3D perception and compliant contact, addresses central challenges in robotic locomotion. Despite recent advances, unstructured tasks such as walking on rocks with sparse footfalls - a task that is easy for humans - remains very challenging for robots. This effort seeks to solve this type of problem with a coarse approach using 3-D perception that uses range sensing to find general curved surface patches in the environment that might be suitable for foot placement. Along with patches on the robot itself, these patches are placed into a novel dynamic spatial patch map with estimated uncertainty that mates robot patches with environmental patches. The fine alignment is handled by a compliant ankle that allows the patches to mesh naturally. This approach is applicable to dexterous manipulation, as well.
Broader Impacts: We are now at a transformative frontier where we must introduce robots to unstructured tasks -- in factories, hospitals, and labs; in outer space; and in the home -- where now only humans can manage the uncertainty. Through building robot analogues to the human process of uncertainty tolerance combining vision and compliance, this work will help enable a new science of high-uncertainty robotics. The proposed education plan is built around modules and open hardware that will be portable to mobile manipulation courses at other institutions, which will be actively promoted. It also aims for broader audiences with specific plans to engage younger students and adult self-learners.
