Legged robots are very beneficial in the context of traversal and navigation through unstructured irregular terrain for the goal of search and rescue. Despite the impressive results achieved in planning and controls of quadruped robots, the weak point of most approaches is in their inability to produce robust high-speed perception-based locomotion. Part of the problem lies in the hardware: MIT Mini Cheetah is the only small-scale quadruped robot able to achieve locomotion speeds of up to 2.5 m/s. In this project, I will build an uncertainty-based perception-mapping-planning pipeline for this robot. The key development is the multi-behavioral path planner that will be used to prime lower-level controllers to produce smooth behavioral transitions, resulting in fast locomotion through varying terrain. My solution combines the concepts of navigational meshes, borrowed from the video game industry, with custom dynamic-resolution terrain maps, which allow for probabilistic height and traversability estimation.

“I’ve been contributing to research and publications in various labs since high school. I have much more to learn – my aim is to go to graduate school to study controls and planning. SuperUROPing at the Biomimetic Robotics Lab is a great next step in learning about research in these fields. Controls and planning for walking robots is difficult and fascinating, and I am excited to get an applied perspective on underactuated robotics.”