Recent years have witnessed tremendous growth in the capabilities of legged robots, with quadrupeds and humanoids demonstrating athletic behaviors that even five years ago were out of reach. Likewise, actively powered lower-limb assistive devices have made great strides in their maturity, with hardware such as the Open-Source Leg broadening access for future breakthroughs.

Despite this progress, the wide variability of real-world environments and users remains a pressing challenge to practical applications. It is fundamentally impossible to train our robots in the lab for anything they may encounter in an open world! As steps toward addressing this challenge, the talk will first discuss recent work on the control of the MIT Mini Cheetah, which considers new computational methods for the robot to reason through its actions on the fly in complex environments. The second part of the talk will then present ongoing research on improving user interfaces for lower-extremity prosthetic limbs to make human/robot interaction more fluid. Collectively, this work expands the ability of these systems to tailor their motions to new environments and users, paving the way for broader adoption in the “wild.”