Principles of Optimality in Robotics and Neuroscience

Optimality is a principal that is useful not only for generating robot motion, but also for modeling the behavior of systems with which robots interact. I will discuss two examples, one in robotics and the other in neuroscience.

First, by application of geometric optimal control, I will show how to establish an appropriate configuration space for robotic manipulation of canonical “deformable linear objects” like an elastic rod (e.g., surgical suture wire) or a planar elastic kinematic chain. This result leads to simple algorithms for manipulation planning that are easy to implement and that work well in practice.

Second, by application of a new framework for inverse optimal control of differentially flat systems, I will show how to identify and characterize “invariant outputs” of human locomotion given observations of unimpaired human walkers. This result leads to a method of control design for an active prosthetic leg that reduces the time required for parameter tuning by a clinician from four hours to four minutes.