Despite many advances in AI, control theory, computing, sensing and actuation, electronics, soft actuators, etc., still, we cannot trust a robot to hold a live hamster without hurting it or to get a robot to walk on an uncalibrated outdoor terrain. It looks like we are missing a few fundamental things that enable living beings to survive in complex natural environments with noisy sensors and slow communication fibers in neural circuits.
We try to understand how physical mechanisms in the body and environment contribute to solving computational problems to do with efficient survival in unstructured environments. We observe that the brain sometimes withdraws from direct closed loop control between sensed states and actions to an indirect mode of involvement where it controls joint impedance to allow physical circuits make meaningful contributions to survive, like when a bicycle rider would stand up with bent knees when riding on a bumpy terrain. Furthermore, we observe that long-term evolution tunes sliding surfaces in joints and accompanying muscles that apparently simplify closed-loop control in the central nervous system (CNS). We call this kind of contributions made by physical mechanisms to simplify computation at the level of CNS to be “morphological computation”. Though most robots still find it difficult to exploit the full potential of morphological computation, they provide a good paradigm to test many underlying hypotheses.
This recent seminar talk gives an overview of things going on in my lab:
Our demo at the 5G World summit in collaboration with Ericsson and the Telecommunications Research group at KCL.