Programming Work

Despite decades of steadily accelerating computational power and recent advances in sensor technology, we have not yet achieved robots that can perform work according to our wishes. One challenge arises from the disappointing power density of available actuators: there are severe limitations to the rate at which work can be performed. But animals, suffering similar constraints, manage to go about their work quite effectively, exchanging energy with their environments to perform numerous and varying tasks in a goal directed manner.  The fundamental problem of robotics remains how to specify and execute programs that encode the goal directed exchange of energy (e.g., forces over motions) within a physical environment.

I will review a two decade agenda of robotics research seeking useful symbols and effective compositional methods that connect them into programs for the specification and execution of physical work.  Our effort is strongly inspired by biology and I will mention some of the ways in which we have tried to identify and borrow principles underlying the use animals make of their bodies in accomplishing their work. These efforts have, in turn, suggested new hypotheses about how animals’ bodies make use of their nervous systems in so doing. Over time, animal nervous systems have given rise to animal brains and I hope this talk may stimulate an exchange of views leading to new hypotheses about how the brain works and the body thinks.