
Robotics: Toward the New Science of Programmable Work
NSSEFF Proposal submitted by Daniel E. Koditschek, PI, University of Pennsylvania
Project Abstract/Summary
The fundamental problem of robotics is how to design and deploy general purpose mechanisms capable of performing user-specified mechanical work within a designated task domain. This statement reveals the three essential challenges to any agenda for such a science of programmable energy exchange: architecture (a theory and practice of design and deployment); environment (a rational framework for task domain models); and language (a formalism for user specification). The architectural problem is to formulate a user’s goal in terms of target energy landscapes intended to govern the coupled agent-environment pairing, yet to do so in a compositional manner. The resources available for engaging the environment are revealed by interpreting at the mesoscale advances in the thermodynamics of information applied to heat engines equipped with feedback loops. Finally, a recent convergence between the computer scientists and mathematicians working at the logical foundations of mathematics affords a new body of tools and techniques that can render the topological symbols of dynamical systems theory as typed expressions in a programming language. The project aims to produce a novel, correct, automated design environment for building robots with unparalleled capabilities along with a new generation of unusually interdisciplinary young scholars, all through the effort to stand up a new discipline on sound mathematical and physical foundations.