The future of robotics as we envision it is to fulfill essential roles in supporting humanity, from mundane and repetitive daily tasks to daring, dynamics feats in support of disaster relief. Reliable execution of these tasks often requires robots to utilize frictional contact in an arbitrary, unstructured manner. Generic algorithms for synthesis and characterization of such behaviors are computationally intractable, as they require selection from and analysis over the combinatorially vast set of contact initiation and termination sequences. By contrast, humans and animals are able to perform stunning displays of athleticism and dexterity with only a corse understanding of contact behaviors—and rigorous definitions of our intuition are a key enabler of high reliability on a limited number of tasks in robotic manipulation and locomotion. The impressive performance of such simple methods suggests that the essence of our robots’ dynamics is often significantly simpler than the constituent expressions of our models, even in the absence of useful human intuition. Discovering these underlying minimal models—either by hand-design or automated synthesis—therefore presents an opportunity to significantly increase the tractability of online planning, estimation, and control for complex systems.

Relevant publications