ABSTRACT

This project considers the problem of planning and controlling motions that involve intermittent physical contact. These motions include such tasks as walking across unexplored terrain, or manipulating an object with poorly known weight, shape, or surface roughness. The project approaches this problem in two ways. The first builds upon findings that humans respond to uncertainty by varying the effective springiness of their limbs. The project will formulate a corresponding approach to robot control by finding the robot limb stiffness that minimizes a probabilistic measure of risk under uncertain initial conditions. That is, the first part of the project will find the best possible outcome of a movement, while taking into account a spread of possible starting points. The second part of the project addresses the challenge of considering all the ways in which small changes to intermittent contacts -- such as when a foot hits the ground, or where a finger touches a tool -- can propagate through a larger task. There are efficient methods to handle such variability when the problem being study has a property called "convexity," which allows for efficient partitioning and search of the space of solutions. Contact problems do not have this desirable property, however the project will explore ways to approximate the true problem by a sequence of convex problems. Walking and grasping robots will increasingly help human co-workers in manufacturing settings, and assist elderly and disabled citizens in everyday tasks. This project will promote the national health and prosperity by improving the performance and reliability of robotic walking and grasping. The results are not limited to robotics and will also be beneficial in bio-mechanics and human motor control research, where they could suggest an explanatory framework for analyzing human behavior.

The project will characterize the optimal mechanical impedance modulation for robust contact interactions and provide a methodology to compute motions that are open-loop robust despite environmental uncertainties. It will leverage recent results in risk-sensitive optimal control and robust optimization to explicitly consider uncertainty about the environment while ensuring low computational complexity. The last but key objective of the project is to conduct extensive robotic experiments with a one-legged jumping robot, a manipulator grasping unknown objects, a quadruped walking and jumping and a humanoid robot climbing up high steps using its arms and legs therefore demonstrating the general applicability of the methodology in realistic and diverse robotic scenarios. The experiments seek to clarify the influence of external disturbances and environmental uncertainty on optimal impedance modulation and robust movements. Additionally, they will shed light on the important factors enabling robust execution of complex tasks in unknown environments. The project will also compare the optimal leg impedance predicted by the established modeling methodology with human walking data.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Please report errors in award information by writing to: awardsearch@nsf.gov.