Multi-contact Variable-Compliance Manipulation in Extreme Clutter.

摘要

We have developed new foundational capabilities for robot manipulation that assume contact across the entire manipulator is inevitable and desirable. Our approach makes use of compliant actuation and full-body force-sensing skin. We developed force-sensitive skin, low-level control algorithms, mid-level control algorithms, and planners that enable robots to reach to locations in extreme clutter, such as foliage and rubble, while haptically generating 3D maps of their surroundings. We have performed experiments with software simulated robots with skin, a hardware-in-the-loop system that simulates skin for a real robot, and real robots with real force-sensing skin covering their arms. In these tests, our novel control systems have enabled robots to perform qualitatively new tasks and outperformed baseline systems both in terms of success rate and keeping contact forces low. Our most recent control system also substantially outperforms our original control system in terms of time to complete (i.e. speed), success rate, and contact forces. Our project has resulted in a new and broadly applicable approach to robot manipulation that enables robots to achieve dramatically improved real-world manipulation performance. We have also produced open source code and open hardware implementations using open standards.