SBIR Phase II: Gripper-integrated proximity, contact and force sensing for collaborative robots

The broader impact/commercial potential of this project is a new generation of smart robotic hands that will manifest a new plateau for industrial manipulation and robotics research. Previously hard manipulation challenges will be available out-of-the-box, enabling industry to create new, complex applications. Integration of sensing, actuation, control, and programming environment into a single low-cost device will decrease complexity, setup time and cost of collaborative robotic solutions, and, together with a new generation of low-cost robotic arms, enable industrial tasks of high commercial value, including bin-picking, kitting and assembly. A simple, graphical programming interface paired with a browser-based scripting interface will make robot programming accessible to a workforce with widely varying educational backgrounds, thereby increasing productivity and efficiency in the workplace and ensuring long-term economic viability of US-based manufacturing. This project also contributes to workforce development by training a new breed of engineers that are equally versed with mechanical, electronics and control, while providing opportunities for a diverse population of students at the University of Colorado. This Small Business Innovation Research Phase II project fundamentally advances the understanding of the interplay between proximity, contact and force sensing; self-supervised learning techniques to improve hand-coded robot controllers; and manufacturing challenges of systems that tightly integrate sensing, actuation, and computation. Building up on a proof-of-concept gripper design resulting from Phase I research that has successfully demonstrated a variety of challenging industrially-relevant bin picking, kitting and assembly tasks, this research will investigate new algorithms for dual-arm assembly problems including improved path planning and collision avoidance, optimization algorithms for automatic parameter tuning to improve grasp and assembly success, and training of deep convolutional neural networks to replace hand-coded segmentation algorithms for object identification and ?actor critics?. Rigorous testing for robustness, reliability and accuracy throughout the project will drive algorithm development. Safety analysis will feed back into hardware and controller design. 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.