CAREER: Generalization and Safety Guarantees for Learning-Based Control of Robots

点击翻译按钮获取中文摘要

The ability of machine learning techniques to process rich sensory inputs such as vision makes them highly appealing for use in robotic systems (e.g., micro aerial vehicles and robotic manipulators). However, the increasing adoption of learning-based components in the robotics perception and control pipeline poses an important challenge: how can we guarantee the safety and performance of such systems? As an example, consider a micro aerial vehicle that learns to navigate using a thousand different obstacle environments or a robotic manipulator that learns to grasp using a million objects in a dataset. How likely are these systems to remain safe and perform well on a novel (i.e., previously unseen) environment or object? How can we learn control policies for robotic systems that provably generalize well to environments that our robot has not previously encountered? Unfortunately, existing approaches either do not provide such guarantees or do so only under very restrictive assumptions. This Faculty Early Career Development (CAREER) project seeks to establish a foundational framework for learning-based control of safety-critical robotic systems with guaranteed generalization and safety. The project will impact challenging application domains such as aerial inspection and manipulation (e.g., for infrastructure repair tasks) and includes activities for (i) engaging regulatory agencies and industry entities in discussions regarding the certification of learning-based robotic systems, (ii) partnering with teacher preparation programs and other educational programs to engage high-school and undergraduate students in robotics, and (iii) widely disseminating materials from a new robotics course which uses hands-on labs with drones. Motivated by the need for guaranteeing the safety of learning-based robotic systems, this project is developing a principled theoretical and algorithmic framework for learning control policies for robotic systems with provable guarantees on generalization to novel environments (i.e., environments that the robot has not previously encountered). The key technical insight of this project is to leverage and extend powerful techniques from generalization theory in theoretical machine learning. The resulting framework provides bounds on the expected performance of learned policies (including ones based on neural networks) across novel environments. The project is developing algorithms (based on convex optimization, gradient-based methods, and black-box optimization) for learning policies that explicitly optimize these bounds. The project also seeks to guarantee the robustness of learned policies to shifts in the distribution of environments that the robot encounters. An important part of the effort is to thoroughly validate the technical approach on hardware platforms including micro aerial vehicles performing navigation, inspection, and aerial manipulation tasks motivated by infrastructure repair applications. This project is supported by the cross-directorate Foundational Research in Robotics program, jointly managed and funded by the Directorates for Engineering (ENG) and Computer and Information Science and Engineering (CISE).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.