Learning and Control for Adaptive and Reactive Robots

Catalog description

For decades, we've envisioned a future where robots seamlessly coexist, collaborate, and cooperate with humans in our everyday lives. Yet, the reality of robots fluidly interacting with humans and other robots in our dynamic, human-centric environments remains elusive. This challenge stems from traditionalist views of how robot motions, tasks and behaviors should be specified, controlled and learned. To overcome this bottleneck, we must change the way we control robots, starting with changing the way we train roboticists. This graduate-level course introduces the fundamental principles of the modern Dynamical Systems (DS) paradigm for motion planning, learning, and control; designed to create adaptive, reactive, easy-to-teach, provably safe and stable robot behaviors in dynamic, ever-changing environments. We will cover a range of topics through the DS lens including, reactive control and motion planning, impedance, admittance and force control, safety-critical control, stability and convergence guarantees for physical human-robot and robot-robot interaction, human-guided learning and efficient learning for interactive robots. Students will learn to formally model and analyze controllers and learning algorithms for efficient, safe and adaptive robots interacting with humans and the physical world through homework, literature reviews and a final project.