|
In this talk I'll describe our efforts in using computational tools from optimal control theory (including machine learning and motion planning algorithms) to design efficient and agile control systems for locomotion. I'll start by describing the passive dynamics of walking, and demonstrate that approximate optimal control can be used to design nonlinear control solutions that allow minimally-actuated bipeds to walk efficiently and dynamically, even over rough terrain. In many cases, the performance of the algorithms can be improved dramatically by exploiting knowledge about the dynamics of the plant. Then I'll describe a new line of work applying these ideas to bird-scale aerial vehicles, and argue that model-free learning methods can design high-performance control solutions in even very complicated fluid dynamic regimes. Our initial evidence includes a robotic bird which flies with flapping wings and an airplane that can land on a perch. |