Inverse Optimal Control for a Simple Stepping Task

We consider a method of characterizing cost functions underlying legged locomotion. We use a trajectory optimization approach, allowing a separation of the constraints associated with walking, which walkers must follow, from various proposed cost components, which reflect preference. Here we present results on synthetic walking data, where we generate stepping motions using known cost functions, and use an iterative trajectory optimization process which requires only a single demonstrated optimal trial to infer the cost function weightings that were used to generate it.