Optimizing robot model parameters for highly dynamic tasks

This work addresses the problem of designing a robot to perform dynamic tasks, like running, strictly through changes to the physical model. We find that this problem is formulable as an optimal control problem. We also find that if the motion can be prescribed for the robotic system as a function of its physical parameters, then that optimal control problem can be solved quickly. We demonstrate an implementation used to get a manually adjustable quadrupedal robot to run.