A convex, smooth and invertible contact model for trajectory optimization

摘要: Trajectory optimization is done most efficiently when an inverse dynamics model available. Here we develop the first of contact defined in both forward and directions. The impulse solution to a convex problem: minimize kinetic energy space subject non-penetration friction-cone constraints. We use custom interior-point method make problem unconstrained; this key defining consistent way. resulting has parameter which sets amount smoothing, facilitating continuation methods for optimization. implemented proposed solver our new physics engine (MuJoCo). A full Newton step trajectory 3D walking gait takes only 160 msec, on 12-core PC.