Researchers have introduced LEAP, a trajectory optimization method designed for legged and manipulating robots. The approach addresses the challenge of generating dynamically feasible motions, which requires solving a large, nonlinear, non-convex optimal control problem. A key difficulty is that contact wrenches must satisfy friction-cone constraints at all times.

LEAP combines gradient flow with an augmented Lagrangian formulation. This combination is intended to make the constrained optimization problem more tractable than standard techniques. The paper's abstract frames the method as a way to handle contact constraints directly during trajectory generation.

Because only the abstract was available for this digest, there is no second source to compare or contrast. The claims are as reported by the authors.