Robot manipulators that work against rigid objects must manage both contact forces and motion, but these goals often conflict. Existing approaches typically formulate the problem in task space, then map the result to joint torques. That works well until practical limits—joint stops, collision avoidance, singularities—enter the picture, because those constraints are naturally expressed in joint space, not task space.

Jon Woolfrey's paper proposes solving the control problem directly in generalised coordinates using Gauss' principle of least constraint. The control law becomes a quadratic programming problem, which allows inequality constraints to be appended directly. The theoretical solution is shown to reproduce the correct constrained dynamics, giving accurate contact force control while projecting motion control into the unconstrained directions.

Simulations in MuJoCo compare the approach with the operational space formulation. The proposed method simultaneously holds a desired contact force, keeps trajectory tracking error low, and satisfies practical inequality constraints. The paper is submitted to ICRA 2027 and is available on arXiv.