Control of a climbing robot using real-time convex optimization

This paper presents a controller for a free-climbing robot called Cartesian Force Convex control. This controller computes in real-time control inputs that will move a robot safely along a nominal plan by using a convex optimization technique based on an LP solver to compute torques to joint motors. For the LP solution, a cost function is chosen that increases robustness by weighting tracking performance of the robot’s body against the margins in the constraint variables (e.g. proximity of the contact forces to their friction cone limits). The constraints are formulated to characterize the climbing problem and include both static and dynamic limits. The effectiveness of this controller is demonstrated both with simulations and hardware experiments.