Exploiting robot redundancy in collision detection and reaction

We present a method that allows automatic reaction of a robot to physical collisions, while preserving as much as possible the execution of a Cartesian task for which the robot is kinematically redundant. The work is motivated by human-robot interaction scenarios, where ensuring safety is of primary concern whereas preserving task performance is an appealing secondary goal. Unexpected collisions may occur anywhere along the manipulator structure. Their fast detection is realized using our previous momentum-based method, which does not require any external sensing. The reaction torque applied to the joints reduces the effective robot inertia seen at the contact and lets the robot safely move away from the collision area. If we wish, however, to continue the execution of a Cartesian trajectory, robot redundancy can be exploited by projecting the reaction torque into the null space of a dynamic task matrix so as not to affect the original end-effector motion. This leads to the use of the so-called dynamically consistent approach to redundancy resolution, which is further elaborated in the paper. A partial task relaxation strategy can also be devised, with the objective of keeping contact forces below a user-defined safety threshold. Simulation results are reported for the 7R KUKA/DLR lightweight robot arm.