PRATT, G.: Low impedance robotics

Almost all robots, including current walking robots, embody a “stiffer is better” design and control philosophy inherited from the earliest days of numerically controlled machine tools, where position accuracy in the face of unpredictable force disturbances was of paramount importance. This philosophy is so pervasive that is has become part of the cultural definition of a robot. By contrast, the impedance of animals is low. We cannot hold positions accurately in the face of unexpected force disturbances. What we can do is execute natural tasks, like locomotion, manipulation and prey catching, with breathtaking agility and grace. Previous work has shown that the passive dynamics of an animal’s body acts to self-stabilize and automatically sequence some movements. Unfortunately, in today’s stiff, trajectory controlled robots, such natural dynamics are over-ridden. We have developed a family of low-impedance actuators, mechanisms, and control philosophies that allow our robots to interact with the world more softly – more like animals. In the actuator arena, we have developed a succession of series elastic actuators that incorporate elasticity between the motor/transmission and the load. Used in a feedback configuration, this elasticity gives our actuators low minimum impedance, high force fidelity, high dynamic range, high shock tolerance, and the capacity for energy storage. In the control arena, we have developed physics-based low-impedance techniques and a general programming tool called virtual model control that extends the impedance control paradigm of robot control. Using these control methods on both simulated and real low-impedance walking robots, we have demonstrated that soft, model free control can be effected in parallel with the natural dynamics of the robot. Experiments have shown that these techniques are capable of achieving both stability, performance, and, to the human eye, grace.