Vii. Conclusions

In this paper we have shown that for natural tasks, series elastic actuators can provide many benefits when compared to traditional stiff actuators. These benefits include shock tolerance , lower reflected inertia, more accurate and stable force control , less damage to the environment, and energy storage. Although zero motion force bandwidth is reduced as series elasticity is increased, the force bandwidth for many tasks that involve load motion is improved. This is particularly true for natural tasks that are spring-or damper-like in their output impedance[22]. We have also shown that a simple control system can generate a range of output impedances-not just that of the passive series elasticity, and have demonstrated experimentally that accurate, stable control is possible over a range of generated impedances. Since force is the direct variable of control, energy field methods[16] may also be used with series-elastic actuators. Several avenues are open for future work: In cases where the reduction of zero motion bandwidth cannot be tolerated, one can connect the outputs of two differently tuned series-elastic actuators in parallel. The series elasticity in each actuator isolates its motor's inertia from the output, thus making it possible to connect the output of a high frequency actuator (perhaps with lower ratio gearing and a stiffer series elasticity) to the output of a low frequency actuator (with higher ratio gearing and a less stiff series elasticity). In much the same way a hi-fi speaker uses a tweeter and woofer to cover a larger frequency range, two series-elastic actuators can be configured to cover a wider bandwidth. Because frictional elements may differ in the two motors as well, a parallel connection can also improve dynamic range. This approach is currently being investigated at MIT by John Morrell and Ken Salisbury[17]. Another interesting idea is to use variable-rate springs where modulation of the bias point can effect changes in passive stiffness. This type of mechanism has been studied before[21] and a more sophisticated version is currently being investigated at MIT by Ken Salisbury. In many cases (e.g. tendon manipulators and pneumatic actuators), series elasticity is unavoidable. It is our belief that when such an actuator is to be used in natural tasks, this property should be looked on as a blessing rather than a curse. Mike Binnard at MIT is presently investigating series-elastic control in a small pneumatic robot. The revolute actuators discussed in this paper have been designed …