Parallel passive-elastic elements can reduce the energy consumption and torque requirements for motors in powered legged systems. However, the hardware design for such combined actuators is challenged by the need to engage and disengage the parallel elasticity depending on the gait phase. Although clutches in the drive train are often proposed, compact and low cost solutions of clutched parallel elastic actuators have so far not been established. Here we present the design and control of an initial prototype for a parallel elastic actuator. The actuator combines a DC motor with a parallel spring that is engaged and disengaged by a commercially available, compact and low-cost electric clutch. In experiments that mimic the torque and motion patterns of knee extensor muscles in human rebounding tasks we find that the parallel spring in the prototype reduces the energy consumption of the actuator by about 80% and the peak torque requirement for the DC motor by about 66%. In addition, we find that a simple trigger-based control can reliably engage and disengage the electric clutch during the motion, allowing the spring to support the motor in rebound, to remove stored energy from the system as necessary for stopping, and to virtually disappear at the actuator output level. On the other hand, the hardware experiments also reveal that our initial design limits the precision in the torque control, and we propose specific improvements to overcome these limitations.

In biomechanics and biorobotics, muscles are often associated with reduced movement control effort and simplified control compared to technical actuators. This is based on evidence that the nonlinear muscle properties positively influence movement control. It is, however, open how to quantify the simplicity aspect of control effort and compare it between systems. Physical measures, such as energy consumption, stability, or jerk, have already been applied to compare biological and technical systems. Here a physical measure of control effort based on information entropy is presented. The idea is that control is simpler if a specific movement is generated with less processed sensor information, depending on the control scheme and the physical properties of the systems being compared. By calculating the Shannon information entropy of all sensor signals required for control, an information cost function can be formulated allowing the comparison of models of biological and technical control systems. Exemplarily applied to (bio-)mechanical models of hopping, the method reveals that the required information for generating hopping with a muscle driven by a simple reflex control scheme is only $I=32\phantom{\rule{0.16em}{0ex}}\mathrm{bits}$ versus $I=660\phantom{\rule{0.16em}{0ex}}\mathrm{bits}$ with a DC motor and a proportional differential controller. This approach to quantifying control effort captures the simplicity of a control scheme and can be used to compare completely different actuators and control approaches.