Optimal Control for a Shape Memory Alloy Actuated Soft Digit using Iterative Learning Control

Modeling and control of soft robotic systems remains a challenging and active field of research, especially for robots actuated with shape-memory-alloys (SMA). Difficulties in controlling SMA-actuated robots arise in modeling the nonlinearities of the SMA dynamics as well as accounting for their hysteresis behavior. This article helps address these challenges by using sensorization of the end effector along with SMA actuators combined with a simplified two-axis pendulum model to implement an optimal control scheme on a soft digit. The control stack presented uses a direct collocation method (DIRCOL) for initial trajectory optimization, with the actual state of the soft digit then captured and used to refine the system inputs using Iterative Learning Control (ILC) to increase trajectory tracking accuracy. The system is demonstrated on hardware, achieving very close trajectory tracking performance for two-dimensional motion tasks.