Applications in Controlling Soft Robots
Flowrate and pressure are the two most significant factors to be considered when controlling soft pneumatic robots. To drive these robots accurately, one should be able to continuously regulate these two parameters to the desired values. Therefore, to validate the applications of these MRE valves in controlling soft pneumatic robots, the closed-loop control performance of these valves in both flowrate and pressure regulation was tested. PID controllers were implemented in real-time on a microcontroller. The detailed setup, tuning and the PID parameters can be found in Section \ref{sec_experiment}. The step response of the controlled flowrate and pressure were recorded.
Closed-loop Flowrate Control
For closed-loop flowrate control, the experimental schematic is shown in Figure \ref{763625}(A). Constant pressure was supplied to the inlet of the MRE valve, while the MRE valve outlet was connected to the atmosphere via a flowrate sensor. The MRE valve here behaved like a pneumatic resistor whose resistance value can be continuously altered. The reading of the flowrate sensor q was then compared with a reference value q*, which is a step signal in our test. The error between the reference and the actual value was sent into a PID controller. The reference value was also sent into a forward model obtained in Figure \ref{793304}, which gave the expected amount of electric current required by the valve. The final control signal combined the output of the PID controller and the model. The obtained control signal was then fed into a signal amplifier to control the electric current through the MRE valve. The electric equivalent of the pneumatic circuit used here is presented in Figure \ref{763625}(B). Figure \ref{763625}(C) presents the step response of the controlled flowrate under three different control schemes tested. "PID only" means only a PID controller is used in the loop. In this case, the response time at the rising edge is around 0.5 s while the response time at the falling edge is around 0.8 s. No significant steady-state error is observed, and the root-mean-square error (RMSE) during the steady state is 14.37 mL/min. "Model only" means the PID feedback loop is disabled and the valve is controlled in an open-loop manner with only the forward model obtained from Figure \ref{793304}. A significant error of more than 50 mL/min is observed, caused by the difference between the actual valve behavior and the pre-obtained model. The response time at the rising edge is longer than 2 s due to the viscoelasticity of the MRE membrane. However, the response time at the falling edge reduces to 0.15 s thanks to the feed-forward model. "PID + Model" means the feedforward model is combined with a PID controller to eliminate the model inaccuracy and thermal drift. This scheme shows the best performance among all three schemes. The response times are 0.4 s and 0.1 s at the rising and falling edge respectively. No significant steady-state-error are observed and the RMSE during the steady state is 8.77 mL/min.