Iterative Learning Control for Fast and Accurate Position Tracking with an Articulated Soft Robotic Arm

TL;DR

The effectiveness of the learning approach is demonstrated by a reduction of the root-mean-square tracking error from 13 degrees to less than 2 degrees after applying the learning scheme for less than 30 iterations.

Abstract

This paper presents the application of an iterative learning control scheme to improve the position tracking performance for an articulated soft robotic arm during aggressive maneuvers. Two antagonistically arranged, inflatable bellows actuate the robotic arm and provide high compliance while enabling fast actuation. Switching valves are used for pressure control of the soft actuators. A norm-optimal iterative learning control scheme based on a linear model of the system is presented and applied in parallel with a feedback controller. The learning scheme is experimentally evaluated on an aggressive trajectory involving set point shifts of 60 degrees within 0.2 seconds. The effectiveness of the learning approach is demonstrated by a reduction of the root-mean-square tracking error from 13 degrees to less than 2 degrees after applying the learning scheme for less than 30 iterations.

Figures (6)

• Figure 1: The articulated soft robotic arm used for the experimental evaluation. It consists of two antagonistically arranged soft bellow actuators and a rigid backbone. The inflatable actuators are made from coated fabric and attached to a lightweight structure built from a combination of 3D printed parts and aluminum plates. The low inertia of the one degree of freedom arm enables fast maneuvers.
• Figure 2: The bellow-type actuator made from polyurethane coated nylon in fully inflated state. The actuator consists of twelve single cushions and the front side measures $\unit[100]\times\unit[100]{mm}$ in a deflated state. When fully inflated, the maximum actuation range, is approximately $\unit[190]{^\circ}$. Two tubes connect the actuator to the control valves, while the third one is used to measure the pressure in the actuator.
• Figure 3: Schematic drawing of the articulated soft robotic arm including two actuators, namely A and B, the static and moving prisms, the rod and the revolute joint indicated by the black circle. In the configuration shown, the pressure in actuator A is higher than in actuator B, leading to a deflection in the positive $\alpha$-direction.
• Figure 4: Control architecture with the NOILC scheme in parallel configuration with the PID feedback controller in the outer control loop. The feed forward signal of the PID controller is not depicted for the sake of clarity. The input to the NOILC is the normalized error between the desired, $y_{\text{D}}$, and measured angle, $y^j$. The current value of the correction signal $u^j$ is added to the control input of the feedback controller, $u_{\text{PID}}$, and forms the input to the inner control loops. The pressure controllers (pCtrl A, pCtrl B) adjust the pressures in each actuator (Act A, Act B). These pressures are the inputs to the system (Arm).
• Figure 5: Experimental results of the robotic arm tracking an angular set point trajectory. The top plot shows the tracking performance when the feedback controller is used only (no learning, iteration 0). In this case, the angle initially follows the set point with a steep response, then slightly bounces back and finally reaches the set point. The middle plot shows the improved tracking performance after applying the NOILC scheme for 30 iterations. The bottom plot depicts the total pressure difference applied in iteration 0 (blue) and iteration 30 (red) and reveals the importance of the non-casual nature of the NOILC scheme, resulting in a shifted pressure difference input anticipating the repetitive disturbances. For the entire trajectory, the total pressure difference is never close to its maximum pressure constraint.