Model Predictive Wave Disturbance Rejection for Underwater Soft Robotic Manipulators

Abstract

Inspired by the octopus and other animals living in water, soft robots should naturally lend themselves to underwater operations, as supported by encouraging validations in deep water scenarios. This work deals with equipping soft arms with the intelligence necessary to move precisely in wave-dominated environments, such as shallow waters where marine renewable devices are located. This scenario is substantially more challenging than calm deep water since, at low operational depths, hydrodynamic wave disturbances can represent a significant impediment. We propose a control strategy based on Nonlinear Model Predictive Control that can account for wave disturbances explicitly, optimising control actions by considering an estimate of oncoming hydrodynamic loads. The proposed strategy is validated through a set of tasks covering set-point regulation, trajectory tracking and mechanical failure compensation, all under a broad range of varying significant wave heights and peak spectral periods. The proposed control methodology displays positional error reductions as large as 84% with respect to a baseline controller, proving the effectiveness of the method. These initial findings present a first step in the development and deployment of soft manipulators for performing tasks in hazardous water environments.

Figures (8)

• Figure 1: Piece-wise integration of fluid forcing across the segment body, where the relative motion of the fluid is considered to account for non-steady flows and evaluation is performed at each local point, $s$, along the manipulator.
• Figure 2: Overview of the control architecture deployed for disturbance rejection; it should be noted dependencies have been dropped here in the model for the sake of space.
• Figure 3: The various wave spectra selected for testing, all based on real-world data collected using wave buoys, showing (a) the JONSWAP curve for each spectra and the temporal representation wave cases (b) W1, (c) W2 and (d) W3 (each scaled to $H_s = 3m$).
• Figure 4: When subjected to the (a) wave train, the evolution of the (b) segment tip positions with respect to the base at (0m, 0m), (c) actuation torques and (d) wave loading torques are analysed. Shown are the evolution's for case W3 and highlighted in grey are key points of note where large disturbances are impacting the manipulator.
• Figure 5: Variation in error between the MPC strategy and the baseline feedforward + PD strategy, represented as an RMSE ratio. Shown are the results for cases (a) W1 (b) W2 and (c) W3.