Influence Vectors Control for Robots Using Cellular-like Binary Actuators

TL;DR

This work tackles robust fault-tolerant control for soft robots built from cellular-like binary actuators, where cross-coupling and uncertain dynamics complicate model-based control. It introduces influence vectors, experimentally identified to form a Jacobian-like model $\boldsymbol{J}$ that maps binary actuator switches $\boldsymbol{b}$ to state changes, enabling both static ($\boldsymbol{a}(\boldsymbol{b}) = \boldsymbol{J}\boldsymbol{b}$) and dynamic (sliding-mode) control without a full analytical model. The static controller uses iterative binary recruitment via a genetic algorithm to minimize the resolution error $\boldsymbol{\epsilon}_r$, while the dynamic controller employs bang-bang decisions based on $\boldsymbol{s}=\dot{\boldsymbol{x}}_e + \lambda \boldsymbol{x}_e$ and per-actuator force vectors $\tilde{\boldsymbol{f}}_k$, achieving robust motion tracking. Experimental validation on a 4-DOF, 20-actuator soft robot demonstrates strong fault tolerance to perturbations and actuator failures, with practical control bandwidths and clear pathways to online adaptation and non-binary extensions.

Abstract

Robots using cellular-like redundant binary actuators could outmatch electric-gearmotor robotic systems in terms of reliability, force-to-weight ratio and cost. This paper presents a robust fault tolerant control scheme that is designed to meet the control challenges encountered by such robots, i.e., discrete actuator inputs, complex system modeling and cross-coupling between actuators. In the proposed scheme, a desired vectorial system output, such as a position or a force, is commanded by recruiting actuators based on their influence vectors on the output. No analytical model of the system is needed; influence vectors are identified experimentally by sequentially activating each actuator. For position control tasks, the controller uses a probabilistic approach and a genetic algorithm to determine an optimal combination of actuators to recruit. For motion control tasks, the controller uses a sliding mode approach and independent recruiting decision for each actuator. Experimental results on a four degrees of freedom binary manipulator with twenty actuators confirm the method's effectiveness, and its ability to tolerate massive perturbations and numerous actuator failures.

Figures (25)

• Figure 1: Soft cellular robot concept with pneumatic actuators.
• Figure 2: Soft exploration robots.
• Figure 3: Influence vectors for a planar manipulator: left, a displacement influence vector and, right, a force influence vector.
• Figure 4: Static control scheme, dotted lines are for binary vectors (i.e., vectors of 0's and 1's) and the dashed line is an optional feedback linearization loop.
• Figure 5: Resolution error vector $\boldsymbol{\epsilon}_r$ and approximation error vector $\boldsymbol{\epsilon}_a$.