Control of pneumatic artificial muscles using local cyclic inputs and genetic algorithm

Abstract

Recently, lightweight and flexible soft actuators have attracted interest from robotics researchers. We focused on pneumatic rubber artificial muscle (PAM) as a high-output soft actuator. The high compliance of PAM allows a robot to adapt flexibly to the environment without many external sensors. Although PAM has these characteristics, it is difficult to control because of the nonlinearity between the input and output and the delay of air response. This limits the accuracy of artificial muscles and complicates motion planning. Therefore, we considered that PAM can be driven by simplified control laws, so that the entire system shows emergent motion guided by metaheuristics. We developed a legged robot with two joints driven by PAMs. Each PAM was controlled with a cyclic signal, and the genetic algorithm was applied to optimize these signals. We tested to check whether the behavior of the PAMs is changed by the genetic algorithm using three simple performance indexes. We found out that although the genetic algorithm adjusted the local cyclic inputs appropriately according to each performance index, the time-varying characteristic of PAMs disturbed the monotonic increment of the evaluation values. We also discovered that by only adjusting the input timing, the leg develops a limitation in robustness.