LSTM-NN-Enhanced Tracking Control for PAM-Driven Parallel Robot Systems With Guaranteed Performance

Abstract

Mechanical systems often face unpredictable surrounding situations in applications, which bring lots of intangible uncertainties into system operations. Further, some robot systems, especially, pneumatic artificial muscle (PAM)-driven robot systems, also have accumulative nonlinearities, such as rate-dependent hysteresis, creep, and periodic/regular time-varying parameters, increasing design difficulties of high-accuracy controllers. This paper develops a long short-term memory neural network (LSTM-NN)-enhanced adaptive controller for PAM-driven parallel robot systems with transient and steady-state performance constraints. Specifically, a continuous-time LSTM-NN structure is introduced to recover unknown lumped dynamics, improving the approximation ability of time-dependent terms with accumulative effects. Moreover, a new two-stage error transformation function is designed to flexibly adjust the desired transient and steady-state performance, facilitating better adaptation to task requirements. To our knowledge, this paper proposes the first solution of utilizing the LSTM-NN-based neuroadaptive method for soft actuator-driven robots to enhance tracking accuracy with transient/steady-state performance improvement. The detailed stability analysis and several groups of experimental results on the self-built platform are provided to verify the feasibility and versatility of the proposed method.