Neural Network Self Tuning PI Control for Thin McKibben Muscles in an Antagonistic Pair Configuration

This paper proposes a model free neural network self-tuning proportional integral (NNPI) controller for a biceps-triceps thin McKibben muscle (TMM) platform in an antagonistic pair configuration. The study intends to explore the proposed model independent control strategy for TMMs in an antagonistic assembly for time varying joint angle tracking. In practice, PI controllers are tuned offline to obtain control parameters which suits the system. A change in the desired joint angle specifications may degrade the performance of the controller, hence the gains are no longer adequate. The proposed NNPI controller updates the control parameters in real-time according to the gradient descent method to minimize the error. To test the effectiveness of the proposed method, experiments are carried out on the TMM platform and injected with sinusoidal input signals with two different frequencies. Experiments conducted showed the TMM platform able to produce better accuracy for both conditions by implementing the NNPI control scheme compared to a Proportional Integral (PI) controller and a Model Free Adaptive Controller (MFAC). The control can be very useful in other TMM applications requiring antagonistic muscle actuation.

Automated summary

This paper proposes a model free neural network self-tuning proportional integral (NNPI) controller for a biceps-triceps thin McKibben muscle (TMM) platform in an antagonistic pair configuration. Experimental results show that the proposed controller performs better than the traditional proportional integral (PI) controller and the model free adaptive controller (MFAC) in time varying joint angle tracking.