Human-Machine Co-Learning Design in Controlling a Double Inverted Pendulum

Effective human-machine interaction is an essential goal of the design of human-machine systems. This, however, is often constrained by the fundamental limitation of the human neural control and inability of the machine’s control system in adapting to the time-varying characteristics of the human operator. It is desirable that the control system of the machine can learn to optimize its performance under the behavior change of the human operator. This thesis is aimed at enhancing the machine’s control system with learning capabilities. Specifically, an adaptive control framework is proposed that enables human-machine co-learning through the interaction between the machine and the human operator. A dual inverted pendulum system is introduced as an experimental platform. Simulations are performed to implement the control of the two-joint inverted pendulum using the human-machine co-learning controller. The results are compared with those using a controller without learning ability. The parameters of the two controllers are adjusted to explore the effect of the value changing of each parameter on the control performance. Simulation results indicate the superior performance of the proposed adaptive controller design framework.