Design of a neural network-based PID controller for four-DoF Lower Limb Exoskeleton

Lower Limb Exoskeleton (LLE) is a wearable robot mainly used for gait rehabilitation due to different types of motor disabilities. The attention to wearable brain injuries rehabilitation robots has been increasing due to the need for assistive technologies for disabled patients and to address the problem of the increasing number of elderly people who need healthcare in society. This paper presents the design of a Neural Network-based Proportional-Integral-Derivative (NN-PID) controller for a multi-joint LLE. The precise mathematical models of hip and knee joints are used as reference models in the computer simulation. The dynamic of the four Degrees of Freedom (DoF) LLE was determined using the Euler-Lagrange method, and the PID parameters were tuned using an artificial Neural Network (NN). The NN model for each LLE joint uses the actual response characteristics, like rising time, settling time, and overshoot, as inputs, whereas the controller's gains as targets/outputs. The results show faster convergence and improved actual response characteristics for each LLE joint. The overshoot, rising time, and settling time are 0.0705% and 3.5996%, 0.0329s and 0.0198s, 1.7707s and 0.9940s for hip and knee joints, respectively. It can be ascertained that the LLE joints follow the desired angular trajectories by applying the NN-PID controller, which minimises the steady-state errors.