Online neural sliding mode control with guaranteed stability

Abstract

Model-based control systems allow engineers to leverage known system information in controllers, often resulting in improved control accuracy or robustness. However, these system models can change over time, be difficult to obtain for complex systems, or fail to capture higher-order effects in the system. In this work, we use a highly robust model-based control methodology, sliding mode control, without the need for an a priori model. We use a neural network to learn an approximate system model fully online, with no pretraining. The neural network term is found to increase control accuracy while alleviating chatter, the primary drawback of conventional sliding mode control. The controller is verified by simulating a non-planar 3 degree-of-freedom robotic arm picking up a load, and the proposed controller is shown to outperform conventional model-based sliding mode control, especially when the built-in load margin on the end effector in the conventional model-based controller is exceeded.

Keywords:

• Adaptive control
• neural networks
• learning-based control
• online learning
• sliding mode control

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Jacob G. Elkins is supported by the Department of Defense (DoD) through the National Defense Science and Engineering Graduate (NDSEG) Fellowship Program.