Nonlinear Intelligent Model Predictive Control of Mobile Robots

This thesis presents a framework for an artificial neural network (ANN) model-based nonlinear model predictive control of mobile ground robots. A computer vision analysis module was first developed to extract quantitative position information from onboard camera feed with respect to a prescribed path. Various strategies were developed to construct nonlinear physical plant models for model predictive control (MPC), including the physics-based model (PBM), the ANN trained on PBM-generated data, the ANN trained on test-captured data, and the ANN initially trained on PBM-generated data and then retrained with captured data. All the models predict physical states and positions of the robot in the future horizon using the current control signals and the information obtained by the computer vision analysis. Model predictive controllers based on these models and real-time optimization were also developed, and were able to determine optimal control signals in the future horizon, enabling the robot to follow the designated path. Path following experiments were then conducted on a test track to evaluate nonlinear MPC performance. It is found that both the PBM and the ANN model allow accurate path following through nonlinear MPC with an error metric of 1 cm (i.e., the average deviation of the robot from the designated path). More importantly, incorporating test-collected data into the ANN retraining to consider non-ideal factors not captured by PBM further improves the path following accuracy by 30.0%. The developed framework paves the way towards a new paradigm to develop autonomous robots with anomaly mitigation and system resilience