Physics-Informed Learning Meets Classical Model-Based Control for High-DoF Robot Manipulators Kingston University in United Kingdom

This PhD project at Kingston University offers an exciting opportunity to advance the state of the art in control systems for high-degree-of-freedom (DoF) robotic manipulators. The research aims to bridge classical model-based control and modern data-driven approaches by developing, integrating, and rigorously comparing three families of controllers: model-reference adaptive laws, sliding-mode robust schemes, and Model Predictive Control (MPC) enhanced with Physics-Informed Neural Networks (PINNs).

Experiments will be conducted in the university's Robotics Lab using advanced hardware such as the 7-DoF Franka Research 3 (Franka Emika Panda) and the 6-DoF industrial Fanuc M-10iA manipulators. The project builds on foundational work in adaptive control for manipulators, re-implementing classic schemes to demonstrate robust tracking and guaranteed convergence even under significant parameter uncertainty. Sliding-mode controllers will be explored for their disturbance rejection capabilities, with attention to practical issues like control chattering and adaptation speed.

In parallel, the student will develop MPC frameworks augmented by PINN-based surrogates, training neural networks that embed the manipulator's differential equations directly into their architecture. This approach leverages automatic differentiation to accelerate gradient computations, enabling near-real-time constraint enforcement and trajectory planning. The comparative study will introduce systematic perturbations—such as varying link masses, payloads, and joint friction—to evaluate each method's tracking accuracy, robustness, computational overhead, and tuning complexity.

The fusion of adaptive MPC and learning-informed controllers for high-DoF arms is a novel research direction, with tube-based sliding-mode MPC frameworks hinting at new possibilities for combining robustness and constraint handling. Embedding PINN surrogates into these architectures could further enhance performance under severe model mismatch.

Candidates will gain expertise in MATLAB, Python, deep-learning frameworks (TensorFlow/PyTorch), automatic differentiation, and hands-on robotics experimentation. Deliverables include open-source implementations, benchmark studies, and innovative hybrid control designs, contributing to both academic research and industrial applications in manipulators and autonomous systems.

The societal and economic impact of this research is significant. Improved controller robustness and adaptability will enable robots to track complex trajectories more accurately, require less calibration, and accommodate payload or model changes. In industrial settings, these advances can reduce downtime, lower maintenance costs, and accelerate production cycles, ultimately lowering the cost of manufactured goods. Enhanced safety and energy efficiency will support more flexible cobotic deployments and help factories meet environmental targets.

Funding for this position is available through the Kingston University Graduate School studentships competition for October 2026 entry. Applicants should have a strong background in engineering, computer science, or robotics, with experience in relevant programming and deep-learning tools. The application deadline is March 4, 2026. For further details and to apply, visit the Kingston University PhD Studentships and Faculty research pages.