With the development of intelligent technology, the traditional paradigm of robots operating in fixed-structured environments is being redefined. Robots are increasingly integrated into various aspects of human life, taking on diverse roles in industries such as healthcare, manufacturing, and service industries. These evolving working environments impose new demands on robotic systems, particularly their ability to interact autonomously with their surroundings. To ensure safe task execution, robots must exhibit compliant behaviors that enable them to manage unexpected collisions and adapt their responses dynamically. Furthermore, achieving high-level autonomy requires robots to learn and refine interaction strategies, allowing them to function as intelligent agents rather than passively executing preprogrammed tasks. Beyond safety and adaptability, interaction performance is a critical factor in robotics. Robots must maintain stable and precise interaction behaviors across various applications, such as accurate force tracking in curved surface polishing, precise position control in robotic surgery, and maximizing energy efficiency in robotic cleaning. However, real-world constraints, including the lack of prior environmental knowledge and inevitable external disturbances, pose significant challenges to achieving stable interactions. To address these challenges, this thesis focuses on developing control strategies for robotic interaction with unknown environments by integrating both interaction performance optimization and adaptive compliance. The main contributions are outlined as follows:
1. To improve convergence rate and tracking accuracy in admittance-controlled robots, a robust fixed-time controller based on non-singular sliding mode control is proposed. By introducing a non-singular sliding manifold, the approach eliminates singularity issues in conventional sliding mode control and guarantees fixed-time initial condition-independent convergence. To further enhance robustness against unknown or inaccurate robot dynamics, radial basis function neural networks (RBFNNs) are integrated for adaptive parameter learning, enabling real-time compensation for disturbances and enhancing interaction stability in dynamic environments. In addition, an optimal admittance parameter design is integrated into the framework to ensure performance-oriented interaction, balancing trajectory tracking accuracy with compliant force regulation.
2. For surface scanning tasks in unknown environments, an optimal control strategy is integrated into the admittance control framework to optimize interaction performance. Specifically, considering non-planar surfaces, the environmental equilibrium position is explicitly incorporated into the model, and a generalized iterative formulation is introduced to represent arbitrary surface variations. A discrete adaptive dynamic programming (ADP) approach is developed to learn optimal admittance parameters in real time. The proposed hybrid iteration strategy guarantees both robust initialization and fast convergence, making it well-suited for real-time optimal control applications.
3. Based on the established interactive system, a control strategy better aligned with human operational habits is developed. Inspired by human ultrasound scanning, the workspace of the end-effector is decomposed into orthogonal motion-controlled subspace, handling trajectory tracking along the object surface, and admittance-controlled control subspace, regulating contact force perpendicular to the surface. The orientation regulation of the end-effector based on the 3D point cloud technique is also integrated to better fulfill the task requirements. To address the nonlinear and discontinuous nature of frictional forces, an augmented neural network incorporating sigmoid-jump neurons is applied for friction compensation. Additionally, environmental factors are explicitly considered when determining the optimal admittance model. By redefining the desired trajectory in the admittance subspace relative to the environmental equilibrium position, the end-effector maintains stable interaction with consistent contact force while optimizing interaction performance across varying surfaces.
4. To address the inaccurate representation of linear models for soft environments, particularly viscoelastic materials, a nonlinear Hunt-Crossley (HC) model is adopted to better describe the environmental dynamics, with model parameter identification eliminating dependence on prior environmental knowledge. In the presence of inevitable model uncertainties, a disturbance term is explicitly added to the HC model, and robust control strategies are developed accordingly. Specifically, an auxiliary control system with a modified cost function is introduced to transform the high-complexity robust control problem of the original nonlinear system into a simpler optimal control problem. Finally, the neural network technique is utilized to approximate the optimal solution, thereby avoiding complex mathematical derivations.
5. To develop a robotic ultrasound scanning system to deal with the shortage of expert sonographers, an adaptive ultrasound scanning system is developed builds on previous advances in nonlinear viscoelastic tissue representation and hybrid admittance control. Unlike conventional approaches that rely on linear environment models or fixed interaction parameters, the proposed system adaptively regulates both probe penetration and orientation in real time, ensuring safe contact forces and stable acoustic coupling. A key innovation is the fusion of nonlinear modeling with online admittance adaptation, which enables reliable imaging quality across different tissue types and phantoms while maintaining optimal interaction performance. This contribution advances the practical applicability of robotic ultrasound scanning by bridging theoretical control design with experimental validation in clinically relevant scenarios.
Simulations and experiments confirm that the proposed optimization-based admittance control strategy effectively enhances safe, stable, and efficient robot-environment interactions in unknown settings. This research forms a crucial foundation for advancing intelligent robotic systems and driving the evolution of a smart society.
| Date of Award | 6 Jun 2025 |
|---|
| Original language | English |
|---|
| Awarding Institution | - The University of Manchester
|
|---|
| Supervisor | Guang Li (Main Supervisor) & Long Zhang (Co Supervisor) |
|---|