AI for Resilient Self-Assembling Robotic Matter: Strengths, Vulnerabilities and Defense

About the project

About the Project What happens when a robot can rebuild its body after damage? Resilient modular machines could continue functioning despite losing parts or connections, but their ability to recover also raises a security question: how can people protect themselves from a machine that behaves dangerously and is difficult to disable? This project will investigate both the strengths and vulnerabilities of self-assembling robotic matter, using systems such as Roblets as a starting point. It will examine how AI can support the development of resilient modular robots and how AI-assisted testing can identify weaknesses relevant to defensive intervention and human protection. The robot will be considered as an interconnected mechanical, electrical and control system. Mechanical joints determine its structure and movement. Electrical contacts, such as sockets and tabs, can provide pathways for power and communication between modules. Control methods coordinate the behaviour of the assembled machine. The project will investigate how disruptions to these layers affect the robot's functionality, recovery and controllability. One strand of the research will use AI or optimisation to improve resilience, for example by identifying alternative configurations or recovery strategies after module or connection failures. A complementary strand will develop AI-assisted vulnerability assessment: systematically searching for failure scenarios and adversarial disturbances that reveal the limits of that resilience. The findings will support the investigation of defensive measures, including detection, containment and reliable deactivation of unsafe robotic behaviour. The student will combine simulation with experiments on a modular robotic platform. Controlled tests will examine mechanical damage, electrical connection failures and control disturbances, individually and in combination. Evaluation will consider retained functionality, recovery time, the extent of disruption required to stop unsafe behaviour, and the reliability of proposed protective measures. The emphasis can be adapted to the student's background, including machine learning, electrical interfaces, mechanical module design or control. The research will contribute methods for understanding resilient machines as physical systems and for maintaining human control as their ability to adapt and recover increases. The project will be based in the Sheffield Microrobotics Lab in the School of Electrical and Electronic Engineering at the University of Sheffield. For the detail of our projects, please visit https://shuhei.net/

Research areas