We develop learning-enhanced planning and feedback control methods that satisfy explicit temporal tasks, safety constraints, robustness requirements, and human-interaction objectives.
This direction integrates control theory, formal methods, optimisation, reinforcement learning, discrete-event synthesis, and robot learning. Safety and task satisfaction are treated as design constraints rather than post-deployment evaluation metrics.
Applications include nonlinear and hybrid systems, robotic manipulation, collaborative and wearable robots, manufacturing scheduling, fault-tolerant operation, and coordination of multiple autonomous platforms.
High-level requirements express ordering, deadlines, persistence, recovery, and interaction. We develop semantics and compilation methods that map these requirements into tractable optimisation, barrier-function, Lyapunov, automata, or model-predictive-control constraints.
Neural policies and reinforcement learning adapt well but may fail outside the training distribution. We combine learning with CLF/CBF filters, robust MPC, formal synthesis, recursive feasibility, and uncertainty adaptation so that improvement remains inside an explicit safety envelope.
Manipulation, scheduling, multi-agent coordination, and shared autonomy couple discrete decisions with continuous dynamics. We study hierarchical decomposition, graph reasoning, fault tolerance, real-time repair, and preservation of human agency.
Temporal specifications, human intent, dynamics, faults, and resource constraints
Reachability, barrier margins, recursive feasibility, uncertainty sets, and event structure
MPC, CBF/CLF, reinforcement learning, formal and hierarchical control
Safe motion, resilient operation, coordinated autonomy, and human-centred interaction
