safe-learning

The interface between probabilistic model checking and data-driven policy learning.

Safe reinforcement learning for quadruped locomotion using Control Barrier Functions (CBF) - Zero falls, 99% safety rate, 90% speed retention with provable safety guarantees

CAIL (IROS 2025): constraint-aware behavior cloning with privileged training-time safety supervision for autonomous racing, without an additional safety filter at deployment.

Safe Online Control-Informed Learning (IEEE L-CSS 2025) — learning a robot's dynamics, objective and safety limits online, one measurement at a time, without leaving the safe set.

Executable research program for safe, data-efficient, learning-assisted PID autotuning and adaptive control.

Overview: safety-constrained, data-efficient Bayesian optimisation for robotic assembly (code private)