Contract-based shielding in multi-agent reinforcement learning, with automata and contract utilities, shielded environment wrappers, RL baseline algorithms, and 6 benchmark environments.

Quantitative reward monitors in reinforcement learning (using $\mathsf{LTL}_f$), synthesising dense reward functions for temporally extended goals with formal specifications, including safety-aware properties.

A co-maintained library for multi-agent and single-agent safe reinforcement learning, with common constraints, environments, and reusable research tooling.

Automatically synthesises manufacturing process controllers from ConGolog resource models and situation calculus basic action theories.

Octree-based mesh mapping software for finite element workflows, with parallel indexing and higher-accuracy interpolation strategies.

Automatically synthesises manufacturing process plans from labelled transition system models of recipes and production resources.