CADR-RAS Webinar: Robust coordinated autonomy through Dec-MCTS with communicated behavioural intent

We are developing coordination algorithms for enabling coordinated manoeuvres that are tightly coupled, robust, interpretable, and applicable to defence-relevant missions such as scouting and target tracking. This project aims to address several fundamental challenges revealed through our high-TRL defence projects via three inter-related threads: (1) We have developed a new representation for encoding “behavioural intent” for communicating between robots, formulated as Behaviour Tree messages within the Dec-MCTS coordination algorithm. (2) We are developing neural network models for reinforcement learning of uncertain reward functions with stability certifications from agents collecting sparse data samples. (3) Lastly, we are generalising coordinated information gathering approaches for scenarios where the environment can dynamically influence perception performance, such as where there is variable lighting, smoke, or irregular clutter and occlusions.