Reinforcement Learning
A collection of reinforcement-learning experiments exploring decision-making, navigation and emergent behaviour — including obstacle avoidance and predator-versus-prey environments.
A set of experiments rather than a product: purpose-built environments used to study how an agent learns to act when it is given an objective and nothing else.
The environments included obstacle avoidance, predator-versus-prey scenarios, and simple worlds built specifically to make a policy's behaviour legible while it trained.
The interesting results were rarely the scores. They were the strategies that appeared without being designed — the point of the exercise, and the reason the work fed into later projects that had to make decisions under uncertainty.
Research
Research rather than a product. Nothing was released publicly.