Odyssey launches PROWL agents to test Minecraft world models
Odyssey introduced PROWL, a system of reinforcement learning agents that explore game environments such as Minecraft. The agents receive rewards for surfacing discrepancies in physics, visuals, and actions that world models fail to predict. Detected failures are automatically turned into new training data that refines the models through an iterative loop without manual intervention. Odyssey shared the launch with a 43-second video of agent movement.
Congrats @UCL_DARK's @ahguzelUK on this internship success at @odysseyml!
Introducing PROWL! We’ve built RL agents that explore game environments, tasked with discovering failures in world models across physics, visuals, and actions. Those failures then become training data in an automated loop that advances world model performance.
What if world models could learn by discovery?
Today we’re sharing PROWL: RL agents that explore game environments, simulators, and eventually robots to discover failures in a world model.
This loop of learning is fully automated!