Microsoft Research's Project Paidia team, in collaboration with game developer Ninja Theory, is exploring new opportunities created by reinforcement learning. Its goal is to democratise state-of-the-art reinforcement learning techniques for any developer through Azure Machine Learning, and to provide access to training infrastructure and easy integration tools.
Microsoft Research's project web pages say:
A key focus of our work within the Game Intelligence theme is to develop reinforcement learning agents that can engage and collaborate with players in exciting and novel ways. We went looking for a test bed that could help us really push the boundaries in this space and were fortunate to find Bleeding Edge, the latest video game from Ninja Theory.
A member of the Xbox family, Ninja Theory is located in Cambridge (like our Microsoft Research Lab) and ideally suited for collaboration at the forefront of AI research. We’re working with the Ninja Theory team to enable experimentation on Bleeding Edge, and proud to showcase some of the game play learned by our agents, including examples of tactical and strategic collaboration.
Bleeding Edge is a 4v4 team-based game set in a post-industrial steam-punk world. Teams consist of characters with diverse skill sets (assassins, tanks, and healers) that compete against the opposing team for control of platforms. The key to mastering this fast-paced game is effective team play. Players must make coordinated strategic and tactical choices to win the upper hand against well-organized opponents. Bleeding Edge effectively tests human players’ ability to collaborate, and is ideal for ascertaining whether trained agents can work together as successfully.