Title: Project Malmo: turning Minecraft into a testbed for the future of AI
Bio: Katja is a researcher in the Machine Intelligence and Perception group at Microsoft Research Cambridge. She is the research lead of Project Malmo, which uses the popular game Minecraft as an experimentation platform for developing intelligent technology. Her long-term goal is to develop AI systems that learn to collaborate with people, to empower their users, and help solve complex real-world problems.
Before joining Microsoft Research, Katja completed her PhD in Computer Science as part of the ILPS group at the University of Amsterdam, working with Maarten de Rijke and Shimon Whiteson on smart, self-learning search engines.
Jonathan Blow, Number None
Title: How do you make an AI that designs video games?
There have been a number of efforts to build AI that designs games, but these generally produce results that are not very interesting to play. Why? There are probably a lot of reasons, but one is that these projects focus on reproducing the most basic outward mechanical appearance of a game, ignoring almost everything that gives a game meaning. We take a look at what designers really think about when they make games, and we muse about what an AI for such tasks might look like.
Jonathan Blow is a game designer and programmer who lives in San Francisco. His most recent games are The Witness and Braid. He runs the studio Thekla, Inc., which creates mind-expanding games.
Simon Clavet (Joint with MIG), Ubisoft Montreal
Motion Matching for Realistic Animation in “For Honor”
Motion Matching is a brute force approach for creating responsive character movement from a large database of mocap data. We adopt a declarative animation philosophy, where instead of placing small animations in a big structure, we place event markup on top of long unstructured animations. This manual markup is necessary for logical information that can’t be inferred automatically, like stances and attack types. For navigation, we don’t have to manually organize transition animations like starts, stops, and turns. We just capture 5 or 10 minutes of a person running around and import it directly into the engine. At runtime, we continuously find the frame that best matches the current pose, the desired future trajectory and the properties of desired events. We simply transition with a small blend time to this winning place in the data. The resulting motion is almost indistinguishable from a raw mocap sequence, while being responsive enough for comfortable control.
Title: Avoiding Artificial Stupidity
Kevin Dill is a member of the senior technical staff at Lockheed Martin Rotary and Mission Systems, and the Chief Architect of the Game AI Architecture. He is a veteran of the game and military simulation industries with over 15 years' experience, and has worked on AI for everything from games (including several major hits, such as Red Dead Redemption, Iron Man, and Zoo Tycoon 2) to military training to emotive avatars. His professional interests include improved techniques for behavior specification, tactical and strategic AI, spatial reasoning, and believable characters. Mr. Dill was the technical editor for Introduction to Game AI and Behavioral Mathematics for Game AI, and a section editor for AI Game Programming Wisdom 4 and the Game AI Pro series. He is a prolific author and speaker, and has taught at Harvard University, Boston University, and Northeastern University.
Bio: Mark Walsh is founder of Motional A.i. and creator of Gary the Gull, the first interactive character in VR. With 18 years at Pixar, Mark was the writer-director of Partysaurus Rex and Animation Lead on Academy Award winners Ratatouille and Finding Nemo – specifically helping create the character of Dory.
After contributing to the Incredibles, Bug’s Life, Monsters Inc, and the Good Dinosaur, Mark created ads for Up, Toy Story 3, Visa, Aflac, Target, Dolby, & Peugeot.
Now with Motional, Mark is bringing Pixar empathy, & character to A.i.