Producing a character animation system for games has a multitude of unique challenges – extremely high performance requirements, control responsiveness, and dealing with the complexity of interacting with the environment and other characters. In this talk, Simon will present the current state of the art as used in production – a technique called Motion Matching recently shipped as part of Ubisoft’s fighting game For Honor, which compared to traditional animation systems drastically reduces the complexity while providing more realistic, more diverse, and more detailed animation. He will then show how large amounts of motion capture data used in conjunction with Neural Networks can produce powerful models useful for a number of common tasks in character animation such as fixing corrupt motion data, automatic stylization of motion data, animation compression, and high-quality locomotion on complicated terrain. Finally, he will discuss the intersection between robotics and virtual characters, and how Deep Reinforcement Learning may soon completely revolutionize game animation.