Transforming animation with machine learning (2021) (medium.com)

Embark Studios has significantly transformed its animation workflow by harnessing machine learning to automate the movement of game characters. Instead of manually animating actions, the team developed AI agents trained through reinforcement learning to independently navigate and respond to their environments, mimicking realistic movement behaviors. This approach not only alleviates the bottleneck of traditional animation processes but also enhances gameplay immersion by allowing characters to adapt dynamically to unique in-game scenarios, such as tripping or being hit by objects. This innovation is a game changer for the AI/ML community as it merges advanced machine learning techniques with practical game design, enabling faster content creation with a smaller team. The new workflow shifts the role of animators and designers from detailed manual animation to guiding the behavior of AI agents. This requires a cultural adjustment in thinking about content creation, focusing on high-level descriptions of desired movements rather than precise animation curves. As Embark continues to refine this technology within the constraints of game engines and physics simulation, it sets a precedent for how AI can elevate the creative processes in the gaming industry, ultimately leading to unexpected and emergent gameplay experiences.