← All resources

Reinforcement Learning 04 : Motion Fields For Interactive Character Animation

Auto-discovered

Video / channel Advanced

Animation Tech

Visit resource ↗

Description

Reinforcement Learning 04 : Motion Fields For Interactive Character Animation is a YouTube implementation talk on the Animation Tech channel where Jerome rebuilds ideas from the 2010 Motion Fields paper in a reinforcement-learning-for-animation series.

It covers a motion database of pose-plus-velocity states, similarity weights and nearest-neighbor lookup, integrating blended velocities while anchoring into known data, greedy direction control versus fitted value iteration, separate walk and jog value functions.

It is for technical animation programmers comfortable with skeletons, offline clips, and basic RL vocabulary such as rewards and value functions. PyTorch familiarity helps with batch neighbor queries. The level is advanced relative to introductory blending videos.

The resource is a free YouTube video with live coding, paper math on screen, and demos of wandering and directed locomotion. Prior series episodes supply shared helpers.

It is useful when studying interactive locomotion between motion matching and RL, and when you want a code walkthrough of motion-field construction rather than the paper alone.