🧭 3D Eligibility Traces: TD(lambda) Credit Assignment, Forward and Backward View on a Sparse-Reward Grid World
Four Blender-built boards of value bars on one random walk: the exact value function, and three TD learners with lambda 0, your lambda and lambda 1, with glowing eligibility traces showing how far one reward propagates back, checked against the Bellman solution and the offline lambda-return theorem.
⚙ Under the hood
Four Blender-built boards of value bars on one random walk: the exact value function, and three TD learners with lambda 0, your lambda and lambda 1, with glowing eligibility traces showing how far one reward propagates back, checked against the Bellman solution and the offline lambda-return theorem.
Eligibility TracesTD LambdaTemporal Difference LearningCredit AssignmentReinforcement LearningLambda ReturnValue FunctionGrid World
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install