Live demo

Distributed
Q-Learning

A grid-world navigation agent trained across browser GPU workers. Each round, workers run Q-learning episodes in parallel on the GPU, then the coordinator merges their Q-table updates. Watch the value landscape and greedy policy evolve in real time.

Not started
rl-agent @ waiting… idle
Unexplored Learning Learned Wall Start Goal
Round
0 / 0
Goal rate
Avg steps
Episodes
0
Workers 0
--:--:--warppool rl-agent · v0.3.1
Distributed Q-learning. Workers run episodes on GPU; coordinator merges Q-tables each round.