Near-perfect point-goal navigation from 2.5 billion frames of experience - discu.eu

Reddit

"DD-PPO: Near-perfect point-goal navigation from 2.5 billion frames of experience", Wijmans & Kadian 2020 {FB} [PPO scaling w/many-GPU-envs: synchronous model updates, shortcircuit env rollouts] https://ai.facebook.com/blog/near-perfect-point-goal-navigation-from-25-billion-frames-of-experience/ 6 comments 22/1/2020 reinforcementlearning