The following figures show the average returns of the Rainbow RL algorithm on the Unity Obstacle Tower Challenge.
Each step represents 250,000 frames for the training phase and 125,000 frames for the evaluation phase.
Further information about the challenge: https://www.aicrowd.com/challenges/unity-obstacle-tower-challenge
Obstacle Tower paper: https://arxiv.org/pdf/1902.01378.pdf
Further information about Google dopamine: https://github.com/google/dopamine
Rainbow paper: https://arxiv.org/pdf/1710.02298.pdf