Deep RL Timeline

  1. Atari (DQN)

    [Deepmind]

    Pong Atari gameplay
    Pong
    Enduro Atari gameplay
    Enduro
    Beamrider Atari gameplay
    Beamrider
    Q*bert Atari gameplay
    Q*bert

  2. 2D locomotion (TRPO)

    [Berkeley]

    A red and purple two-legged walker on a checkerboard floor

  3. AlphaGo

    [Deepmind]

    AlphaGo versus Lee Sedol, with the Go board and player shown together

  4. 3D locomotion (TRPO+GAE)

    [Berkeley]

    A humanoid falls on a checkerboard floor at iteration 160 while learning 3D locomotion

  5. Real Robot Manipulation (GPS)

    [Berkeley, Google]

    A PR2 robot manipulates a wooden shape-sorting cube at iteration 15

  6. Dota2 (PPO)

    [OpenAI]

    OpenAI's Dota2 bot plays a one-on-one match in the middle lane

  7. DeepMimic

    [Berkeley]

    A montage of DeepMimic simulated character skills compared with reference motions

  8. AlphaStar

    [Deepmind]

    AlphaStar's StarCraft II MMR results compared with human player percentiles and Grandmaster league

  9. Rubik’s Cube (PPO+DR)

    [OpenAI]

    A black robotic hand with an orange and blue wrist holds a Rubik's Cube