Coordination Ring

coord ring lots of training

same training method as original. just run a lot more. 305 steps

Evaluated Unknown · Uploaded Unknown by Rowan G. Last edited by Rowan G.

agent_0

Observation and actions

{
  "actions": {
    "action_selection": "sample"
  },
  "observation": {
    "agent_view_size": 2,
    "shape": [
      5,
      5,
      30
    ]
  }
}

agent_1

20M timesteps · 19.99M steps

Observation and actions

{
  "actions": {
    "action_selection": "sample"
  },
  "observation": {
    "agent_view_size": 2,
    "shape": [
      5,
      5,
      30
    ]
  }
}

Statistics across all 5 episodes

episodes: 5
episodes with a correct delivery: 5
mean correct deliveries: 17
mean team reward: 340
policy: trained
status: complete
total correct deliveries: 85
total wrong deliveries: 0

Selected episode

Select an episode to inspect its recording.

Evaluation protocol and metric definitions

{
  "action_selection": "sample",
  "backend": "gpu",
  "episode_seeds": [
    1000,
    1001,
    1002,
    1003,
    1004
  ],
  "gif_frame_ms": 100,
  "gif_stride": 1,
  "policies": [
    {
      "agent": "agent_0",
      "chef": 1,
      "color": "red",
      "policy_sha256": "fe793fb313762382b7c2d433d83c2a610e8386386698b4b1ad3067f10c49a21b",
      "training_seed": 901
    },
    {
      "agent": "agent_1",
      "chef": 2,
      "color": "blue",
      "policy_sha256": "fe793fb313762382b7c2d433d83c2a610e8386386698b4b1ad3067f10c49a21b",
      "training_seed": 901
    }
  ],
  "policy": "trained",
  "policy_sha256": "fe793fb313762382b7c2d433d83c2a610e8386386698b4b1ad3067f10c49a21b",
  "recorded_episode": 0,
  "recording_coverage": "all_episodes_full_resolution",
  "replay_frame_ms": 100,
  "replay_schema_version": 1,
  "replay_stride": 1,
  "reward": "shared raw environment reward counted once; no shaping",
  "state_action_convention": "steps[0] initial; steps[t] after attached action t",
  "upstream_revision": "976aeb152cb184a5095021968bba94da96eb6394"
}

Effective environment configuration

{
  "action_meanings": {
    "0": "right",
    "1": "down",
    "2": "left",
    "3": "up",
    "4": "stay",
    "5": "interact"
  },
  "config": {
    "agent_view_size": 2,
    "indicate_successful_delivery": false,
    "layout": "coord_ring",
    "max_steps": 400,
    "negative_rewards": true,
    "observation_type": "default",
    "random_agent_positions": true,
    "random_reset": false,
    "sample_recipe_on_delivery": false,
    "start_cooking_interaction": false
  },
  "layout": "coord_ring",
  "name": "overcooked_v2",
  "renderer": {
    "cook_ticks": 20,
    "kind": "overcooked"
  }
}

Evaluation code and dependency provenance

{
  "git_commit": "5632bd7dc714a85b138cfdea5ea3976bdbcd5e03",
  "git_dirty": true,
  "sources": {
    "algorithm_sha256": "5cbd7c4e2faba93004cb701e5ea8ec7778c132aea6ed7862b083fde21c22fca5",
    "environment_sha256": "a535b216a921887cb5122f2c8150ce81f469522141bb37418af9d92086b573bd",
    "evaluation_core_sha256": "4389100d10a4210408d0901035c968e42577c2c0332f09f033aaad2525cbf293"
  },
  "training_checkpoint_provenance": {
    "run:3e36ab97-b1a8-4aaa-8359-c5f39e7c6070:sha256:fe793fb313762382b7c2d433d83c2a610e8386386698b4b1ad3067f10c49a21b": {
      "training_resumes": []
    }
  },
  "upstream_revision": "976aeb152cb184a5095021968bba94da96eb6394"
}

Evaluation device (separate from training device)

{
  "architecture": "x86_64",
  "backend": "gpu",
  "devices": [
    {
      "kind": "NVIDIA GeForce RTX 3070",
      "platform": "gpu"
    }
  ],
  "logical_cpu_count": 16,
  "os": "Linux",
  "os_release": "6.18.40.1-microsoft-standard-WSL2",
  "python": "3.12.3",
  "versions": {
    "flax": "0.10.2",
    "jax": "0.4.38",
    "jaxlib": "0.4.38",
    "numpy": "1.26.4",
    "optax": "0.2.4"
  }
}