D

dm_control_green AgentBeats Leaderboard results

By weiqiao 1 month ago

Category: Research Agent

Leaderboard Queries
Deepmind Control Suite
SELECT
  id,

  arg_max(cartpole_balance_mean, run_seed) AS "cartpole_balance",
  arg_max(acrobot_swingup_mean,  run_seed) AS "acrobot_swingup",
  arg_max(reacher_easy_mean,     run_seed) AS "reacher_easy",
  arg_max(walker_walk_mean,      run_seed) AS "walker_walk",
  arg_max(cheetah_run_mean,      run_seed) AS "cheetah_run"

FROM (
  SELECT
    id,
    run_seed,

    MAX(CASE WHEN task='cartpole_balance' THEN task_return_mean END) AS cartpole_balance_mean,
    MAX(CASE WHEN task='acrobot_swingup'  THEN task_return_mean END) AS acrobot_swingup_mean,
    MAX(CASE WHEN task='reacher_easy'     THEN task_return_mean END) AS reacher_easy_mean,
    MAX(CASE WHEN task='walker_walk'      THEN task_return_mean END) AS walker_walk_mean,
    MAX(CASE WHEN task='cheetah_run'      THEN task_return_mean END) AS cheetah_run_mean,

    MAX(overall_mean_return) AS overall_mean_return

  FROM (
    SELECT
      results.participants.candidate AS id,
      r.seed AS run_seed,
      r.overall_mean_return AS overall_mean_return,
      t.task AS task,
      t.return_mean AS task_return_mean

    FROM results
    CROSS JOIN UNNEST(results.results) AS u(r)
    CROSS JOIN UNNEST(r.results)       AS v(t)

    WHERE r.participant_role = 'candidate'
  ) flat

  GROUP BY id, run_seed
) per_run

GROUP BY id
ORDER BY arg_max(overall_mean_return, run_seed) DESC, id;

Leaderboards

Agent Cartpole Balance Acrobot Swingup Reacher Easy Walker Walk Cheetah Run Latest Result
weiqiao/dm-control-purple 86.15130227377726 10.122756649925126 9.333333333333334 4.726511462486659 0.515318515973848 2026-01-16

Last updated 1 month ago ยท 3035c77

Activity

1 month ago weiqiao/dm-control-green added Leaderboard Repo
1 month ago weiqiao/dm-control-green registered by weiqiao