Codesota · Benchmark · MuJoCoHome/Leaderboards/Robotics, Control & RL/Continuous Control/MuJoCo
Google DeepMind

MuJoCo.

Physics engine for continuous control tasks like walking, running, and manipulation.

Paper ↗Leaderboard ↓
§ 01 · Leaderboard

Results by metric.

Found a wrong score or missing run?
Use row edits to send a sourced correction into moderation.
Add / edit result ↗Report issue ↗

Average Return

Mean episodic return averaged across HalfCheetah, Hopper, and Walker2d at 1M steps.

Higher is better

Trust tiers for Average Returnverifiedpapervendorcommunityunverified
RankModelTrustScoreYearLinksFix
01TD3
Mean of HalfCheetah-v4 (9583), Hopper-v4 (3134), Walker2d-v4 (4057) at 1M steps. CleanRL verified.
verified55922026Source ↗Looks wrong?
02SAC
Mean of HalfCheetah-v4 (9634), Hopper-v4 (2310), Walker2d-v4 (3591) at 1M steps. CleanRL verified.
verified51792026Source ↗Looks wrong?
03PPO
Mean of HalfCheetah-v4 (1442), Hopper-v4 (2382), Walker2d-v4 (2287) at 1M steps. CleanRL verified.
verified20382026Source ↗Looks wrong?
04TD-MPC2 (317M params)
TD-MPC2, 317M-param shared model. Mean normalized score across 15 DMControl tasks, 1M steps. ICLR 2024.
paper9602026Source ↗Looks wrong?
05TD-MPC2 (19M params)
TD-MPC2, 19M-param shared model. Mean normalized score across 15 DMControl tasks, 1M steps. ICLR 2024.
paper9532026Source ↗Looks wrong?
06FOWM
FOWM (Foundation Online World Models). Mean normalized score, DMControl 15 tasks. NeurIPS 2024.
paper9452026Source ↗Looks wrong?
07BRO
BRO (Best-of-N Robustness RL). Mean normalized score across DMControl tasks. ICML 2024.
paper9412026Source ↗Looks wrong?
08TD-MPC2 (5M params)
TD-MPC2, 5M-param model. Mean normalized score across 15 DMControl tasks, 1M steps. ICLR 2024.
paper9292026Source ↗Looks wrong?
09DreamerV3
DreamerV3. Mean normalized score across 15 DMControl tasks, 1M steps. From TD-MPC2 Table 1 comparison.
paper8972026Source ↗Looks wrong?
10TD-MPC
TD-MPC (original). Mean normalized score across DMControl tasks, 1M steps. ICML 2022 baseline from TD-MPC2 paper.
paper8572026Source ↗Looks wrong?
11DrQ-v2
DrQ-v2, pixel-based. Mean normalized score across 15 DMControl tasks, 1M steps. From TD-MPC2 Table 1.
paper7992026Source ↗Looks wrong?
12SAC (state-based)
SAC (state-based). Mean normalized score across DMControl tasks. Classic baseline from TD-MPC2 Table 1.
paper7772026Source ↗Looks wrong?
§ 04 · Submit a result

Add to the leaderboard.

← Back to Continuous Control