task_id
stringclasses 6
values | env
stringclasses 6
values | suite
stringclasses 1
value | type
stringclasses 2
values | obs_dim
int64 11
376
| act_dim
int64 2
17
| max_steps
int64 50
1k
| baseline_ppo_return
float64 -4.2
5.8k
|
|---|---|---|---|---|---|---|---|
hopper-v5
|
Hopper-v5
|
gymnasium
|
locomotion
| 11
| 3
| 1,000
| 3,183.2
|
halfcheetah-v5
|
HalfCheetah-v5
|
gymnasium
|
locomotion
| 17
| 6
| 1,000
| 5,803.9
|
walker2d-v5
|
Walker2d-v5
|
gymnasium
|
locomotion
| 17
| 6
| 1,000
| 4,918.5
|
ant-v5
|
Ant-v5
|
gymnasium
|
locomotion
| 27
| 8
| 1,000
| 886.6
|
humanoid-v5
|
Humanoid-v5
|
gymnasium
|
locomotion
| 376
| 17
| 1,000
| 573.8
|
reacher-v5
|
Reacher-v5
|
gymnasium
|
manipulation
| 11
| 2
| 50
| -4.2
|
MuJoCo Gymnasium v5 Benchmark
PPO evaluation benchmark on MuJoCo Gymnasium v5 locomotion and manipulation environments.
Tasks
| Task ID | Name |
|---|---|
| hopper-v5 | Hopper-v5 |
| halfcheetah-v5 | HalfCheetah-v5 |
| walker2d-v5 | Walker2d-v5 |
| ant-v5 | Ant-v5 |
| humanoid-v5 | Humanoid-v5 |
| reacher-v5 | Reacher-v5 |
Usage
Submit evaluation results to this benchmark by adding .eval_results/*.yaml files
to your model repo, referencing this dataset's task IDs.
© 2026 ParamTatva.org
- Downloads last month
- 11