version int64 | codex_model string | codex_reasoning_effort string | environments dict | episodes list |
|---|---|---|---|---|
2 | gpt-6-astra | high | {
"libero": {
"benchmark": "libero-plus",
"env_root": "<LIBERO_PLUS_ROOT>",
"sim_python": "<SIM_PYTHON>"
}
} | [
{
"environment": "libero",
"seed": 0,
"init_state_index": 0,
"max_native_steps": 600,
"depth_estimation": true,
"task": "libero_goal/106",
"reference": {
"video": "libero_goal/references/base_28/videos",
"trajectory": "libero_goal/references/base_28/trajectory.json",
"t... |
LIBERO-Plus Astra High: 280 assigned trials
All 280 entries from the assigned wu_zenan queue, one trial per task variant. Includes successes and failures.
| Suite | Valid trials | Successes | Success rate |
|---|---|---|---|
| libero_goal | 81 | 57 | 70.37% |
| libero_spatial | 57 | 46 | 80.70% |
| libero_object | 71 | 70 | 98.59% |
| libero_10 | 71 | 34 | 47.89% |
Protocol: GPT-GEAR; gpt-6-astra with high reasoning effort via Codex ChatGPT login; seed=0, init_state_index=0, max_native_steps=600, depth_estimation=true. ICL contains official demonstration video frames sampled at1 FPS, trajectory and text.
Each suite directory contains episodes/task_ID/attempt-XXXX/trajectory.zip (offline HTML, public tool events, observations and video), result.json and configuration.json; references/ contains the corresponding demonstration videos, trajectories and text. See results.csv and evaluation_config.json for ordering and configuration.
Source: https://github.com/6UOOON9/GPT-GEAR ; official reference source: https://huggingface.co/datasets/uooon/reference_demo . Environment: LIBERO-plus commit4976dc30028e805ff8094b55501d532c48fec182.
Public trajectory exports exclude hidden reasoning, system/developer messages and credentials. Local raw records are retained separately. Aggregation uses official env.check_success and valid completed trials only. Simulation steps are not model-call counts.
- Downloads last month
- 97