eval/scene_pusht.json

{
  "schema_version": "0.1",
  "source": {
    "paper": "https://arxiv.org/abs/2303.04137",
    "arxiv": "https://arxiv.org/abs/2303.04137",
    "model_variant": "diffusion",
    "evaluated_by": "OpenRAL:openral benchmark scene",
    "reproduced_locally": true,
    "reproduction_planned": null,
    "reproduction_cli": "openral benchmark scene --config scenes/benchmark/pusht.yaml --rskill rskill://rskills/diffusion-pusht",
    "table": null,
    "status": "reproduced"
  },
  "benchmark": {
    "name": "pusht",
    "dataset": null,
    "protocol": "1 episodes, seed=0, success_key=is_success, max_steps=300",
    "robot": "pusht_2d",
    "simulator": "pusht"
  },
  "eval_config": {
    "n_episodes": 1,
    "seeds": [
      0
    ],
    "success_key": "is_success",
    "max_steps": 300,
    "vla_id": "diffusion",
    "weights_uri": "rskill://rskills/diffusion-pusht"
  },
  "results": {
    "pusht/0_success_rate": 1.0,
    "avg_success_rate": 1.0,
    "n_tasks": 1,
    "n_episodes_per_task": 1,
    "n_episodes_total": 1,
    "mean_step_latency_ms_avg": 246.9849692727758,
    "mean_coverage_iou": 1.0
  },
  "baselines": {},
  "trace_id": null
}