Spaces:
Running
Running
| { | |
| "harness": "stock Codex CLI 0.160.0", | |
| "model": "gpt-6-astra", | |
| "effort": "high", | |
| "seed": 0, | |
| "episodes_per_task": 1, | |
| "mode": "stepped", | |
| "safety_timeout_sec": 28800, | |
| "fresh_workspace": true, | |
| "resume_trajectory": false, | |
| "source_jobs": [], | |
| "families": [ | |
| { | |
| "id": "task01", | |
| "name": "RoboCasa", | |
| "total": 10, | |
| "completed": 10, | |
| "pending": 0, | |
| "successes": 2, | |
| "valid_results": 10, | |
| "success_rate": 0.2, | |
| "input_tokens": 63285354, | |
| "cached_input_tokens": 62436736, | |
| "output_tokens": 204733, | |
| "usage_complete": 10, | |
| "control_frequency_hz": 20, | |
| "max_control_steps": 6000, | |
| "preflight_results": 0, | |
| "formal_results": 10, | |
| "modified_results": 10, | |
| "original_results": 0 | |
| }, | |
| { | |
| "id": "task02", | |
| "name": "LIBERO Long", | |
| "total": 10, | |
| "completed": 10, | |
| "pending": 0, | |
| "successes": 9, | |
| "valid_results": 10, | |
| "success_rate": 0.9, | |
| "input_tokens": 12586499, | |
| "cached_input_tokens": 12106752, | |
| "output_tokens": 93969, | |
| "usage_complete": 10, | |
| "control_frequency_hz": 20, | |
| "max_control_steps": 6000, | |
| "preflight_results": 0, | |
| "formal_results": 10, | |
| "modified_results": 5, | |
| "original_results": 5 | |
| }, | |
| { | |
| "id": "task03", | |
| "name": "RoboTwin", | |
| "total": 10, | |
| "completed": 10, | |
| "pending": 0, | |
| "successes": 7, | |
| "valid_results": 10, | |
| "success_rate": 0.7, | |
| "input_tokens": 15621828, | |
| "cached_input_tokens": 15138688, | |
| "output_tokens": 105566, | |
| "usage_complete": 10, | |
| "control_frequency_hz": 25, | |
| "max_control_steps": 7500, | |
| "preflight_results": 0, | |
| "formal_results": 10, | |
| "modified_results": 7, | |
| "original_results": 3, | |
| "deferred_tasks": 0 | |
| }, | |
| { | |
| "id": "task04", | |
| "name": "RoboDojo", | |
| "total": 42, | |
| "completed": 42, | |
| "pending": 0, | |
| "successes": 31, | |
| "valid_results": 42, | |
| "success_rate": 0.7380952380952381, | |
| "input_tokens": 259078099, | |
| "cached_input_tokens": 254678912, | |
| "output_tokens": 971405, | |
| "usage_complete": 42, | |
| "control_frequency_hz": 25, | |
| "max_control_steps": 7500, | |
| "preflight_results": 0, | |
| "formal_results": 42, | |
| "modified_results": 34, | |
| "original_results": 8 | |
| }, | |
| { | |
| "id": "task06", | |
| "name": "SIMPLE G1", | |
| "total": 12, | |
| "completed": 12, | |
| "pending": 0, | |
| "successes": 7, | |
| "valid_results": 12, | |
| "success_rate": 0.5833333333333334, | |
| "usage_complete": 12, | |
| "control_frequency_hz": 50, | |
| "max_control_steps": null, | |
| "max_control_steps_by_task": { | |
| "task06/01": 10000, | |
| "task06/02": 15000, | |
| "task06/03": 15000, | |
| "task06/04": 15000, | |
| "task06/05": 15000, | |
| "task06/06": 15000, | |
| "task06/07": 15000, | |
| "task06/08": 10000, | |
| "task06/09": 15000, | |
| "task06/10": 15000, | |
| "task06/11": 15000, | |
| "task06/12": 15000 | |
| }, | |
| "preflight_results": 0, | |
| "formal_results": 12, | |
| "modified_results": 0, | |
| "original_results": 12, | |
| "input_tokens": 106763761, | |
| "cached_input_tokens": 105331968, | |
| "output_tokens": 303566 | |
| } | |
| ], | |
| "usage_accounting": "reported-responses", | |
| "retry_policy_note": "The first six completed results use original Codex request defaults. Remaining tasks and two authorized fresh replacements use a 50-retry request policy. All new RoboCasa and LIBERO runs use explicit 50-retry SSE request policy; no automatic episode retry.", | |
| "family_protocols": { | |
| "task01": { | |
| "episodes": 1, | |
| "seed": 0, | |
| "max_steps": 6000, | |
| "control_frequency_hz": 20 | |
| }, | |
| "task02": { | |
| "episodes": 1, | |
| "seed": 0, | |
| "init_state_index": 0, | |
| "max_steps": 6000, | |
| "control_frequency_hz": 20 | |
| }, | |
| "task04": { | |
| "episodes": 1, | |
| "seed": 0, | |
| "max_steps": 7500, | |
| "control_frequency_hz": 25 | |
| }, | |
| "task03": { | |
| "episodes": 1, | |
| "seed": 0, | |
| "max_steps": 7500, | |
| "control_frequency_hz": 25, | |
| "safety_timeout_sec": 28800, | |
| "execution_mode": "stepped", | |
| "stop_on_success": true, | |
| "deferred_tasks": [], | |
| "native_tcp_tasks": [ | |
| "robotwin/lift-pot-aloha-agilex" | |
| ], | |
| "runtime_provenance": "See each episode: the other nine tasks retain their previously measured runtime." | |
| }, | |
| "task06": { | |
| "episodes": 1, | |
| "seed": 0, | |
| "control_frequency_hz": 50, | |
| "max_control_steps": null, | |
| "max_control_steps_by_task": { | |
| "task06/01": 10000, | |
| "task06/02": 15000, | |
| "task06/03": 15000, | |
| "task06/04": 15000, | |
| "task06/05": 15000, | |
| "task06/06": 15000, | |
| "task06/07": 15000, | |
| "task06/08": 10000, | |
| "task06/09": 15000, | |
| "task06/10": 15000, | |
| "task06/11": 15000, | |
| "task06/12": 15000 | |
| }, | |
| "timeout_s": 28800, | |
| "mode": "stepped", | |
| "instruction_policy": "original_native", | |
| "max_steps": null, | |
| "execution_mode": "stepped", | |
| "stop_on_success": true, | |
| "sim_mode": "mujoco_isaac", | |
| "dr_level": 0, | |
| "arm_gravity_compensation": true, | |
| "recording": true, | |
| "model": "gpt-6-astra", | |
| "reasoning_effort": "high", | |
| "service_tier": "default", | |
| "max_request_retries": 50, | |
| "automatic_episode_retries": 0, | |
| "fresh_session": true | |
| } | |
| }, | |
| "robotwin_harness": "Stock Codex CLI 0.160.0, gpt-6-astra, high, default service tier; current RoboEnv SDK/CLI; no Kinex agent runtime.", | |
| "native_tcp_repair_note": "Lift pot uses native RoboTwin TCP frames after the RoboEnv correction, with the existing reviewed instruction, unchanged native verifier, seed 0, 7500 action frames at 25 Hz, and one fresh stock Codex 0.160.0 / GPT-6 Astra high episode. Other results retain their historical runtime and protocol.", | |
| "simple_comparability": "Kinex 0.10.3 and ordinary Codex CLI 0.160.0 share the frozen RoboEnv source, native task instructions, seed and protocol. Each task records its measured budget and runtime. Only earlier native failures were rerun at 15000 steps; retained successes use 10000 steps. The current selection mixes budgets and is not a uniform-budget agent comparison. Prior interrupted attempts are retained separately; fresh sessions inherit no trajectories or resources. Wall time is not a controlled speed comparison." | |
| } | |