collecting_childrens_toys 200, search v2 replay
BEHAVIOR-1K 2026 challenge, issue 74, recov50 scripted recoveries, seed 0, Oct 23 to 29, run v2_smoke
failed The phase is to pick up the dice from the bed. The dice was out of view at the restore, and the search found it with the head camera after 8 s. The reach plateaued 5 times and the stall reset fired twice. It switched to the left arm after the plateaus. It ended failed: timeout in approach, goal error 0.19 m.
Other search v2 runs of this case: v2c_fixed, failed, v2d_orig_grab, pass, exp_A, pass, exp_B, pass, exp_C, pass, expC_harm, pass, v3_tuned, pass, v5b_on_a, pass, v5b_off_a, pass, v6_dev, pass, v6b_dev, pass, v6c_dev, pass.
The clips play at 2 times sim speed by default, and the speed menu changes that. The timeline shows the recoverer's stage, whether the target is inside the head camera's view while it reaches (from the gaze log), and the labelled events. Click a row to jump there. The robot panel sums up the attempt and the phases list follows the stages.
- Task
- -
- Instruction
- -
- Step
- -
- Sim time
- -
- Q
- -
- Driving
- -
Timeline
Q over time
Robot
- State
- -
Phases
The instruction follows the current phase and moves on when the phase detector confirms it. A typed instruction overrides it until you go back to the task sentence.
Last command
| part | 1 | 2 | 3 | 4 | 5 | 6 | 7 | gripper |
|---|
Look-around
Events
- Nothing yet.
Stage timeline
| stage | from | to | detail |
|---|---|---|---|
| Search | 0:00 | 0:06 | |
| Drive to the target | 0:06 | 0:08 | |
| Approach | 0:08 | 0:23 | 1 base move, 1 contact |
| Stall reset | 0:23 | 0:29 | stall check in approach at 0.49 m: touching bed_ivdnny_0; back the hand off 0.08 m; backed off (0.066 m left); arm back to its start pose; arm at its start pose, replan |
| Approach | 0:29 | 1:09 | 4 base moves, 3 reach plateaus, 3 contacts, switch to the left arm |
| Stall reset | 1:09 | 1:18 | stall check in approach at 0.26 m: other arm; back the hand off 0.08 m; backed off (0.045 m left); arm back to its start pose; arm at its start pose, replan |
| Outcome | 1:18 | 1:18 | failed: timeout in approach, goal error 0.19 m |
Plan and path maps
The plan map is where the search meant to look, from the restore. The path map is where the base went: the search path, then the recovery path.
Earlier attempts
| run | outcome | cause | length |
|---|---|---|---|
| pilot | failed | not visible after looking | 0:19 |
| a_cont | failed | not visible after looking | 0:19 |
| search_smoke | pass | - | 0:17 |
Old clips of this case: search v1 before and after, failure clip, not visible.
All cases. Results come from the attempt's result.json in the recov50 v2_smoke run, seed 0.