Recorded 2026-11-05 on ankac/alm-dat-main at 59bdf62d0 (today's main, the #4222 storage, ALM-DAT with decision D3). Same recipes as the campaign, 1280x720 per panel, 30 frames per second. The metric floor of that revision was still in place; the final metric rule (D15) came after these runs.
alm_dat, both towels are held to the release like under VBD. With the single forced detection they are lost at the lift, at 5.50 s and 6.47 s, the same instants as in the campaign. Nothing else differs between the two ALM-DAT columns.r5_franka_towel_contact_v2Panels, left to right: VBD | ALM-DAT with two detections per substep | ALM-DAT with one detection per substep
| Configuration | Outcome | Self-crossings, max pairs (frames with / frames) | Jitter [mm] | Wall [min] |
|---|---|---|---|---|
| VBD, benchmark default | held | 1 (5/2100) | 22.1 | 34 |
| ALM-DAT, two detections per substep (D3) | held | 7 (34/2100) | 28.6 | 27 |
| ALM-DAT, one detection per substep (the removed forcing) | lost at 5.50 s | 11 (2027/2100) | 10.4 | 27 |
r7_h1_towelPanels, left to right: VBD | ALM-DAT with two detections per substep | ALM-DAT with one detection per substep
| Configuration | Outcome | Self-crossings, max pairs (frames with / frames) | Jitter [mm] | Wall [min] |
|---|---|---|---|---|
| VBD, benchmark default | held | 20 (160/2222) | 25.1 | 80 |
| ALM-DAT, two detections per substep (D3) | held | 26 (162/2222) | 43.0 | 37 |
| ALM-DAT, one detection per substep (the removed forcing) | lost at 6.47 s | 28 (2128/2222) | 19.0 | 30 |
Built 2026-11-06 from notes/collision-benchmark-alm-dat/video-d3/.