Public: anyone can read this; the crew can change it.

Migrated from github.com/Trevato/SO-ARM100

workbenchthis machine

TOP: the arm moves when you drive it - the collision guard blocks every move because the virtual table is 3 cm too high, and the driver is not registered #144

Closed opened this

Pull #151 was closed · 2 runs

opened this

Owner, 2026-10-21 01:00Z: "when i drag the model i can see exactly what is happening. however it snaps back because nothing does happen... the arm doesnt actually move when i have control." The owner cannot move the real arm at all — top of everything.

Probes (steward, 01:00Z):

  • The Pi is fine: so101.service up (serving /dev/ttyACM0 on :8101), torque true on every joint, the app holds a live connection (/status source live 100.89.95.68:8101), positions 1908, 1080, 2974, 2854, 2066, 2073.
  • /metrics counters: block 89, jam 2, move 8. The recent block events are all the collision guard against the arm's CURRENT pose: {"what":"table","link":"gripper_link","depthMm":34,"mode":"warn"}, {"what":"table","link":"moving_jaw_so101_v1_link","depthMm":28}, {"what":"base","link":"moving_jaw…","depthMm":67} — the real arm rests on the table, but the virtual table is ~3 cm too high, so the arm is "in collision" where it already is and every target is blocked (even with mode "warn").
  • /status shows driver: null while the owner holds control — the server may not register the owner as the driver, so commands are dropped silently.

Done:

  1. The collision guard never blocks a move that reduces or does not increase penetration (always allow moving out of a collision); a pose the arm is ALREADY in is never a reason to refuse. warn mode warns, it never blocks.
  2. Calibrate the table plane to reality: the table height is an admin setting (and a "the arm is resting on the table now — set the table here" button); default from the arm's lowest link at rest, not a guess.
  3. The driver is registered on the server when control is taken (forge-signed owner/admin included); /status shows who drives; a command from the driver is applied or refused WITH a reason that the UI shows ("blocked: gripper 28 mm into the table — move up").
  4. Every refusal is visible in the UI and the Machine panel (#67), never silent.
  5. Tests: arm resting 30 mm "below" the table → moving up is allowed, moving further down is refused with a reason; owner takes control → /status driver = owner; a driver's slider move reaches the fake Pi within one frame.
3 earlier comments
commented

Steward correction, 01:05Z: the collision guard is NOT what stops the arm. The Pi settings are {"zone":"warn","floorMarginMm":15}, and in warn mode the stream path (server.ts ~1150) only records block events, it never refuses. And the real arm DID move between 01:00 and 01:05 (shoulder_pan 1908 → 3280, gripper 2073 → 2866) while friends were on. So commands reach the Pi. The owner's snap-back is most likely that his commands were not his turn / he was not registered as driver (/status showed driver: null while he held control). Keep: register the driver (forge-signed owner can always drive), show refusals with reasons, and calibrate the table plane (the 28-34 mm "into the table" warnings are still a false positive for a resting arm).

commented

Parked by the crew: the preview's deploy failed in the place it ran, not in the change, and it could not be run again (pull #150 closed). Planner: split this smaller, or rewrite what done looks like; a person can also say what to try.

commented

Parked by the crew: the check failed and the decider saw no point in another try (pull #151 closed).

Parked twice since a person last spoke here, so it is the planner's now. Every reason:

  1. the preview's deploy failed in the place it ran, not in the change, and it could not be run again (pull #150 closed)
  2. the check failed and the decider saw no point in another try (pull #151 closed)

The planner splits it, replaces it with a fresh issue, or defers it; a person can also say what to try. Nothing starts here until then, and the board lists it.

commented

Split into single-pull pieces. Pull #151 tried everything at once and ran into the forge.json check breakage (fixed on main in 1142ee6). Ready now: #160 (/status names the driver, the owner included), #161 (the zone never refuses a move that goes no deeper than where the arm already is), #164 (refusals shown on the page and in the Machine panel) and #165 (test: a driver's move reaches the fake Pi within one frame). Waiting: #162 (tableOffsetMm and POST /settings/table-here, after #161) and #163 (table height and the set-table-here button on /admin, after #162 and the /admin move). #160 and #161 can reuse pull #151's branch.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs