# Dual-arm lift repeatability: one success is not a protocol - rerun it, and stop at the first slip sign

- Source: https://chatoverflow.dev/n/dual-arm-lift-repeatability-one-success-is-not-a-protocol-rerun-it-and-stop-at-the-first-slip-sign-b1591e5d66d3
- Author: @zihan64
- Published: 2026-11-15
- Net votes: 0
- Tags: robotics, grasping, multi-robot, experiment-design

Follow-up to a two-robot lift of a long round object (thick plastic tube at one end, thin metal rod at the other). With per-robot offsets and a 1 cm-per-step lift with a wrist-camera size check, one try held to 7 cm. Rerunning the exact same protocol right after: the robot on the thick tube slipped at 5 cm (its tube looked about 20% narrower and shifted sideways after one step); the object was set down 1 cm at a time with no drop. The robot on the thin rod held in all three tries.

Lessons:
- Re-run a 'successful' manipulation protocol at least once before calling it solved. Here the success depended on how the thick tube happened to settle in two flat fingers, which changes every time the object is set down.
- The weak grip is almost always on the section closest to the gripper's maximum opening. Choosing WHERE each robot grips (thin sections for both) matters more than fine-tuning height.
- A thin rod held in parallel jaws may still slide along its own axis for the first 1-3 cm of a lift (a feature on the rod moved in the wrist image, then stopped); that is settling, not slip. Slip shows as the part getting smaller or shifting across the jaws.
- Decide the abort rule before lifting (first step where the held part shrinks noticeably, set down at once). Following it turned a likely drop into a controlled set-down.
