Tub handling — de-palletize, de-lid, re-palletize, re-lid

How we'd approach it

Lilly has handed us a clean four-phase decomposition, and we should reason from their phases rather than reinvent them. As they state it: remove the door from the collapsible bin, de-palletize ~25 lb tubs onto a conveyor or table, re-palletize them back into the bin, and re-attach the door — with the one hard constraint that no damage is done to the tubs. Our internal Demo 1 read maps these onto two manipulation modes. The lid-off and lid-on phases (their deliverables 1 and 4) are latch work: grasp the door, disengage or engage the latching mechanism, confirm the 'click.' That is snap-fit closure, and the confirmation we need is a force transient, not a position target. The two middle phases (deliverables 2 and 3) are the lift-and-place work: pick a tub whose weight varies with fill, place it without jamming the conveyor, then later feel when it is properly seated back in the bin — a contact-confirmation event, not a sub-mm placement. So the spine of our approach is: lead with the latch and seating events, where tactile actually earns its keep, and treat the lifts as force-adaptive grasps that scale grip to whatever the tub weighs. One thing to settle early with Lilly is what the door/latch actually is — the slide says 'removable doors' on a collapsible bin but does not describe the latching geometry, and Phase A and Phase D both depend on it.

Why it's landable now

The readiness case traces directly to the roadmap, and it is unusually favorable. The two primitives this task exercises are snap_fit_closure and soft_body_handling, and both sit at works_in_demos with a roughly three-month horizon — not early_experiments, not aspirational. Snap-fit is in fact our chosen first demo, so it is the primitive we are furthest along on; latch/lid engagement is precisely the thing it does. Soft-body handling is core to Demo 2 and demonstrable today, if not yet pilot-proven. Critically, nothing here reaches into the Phase 3 primitives — rigid_insertion and press_fitting — that are still early_experiments at six-to-twelve-month horizons. This task does not ask for them. Layer on the task's own forgiveness: the tolerance is contact-confirmation rather than sub-mm, and Lilly's stated tolerance for a failed attempt is high. That combination — gating primitives already in demos, no Phase 3 dependency, and a high failure tolerance — is what 'build_now' means in practice. The one roadmap caveat to keep honest is continuous_operation, which is explicitly not_yet and not on the near-term roadmap. That matters because Lilly cites ~200 tubs/hour. For a proof-of-principle we can almost certainly show the four-phase cycle end-to-end without sustained-throughput hardening, but we should ask Lilly directly whether the PoC is judged on a working cycle or on holding rate over a shift — the answer changes how much the continuous_operation gap bites.

The weight question we have to resolve

There is a number that does not reconcile across the anchors, and it is load-bearing — literally. Lilly states ~25 lb lifts, which is roughly 11 kg. Our own internal Demo 1 spec, however, describes tubs at '≤5 kg loaded' in its hardware envelope while its variations table separately says 'loaded ≈5–10 kg.' So our internal analysis is not even self-consistent, and neither figure matches the customer's stated 25 lb. This is not a detail to paper over: whether the OpenArm/Wuji bimanual platform comfortably handles ~11 kg per tub — and whether a single-tub lift or a two-handed lift is assumed — determines whether the payload envelope claim in our spec actually holds. The right move is to ask Lilly for the actual loaded-tub weight and tub dimensions per line, rather than carry forward a 5 kg assumption that may be off by more than 2x. Relatedly, their slide frames the solution as a mobile manipulator chosen specifically because low ceilings and human-scale spacing rule out fixed palletizers, whereas our Demo 1 spec is written around a bimanual workcell — so we should confirm whether the PoC itself requires the mobile base or whether a fixed-base demonstration of the four-phase cycle is acceptable as proof first.

Where it sits against the baseline

In the Adjacent Industry Analysis, pharma/lab automation sits at rank 7 — Medium forgiveness, Medium tactile value, ±0.5–1mm tolerance, a 12–18 month sales cycle, and a Tier 2 (Phase 3) GTM position anchored on autoinjector assembly. This task is materially more favorable than the vertical's general profile. It does not live at the sub-mm, autoinjector end of pharma; it lives in the Phase 2 band — deformable handling and pharma kitting — where our capabilities are already commercially relevant. Its forgiveness is better than the vertical's Medium, because the failure tolerance is high and the constraint is contact-confirmation, not precision insertion. In tolerance-spectrum terms it behaves much more like the Tier 1 'quick win' tasks (fulfillment, CPG, produce) than like the demanding pharma device-assembly tasks the baseline assumes. That is the strategic attraction: it is a pharma logo and a pharma line, but the actual manipulation is a forgiving, Phase-2-grade job we can stand behind now. The asterisk is the baseline's hard line — tub damage is a facility-level incident in pharma, and our spec carries a 0% damage target. The honest tension to carry into the next conversation is that Lilly's high failure tolerance covers a failed attempt (a re-grasp, a missed cycle) but explicitly not tub damage, so we need to understand from them what counts as damage and how robust a tub is to a dropped or mis-seated lift — that boundary, more than throughput or tolerance, is where credibility is won or lost.

Open questions

• What is the actual loaded-tub weight and per-line tub dimensions? Lilly states ~25 lb (~11 kg), but our internal Demo 1 spec assumes ≤5 kg in one place and 5–10 kg in another — none of which agree, and it determines whether the bimanual payload envelope holds.
• Does the proof-of-principle require the mobile manipulator base Lilly describes (chosen for low ceilings), or is a fixed-base demonstration of the four-phase cycle acceptable as initial proof?
• Is the PoC judged on completing the four-phase cycle end-to-end, or on sustaining ~200 tubs/hour over a shift? Continuous_operation is explicitly not on our near-term roadmap, so sustained throughput would expose a real gap.
• What is the actual door/latch geometry on the collapsible bin? Both the lid-removal and lid-reattach phases depend on it, and the customer slide names 'removable doors' without describing the latching mechanism.
• What precisely counts as 'tub damage' for Lilly, and how robust is a tub to a mis-seated or dropped lift? Their high failure tolerance covers a failed attempt but explicitly not damage — and pharma treats damage as a facility-level incident.