World-Model Robotics Platform Board Adviser

The mandate

The board has not decided whether a learned world model should become the common representation for perception, prediction and planning or remain a research component beside modular production controls. Laboratory sequences show stronger long-horizon prediction, but proposals do not reconcile simulation bias, compute demand, intervention, rare warehouse interactions and the cost of replacing inspectable interfaces already deployed.

The adviser will reserve three working days each month for architecture evidence, a chair or committee session, and preparation or individual challenge with research and operations leaders. Five committee meetings are included; urgent questions on a partner or stage-gate decision receive acknowledgement within one business day and a reasoned view within three.

The appointment lasts nine months through the next platform-capital decision. The committee chair may ask the board to renew it once for up to three months if a named pilot or architecture comparison remains incomplete; research uncertainty alone is not grounds for continuing the appointment.

This role carries no line authority, voting right or executive responsibility for architecture, research claims, model release, robot safety, customer deployment, staffing or investment. Accountable technology and operations leaders make those decisions and must preserve their reasoning; the adviser cannot present a challenge or recommendation as product approval.

Concurrent academic and non-competing robotics work is permitted after disclosure. A board role with a direct warehouse-automation rival, material interest in a shortlisted model or simulation provider, sponsored research tied to a preferred architecture, or confidential work for a pilot customer creates a conflict requiring restricted participation, recusal or termination.

Why the board wants this voice

Directors understand industrial robotics and capital allocation but lack someone who has moved learned environment representations from research into bounded physical operation. Management’s architecture debate currently pits novelty against legacy without a common evidence standard. The chair wants an independent builder who can preserve useful modular controls while forcing a platform-scale learning claim to prove transfer and economics.

What you will own

• Press researchers to define which state, prediction and planning decisions the world model changes and which deterministic controls remain authoritative.
• Test evidence across warehouse layout, object mix, occlusion, human behaviour, robot interaction, sensor degradation and operating horizon.
• Challenge simulation and offline sequence gains through controlled physical trials, counterfactuals, calibration and failure-mode comparison.
• Shape stage gates separating representation research, shadow prediction, constrained planning input and any direct influence on robot motion.
• Probe architecture economics for training data, accelerator use, serving latency, fleet update, validation, observability and modular fallback.
• Evaluate build, licence and partnership choices for model ownership, learning rights, interface portability, safety evidence and supplier exit.
• Frame the final board paper with funded hypotheses, protected modular boundaries, rejected claims, residual uncertainty and next-stage proof.

Candidate qualifications

• Led world-model, predictive representation or learned-planning research into physical robot or autonomous-system trials.
• Compared end-to-end learned and modular architectures using transfer, intervention, observability, safety and lifecycle evidence.
• Identified a simulation or sequence-prediction gain that failed to improve a real embodied decision.
• Governed learned models beside deterministic controls and can explain where arbitration and fallback remained necessary.
• Built an investment case including data, compute, validation, fleet and engineering costs rather than research benchmarks alone.
• Advised a robotics board without using supplier, laboratory or customer interests to predetermine the architecture conclusion.

Non-negotiables

• Can attend all five Shanghai committee meetings and both designated laboratory, pilot or ecosystem reviews.
• Will disclose robotics equity, sponsored research, customer work and relationships with model, compute or simulation providers.
• Accepts that engineering and safety executives retain architecture and deployment authority and may reject the advice.
• Has deployed learned representations on physical systems; language-model strategy or conventional warehouse automation alone is insufficient.

1. 49 words maximum. Describe one world-model or learned-planning result that failed to transfer from simulation to a physical system.
2. 49 words maximum. Which current laboratory, vendor, customer or investment relationship could constrain your independence here?
3. 49 words maximum. Confirm the Shanghai cadence and name the evidence required before a world model influences robot motion.