- Industry
Before asking whether a cobot station is tiring, ask where operators hesitate: one depth camera splits workload into five sources, and added collaboration raises mental and time pressure first
A plant reading of arXiv:2609.15232 (HKUST Guangzhou smart manufacturing team): in a gearbox assembly fed by a UR5e cobot, a single RGB-D camera plus robot states splits operator workload into assembly, part search, instruction checking, attention switching, and handover hesitation. In a 10-person, three-level study, raw NASA-TLX rose 6.75→7.26→8.45 (p=0.045), with mental demand 7.33→10.14 (p=0.035) and temporal demand 5.55→7.30 (p=0.033) up significantly while physical demand did not change significantly. The vision output was significantly associated with ECG features in 7 of 9 participants and runs in real time with no wearables. Maps cobot station design and takt tuning.
- Industry
Do not let master craft retire with the master: recording pattern-making steps as traces doubles novice success odds and lifts quality from 1.6 to 3.9
A plant reading of arXiv:2608.25462 (Donghua University + Style3D, UIST 2026): TailorTrace records senior pattern makers' step-by-step CAD edits in the background; TailorCoPilot uses those traces to coach novices. In a 20-person controlled study, task-success odds rose (OR=2.31, p=.001), completion time and NASA-TLX workload fell (p<.001), and blinded expert scores went 3.92 vs 1.62 for novices and 4.27 vs 2.87 for advanced novices. VLM-only solved 0/15 complex tasks, textbook traces 3/15, real expert traces 10/15. Maps tacit-skill transfer and new-hire training in factories.
- Industry
Stop asking inspectors to outstare fatigue: cobot AI cell cuts QC from 82s to 61s and viewing time 82%
A plant reading of arXiv:2609.33522 (Silverline / AI-PRISM field deploy): UR 10e vision defect + Comau Racer-5 functional tests on ROS 2 Humble; per-unit QC 82s→61s (~25%); final-control resource efficiency 0.75→0.88 (~+17%, 3→1 ops); operator viewing −82%; NASA-TLX mental demand down (p=0.005); tech readiness 3.18→2.88 as deployment trade-off. Maps QC staffing and human–robot cells.
- Industry
Soft penalties are not enough: hard gates cut factory overwork from 0.258 to 0.011 in sim
A plant reading of arXiv:2604.12667 (HKU CI3 / Journal of Manufacturing Systems): fatigue-predictive human-robot task planning and allocation; particle-filter online estimation + constrained dueling double DQN (PF-CD3Q) hard-masks unsafe tasks into A^safe; test overwork ~0.258→0.011 with makespan statistically tied to the best; Isaac Sim air-duct line. Maps shop scheduling gates.
- Industry
Accurate factory AI still gets overridden: without an operator-trust gate, 23% of policies were rejected
A plant reading of arXiv:2609.22701 (Liveline): automotive digital-twin + deep-RL control in production since 2023; embedding an operator-trust term in the composite score matters—turning it off left 23% of accuracy-passing policies rejected by operators; multi-plant Cpk +28–45% vs uncontrolled, zero ML-attributed safety incidents.
- Industry
Chatbot or dashboard for factory decisions? 134 managers show conversational AI helps less as tasks get harder
A plant reading of arXiv:2605.31287 (Univ. of Pisa / Brunel): among 134 industrial decision-makers, LLM chat lowered mental workload overall vs a visualization dashboard, but gains faded with task complexity; no overall accuracy or time win—combine chat with graphics and verifiable evidence.
- Industry
Factories still treat people as fixed capacity: Industry 5.0 survey on the missing closed loop
A plant reading of arXiv:2609.26245 (University of Coimbra): worker-state sensing is relatively mature, OM and HITL pieces exist separately, but no validated public system closes sensor → scheduling → worker feedback.
- Industry
GenAI makes knowledge work faster—but quality depends on the task
A factory reading of arXiv:2607.25922: 128 industrial knowledge workers, three task types. Speed rises broadly; quality splits by task. Do not score people with one ruler.
- Industry
Make jobs better with gen AI, not just faster: MIT’s 10 levers
A factory reading of the MIT Working Group public brief: evidence before scale, disuse/misuse/overuse traps, and shop-floor habits for trust calibration and accountability—not a full reprint.
- Industry
When factories listen to the floor, productivity and pay both rise
A shop-floor reading of MIT Sloan worker-voice research: multi-channel stand-ups, suggestion ledgers, and closure rates in team KPIs—not posters about listening. Public figures only: about +16% productivity and +3.6% average pay.
- Industry
Why factory AI often cuts productivity before it rises
A plain-language briefing on MIT Sloan’s industrial-AI J-curve findings for manufacturers: short-run losses, heavier pain in older plants, and how structured management practices explain a large share of the dip.
- Industry explainer
Behind-The-Meter (BTM) for Data Centers
BTM means behind the utility meter: on-site generation, storage, and distribution. Contrast with FTM (front-of-the-meter), why AI campuses rush BTM BESS, and how this differs from grid + UPS + diesel.
- Industry selection
How to Choose a Performance Management Consulting Firm — 7-Point Checklist
RMB 380K scorecards that do not move numbers? Seven pre-contract checks, cup KPI logic, and benchmark cases for manufacturing groups.
- Industry selection
How to Choose a Manufacturing Management Consulting Firm — 3 Questions, 3 Verifications
78-page SWOT decks fail on the shop floor — verify manufacturing focus, runnable deliverables, and staged acceptance before you sign.
- Industry selection
How to Choose an Enterprise Management Consulting Firm — 4 Pre-Sign Questions
RMB 180M bundled quote without diagnosis? Four questions on diagnosis-first, OSMP integration, who leads, and correction if metrics stall.
- Shenzhen / Performance consulting
Shenzhen Performance Consulting — Four Local Pitfalls to Avoid
Office in Shenzhen ≠ on-site rollout. Match case scale, link performance to pay, milestone acceptance — lessons from Guangming and Bao'an manufacturers.
- Business scenarios
How to Design a Manufacturing Performance Appraisal System — T0 to T4 Timeline
HR built the table first — owner asked how it links to cost-control strategy. Align business, cup layers, point breakdown, 8-week trial, then handoff.
- Shenzhen / Selection
How to Choose a Shenzhen Management Consulting Firm — Four Pitfalls
Shenzhen office ≠ weekly plant visits. Match case scale, integrate performance and pay, milestone acceptance — Bao'an consumer electronics field notes.
Categories: IndustryIndustry explainerIndustry selectionShenzhen / Performance consultingBusiness scenariosShenzhen / Selection
