Painting, converted from an
inspected trade to a verified process.

Every coating manufacturer publishes a specified film thickness; applying under it voids warranties and causes premature failure, applying over it wastes material and causes sagging and cracking. Today this is verified by spot-checking after the fact — if at all.

PaintForge is engineered around in-process film thickness sensing feeding a control loop that modulates pump flow and arm velocity in real time, targeting a ±2 mil tolerance across every pass. The output is not just a painted wall — it is a thickness map of the entire surface, attached to the job record.

Status: core engineering target for 2026 pilot units. Tolerance figures are design targets pending field validation.

The robot builds a depth-vision digital twin of the space — walls, ceilings, openings, obstacles — before a drop of paint moves. Coverage is then solved as a path-optimization problem: minimize overspray and travel time subject to overlap, standoff-distance, and thickness constraints. The operator reviews the plan as a coverage heatmap and approves it in the app.

After execution, the same twin holds the as-applied record: thickness per zone, environmental conditions per coat, photos per wall. Plan and result live in one model.

Every verified square foot compounds. Over time, the platform accumulates a dataset no manual crew can produce: measured thickness, substrate, product, temperature, humidity, and cure conditions — per zone, per job. That dataset is the foundation for asset-lifecycle services: predicting recoat windows, flagging failure risk, and giving building owners a maintenance record instead of a paint invoice.

Design principle: customers own their job data. Aggregate learning improves the platform; individual records belong to the contractor and their client.

The economics that work on real sites — proven by every serious player in this category — are hybrid: the robot handles the high-volume, repetitive, physically punishing spraying; skilled painters own prep, masking, edges, detail work, and final QA. One operator supervises the robot while the rest of the crew does the work that actually requires human judgment.

The trade gets more attractive, not smaller: less time on ladders breathing overspray, more finished square feet per crew, and a new operator skill that commands a premium.

These are open problems we are actively working on — labeled as research, not shipped features:

• Learning-based spray-path optimization under airflow disturbance: adapting pass spacing and gun velocity when HVAC, drafts, or open facades disturb the spray fan.
• Semantic job-site mapping: classifying surfaces (drywall vs. concrete vs. glazing vs. fixtures) from depth + RGB so masking zones are proposed automatically instead of drawn manually.
• Thickness estimation fusion: combining wet-film sensing with flow-rate integration and post-cure spot verification to certify dry-film thickness without full manual gauging.
• Multi-robot job scheduling: sequencing several robots and coats across floors against dry-time constraints, treated as a constrained scheduling problem.

The founding team brings a graduate research background in industrial automation and machine learning (M.Sc., RWTH Aachen University) — the methods above are extensions of published approaches in robotics and control, applied to a trade that has never had them.