The Trenton, Michigan, engine plant — where Stellantis has built V6 powerplants for nearly 75 years — has cut robotic bolt non-engagement from 2.55 percent to 0.24 percent after deploying AI-guided vision from Inbolt. Overall part rejections dropped 90.5 percent, and the project paid for itself in 12 months.

What problem did Stellantis solve?

At the Trenton complex's moving assembly line, a six-axis robot tightens a series of bolts on 3.2-liter and 3.6-liter V6 engines. A lift-and-locate system stops and positions each engine for the robot, but engine placement varied too much for the robot's repeatability tolerance. The robot missed 2.55 percent of bolts.

That forced a manual operator at the end of the line to catch the failures, adding labor cost. The wider issue: traditional robots only go where they are programmed. When parts drift out of position, robots cannot reliably find fastening locations. Lift-and-locate hardware compensates but demands infrastructure changes, increases total cycle time, and loses repeatability as components wear or tolerances stack up.

How did Inbolt's system fix it?

Inbolt, a startup specializing in real-time guidance systems for robots, mounts a standard 3D camera directly on the robot. The camera captures depth data of an assembly station in real time; onboard AI determines each part's exact position and orientation across six axes — X, Y, Z, yaw, pitch, and roll — and updates the robot's path on the fly.

The AI model is pretrained and fine-tuned on CAD data, letting it generalize across a wide range of part geometries without fixed lighting or controlled environments. Stellantis deployed the system at Trenton without changing existing infrastructure.

Since implementation:

• Non-engagement rate: 2.55% → 0.24%
• Part rejections: down 90.5%
• Rework: virtually eliminated
• ROI: 12 months

A human operator remains at the end of the line but now intervenes far less often.

Where else is Stellantis using Inbolt?

Stellantis has expanded the technology to three additional sites:

• Detroit Assembly Complex – Jefferson: An adhesive dispensing station previously experienced misalignments as small as 4 millimeters between the robot's programmed path and the workpiece position, triggering manual cleanup. Inbolt's system now continuously detects workpiece position and adjusts the robot path in real time. Unplanned downtime fell 70 percent; cycle-time consistency improved; ROI hit six months.
• Szentgotthárd, Hungary: A vision-guided robot now picks cylinder heads from pallets and loads them onto the assembly line. Cylinder heads arrived stacked unevenly on flat surfaces, and the autonomous mobile robot delivering pallets stopped in non-repeatable positions. Inbolt's system locates each head, realigns the robot trajectory, and ensures accurate pick-and-place — eliminating the need to upgrade dunnage and cutting total automation cost.
• Valenciennes, France (transmission plant): Engineers automated gear loading and unloading on a machining line where raw parts arrived with non-repeatable positioning and pallets warped from heat treatment. Stellantis uses cobots mounted on mobile carts for flexibility, but cart position varies each time. Inbolt recalculates pallet and pin positions in real time, adjusting cobot motion to avoid collisions and ensure correct insertion.

What to watch next

Stellantis has now deployed Inbolt guidance at four sites spanning the U.S., Hungary, and France — covering engine, transmission, and body assembly operations. Watch for further rollout across Stellantis' powertrain and assembly footprint, particularly applications that previously required lift-and-locate infrastructure. Whether the 12-month and six-month ROI benchmarks hold at higher-volume programs will determine how widely the startup's pretrained-vision approach scales inside the OEM.