Vehicle-assembly autonomy for body, paint, trim, final and end of line

Every station, every minute, inside spec.

Chassira watches the line the way a launch engineer would if they could stand at every station at once. Its agents read weld traces, framing measurement, booth conditions, torque curves and test results per body, then correct what drifts inside limits you set, before the body moves on.

Takt remaining at this station

About 60 seconds per vehicle.

A car plant has about sixty seconds to get each station right.

Then the body moves, and whatever drifted moves with it: a weld out of its window, a framing datum creeping toward the tolerance edge, a booth too humid for a dark colour block, a fastener off its torque curve. Downstream it becomes rework, line stops that cost tens of thousands of dollars a minute, warranty and recall exposure.

~60 s
the takt a car plant launches a vehicle to
~30,000
parts per vehicle, built to millimetre geometry and newton-metre torque
~90 M
vehicles shipped by automakers each year

Industry figures from our problem statement, not Chassira results.

ST10

Body-in-white

Weld, join and frame the body to crash-safety geometry.

  • Stamp-and-Form
  • Weld-and-Join
  • Frame-and-Geometry
Side view of a body-in-white with weld spots along the sills and pillars, and a callout on the left B-pillar datum at framing station FS-1 corrected by 0.3 millimetres. FS-1 left B-pillar datum offset corrected 0.3 mm W-14 spot welds, trace per joint
Illustrative scenario, Plant BIW-2. The framing correction stayed inside the qualified envelope; gap-and-flush vision confirmed it on the next 25 bodies.

What drifts

Weld current and force wander, expulsion climbs on one robot, a new coil lot springs back differently at the press, and the framing datum walks toward the tolerance edge. Sampled audit finds it hours later.

What Chassira does

Scores joint quality per weld from the servo-gun trace and seam vision instead of per destructive sample. Corrects framing offsets inside the envelope. Proposes the weld schedule change, then waits for a named body-shop engineer, because weld schedules are crash-safety critical.

ST20

Paint

Pretreat, e-coat, film build, flash and cure in a booth that never holds still.

  • Paint-and-Finish
  • Quality-and-Anomaly

Booth B-2, dark colour block

Illustrative scenario. The defect model predicted dirt on horizontal panels; the agent trimmed humidity 2 points and rebalanced airflow before the block arrived.

Chart: booth humidity runs high and is trimmed down 2 points before the dark colour block; film build stays inside its window throughout; predicted dirt risk on horizontal panels is high before the trim and low after it. humidity trimmed 2 pt booth humidity film build, inside its window dirt risk high dirt risk low shift time, toward the dark colour block

What drifts

Booth temperature, humidity and airflow, e-coat chemistry and oven profiles move through the shift. The result shows up later as dirt, runs, orange peel and film-build faults, then as repaint.

What Chassira does

Predicts the defect map a body will carry from booth and chemistry state, moves booth setpoints inside the qualified window ahead of it, and localises every defect that still appears to zone, robot and condition.

ST30

Trim

General-assembly torque, fluid fill and glazing, vehicle after vehicle.

  • Trim-and-Fasten
  • Sequence-and-Flow

What Trim-and-Fasten reads on every vehicle

  • Torque toolstrace per fastener, scored against the specification curveread only
  • Fluid fillvolume and pressure per circuitsetpoints in window
  • Glazingbead and placement visionroute or hold
  • Sequencebuild order, JIS feeding, AGV and AMR state, buffersmoves inside agreed limits

What drifts

A tool loses calibration, a fill head under-delivers, a bead wanders on the glass. Each one is small at the station and expensive at end of line or in the field.

What Chassira does

Predicts mis-torque, under-fill and glazing faults per vehicle and routes or holds the vehicle before the fault travels. Torque specifications stay with the process engineer. Sequence-and-Flow prices every sequence and buffer move in takt minutes and rework.

ST40

Final

Marriage of body, chassis and pack: the joints nobody can see once they are made.

  • Powertrain-and-Battery
  • Trim-and-Fasten
Diagram: a vehicle body lowered onto a chassis and battery pack at the marriage station, with the pack's fasteners marked and one fastener flagged. fastener 7 pack
Cell-to-pack joining, coolant and high-voltage integrity and torque-to-yield traces are scored per unit at the marriage station.

What drifts

Torque-to-yield traces, cell-to-pack joints, coolant and HV integrity. On an EV line standing up faster than engineers can staff it, one bad joint is a recall question, not a rework ticket.

What Chassira does

Flags or holds units before they leave the station, with the trace attached. Torque specifications and battery-joint parameters are never changed by an agent: it proposes, a named engineer decides.

ST50

End‑of‑line test

Prove the vehicle, rather than discover it.

  • EOL-and-Calibrate
  • Quality-and-Conformance

Test stand, one vehicle

  1. Wheel alignmentPass
  2. ADAS sensor calibrationRetest, inside takt
  3. Headlamp aimPass
  4. LeakPass
  5. Brake and rollPass
  6. Functional testPass

Illustrative. In the BIW-2 scenario two ADAS retests were sequenced inside takt and the shift was released with full as-built records.

What drifts

Calibration and test failures that were set up far upstream, in the body, the paint or the fasteners, and only surface when the stand runs.

What Chassira does

Predicts failures from the upstream build record, sequences retests and rework inside takt, and writes the as-built record per vehicle against IATF 16949 and APQP criteria. It never passes a vehicle a test has failed; verdicts and limits are read only.

The gate before the customer

Gap and flush, and torque: two findings from one illustrative shift on Plant BIW-2, and the argument the plant had to settle between them.

Gap and flush, left B-pillar

Gauge: the gap-and-flush reading had trended toward the tolerance edge and returned to near nominal after the framing correction. nominal tolerance edge beforeafter
  1. TrendIn-line measurement shows the left B-pillar datum creeping toward the tolerance edge.
  2. MoveFrame-and-Geometry asks to hold FS-1 for one cycle and apply a 0.3 mm datum correction, inside its qualified envelope.
  3. ProofGap-and-flush vision confirms the correction on the next 25 bodies.

Torque to yield, pack PK-1182

Torque against angle: the specification band rises to a plateau; the trace for fastener 7 leaves the band early and plateaus low. angletorque spec band fastener 7
  1. TraceFastener 7 on pack PK-1182 falls outside the specification curve at marriage.
  2. HoldPowertrain-and-Battery holds the unit and routes it to rework with the trace attached.
  3. LimitThe torque specification is untouched. No agent can write it.

At the same moment, Sequence-and-Flow wanted to pull six bodies forward through FS-1 to recover three minutes lost to an andon stop.

The orchestrator scored both requests against the run goal, geometry inside targets. A body out of geometry outranks three minutes of takt. Framing kept the station for one cycle; the minutes came back through the paint-shop buffer instead.

Illustrative scenario, not a customer result.

Ramp it like a new model launch: one process first

A plant does not hand a line to software on day one, and it should not. A Chassira pilot is a paid deployment on one process or area at the Process tier, with the success metric written into the order form: weld or geometry rework, downtime minutes, paint defect or repaint rate, warranty or PPM, or launch quality. The baseline comes from your own records, not from our telemetry.

Onboarding is sized at 150 to 600 GPU-hours per plant and 50,000 to 250,000 synthetic variants per vehicle program. Both are design targets.

Step chart of a pilot ramp: shadow, then assist, then bounded autonomy on low-risk loops, then supervised weld-schedule adaptation for qualified joint families. Each step is gated. L1 shadowL2 assistL3 boundedL4 supervised pilot startgated by ROI and safety
  1. L1 Shadow. Reads everything, predicts and recommends, writes nothing. The baseline is measured here.
  2. L2 Assist. Proposes moves with evidence and confidence; your engineer approves each one.
  3. L3 Bounded. Writes on low-risk loops inside the envelope, once ROI and safety gates are met.
  4. L4 Supervised. Weld-schedule adaptation, only for qualified joint families, with a named approver.

Start on one process. Run the whole plant when it pays.

Billing period

Process

$18,000

per production process or area, per month, billed monthly

Weld, paint or assembly control, or defect-inspection AI, for one process. For plants proving the number on one wedge.

  • One process or area, one module
  • Plant-edge inference node included
  • Controller and sensor reads, guarded writes
  • Run log, audit trail, as-built record
Start a pilot

Plant

$140,000

per plant, per month, billed monthly

All ten modules from stamping to end of line, the twin, conformance and traceability, robotic cell and inspection coordination.

  • Every control module on every process
  • Quality-and-Anomaly on every line
  • Vehicle-and-line twin and sequencing
  • 24/7 support, quarterly model review
Talk to us

Enterprise

Custom

land $1M to $15M ACV

Multi-plant OEM and platform fleets, with custom weld, paint, geometry and battery models kept private to each plant and program.

  • Fleet-wide autonomy policy
  • Plant edge fleet, on-prem option
  • SLAs, SSO and RBAC
  • Dedicated deployment engineer
Enterprise details

Annual prepay carries a 15 to 20% discount. Outcome-based components are available on rework, downtime, warranty and first-time-through. Full pricing

Asked on the line walk

What plant managers, launch engineers and quality leads ask first. More in the full FAQ.

Walk us down your line.

Pick the station that costs you most: body-shop weld and gap-and-flush, paint defects and film build, or torque and battery-joint conformance. We will scope a paid pilot against a baseline agreed in writing.

Pre-launch. No customers, pilots or logos are claimed. We are looking for two to four design-partner plants.