A control room for the shoe factory

Perception at every station, a planner that owns the whole lot, agents that hold cutting, moulding, lasting and bonding against real drift, and an audit trail a brand auditor can read.

Book a line assessmentExplore the agents
  • Factory edge + cloud
  • Graduated autonomy
  • Immutable audit log

The factory orchestrator plans the lot, the agents run the stations

A lot enters as a goal — 1,200 pairs of Velo Trainer 2 at target yield, fit and bond strength. The orchestrator decomposes it into station-level work and hands each step to the agent that owns it.

RUN-4417 · Velo Trainer 2 · Line B · plant HCMC-03agent graph
Twin simulate — Yield-and-Energy — SUCCEEDED Twin simulate twin.simulate_run ✓ SUCCEEDED Cut and nest — Cut-and-Nest — SUCCEEDED Cut and nest cut.nest_optimize ✓ SUCCEEDED Stitch upper — Stitch-and-Upper — SUCCEEDED Stitch upper stitch.tension_control ✓ SUCCEEDED Mould sole — Mold-and-Sole — SUCCEEDED Mould sole mold.density_control ✓ SUCCEEDED Last upper — Last-and-Bond — SUCCEEDED Last upper last.fit_control ✓ SUCCEEDED Bond sole — Last-and-Bond — APPROVAL Bond sole bond.window_control ! APPROVAL Inspect pair — Quality-and-Fit — SUCCEEDED Inspect pair quality.inspect ✓ SUCCEEDED Finish and pack — Robot-and-Assembly — SUCCEEDED Finish and pack robot.finish_pack ✓ SUCCEEDED
View as table
Text equivalent — nodes, owning agent, dependency and status
#NodeAgentDepends onStatus
01 Twin simulate Yield-and-Energy — SUCCEEDED
02 Cut and nest Cut-and-Nest Twin simulate SUCCEEDED
03 Stitch upper Stitch-and-Upper Cut and nest SUCCEEDED
04 Mould sole Mold-and-Sole Stitch upper SUCCEEDED
05 Last upper Last-and-Bond Mould sole SUCCEEDED
06 Bond sole Last-and-Bond Last upper APPROVAL
07 Inspect pair Quality-and-Fit Bond sole SUCCEEDED
08 Finish and pack Robot-and-Assembly Inspect pair SUCCEEDED

One lot, eight steps, one human decision

The loop above is not a diagram — it is RUN-4417 as the orchestrator actually executed it on Line B. Every step carries its own duration, evidence and status.

  1. 01 twin.simulate_run SUCCEEDED 2m 40s

    Simulated the whole lot in the shoe-and-line twin before a single hide was cut — nest layout, mould recipe and bonding process window for EVA midsole and rubber outsole at 71% relative humidity.

    inputs · outputs

    in: pattern velo-trainer-2, hide batch LH-2291, mould set M-14 · out: nest v7, recipe R-08, bond window 62–68 °C

  2. 02 cut.nest_optimize SUCCEEDED 4m 12s

    Scanned 64 hides, marked 1,842 scars and brand marks, then solved a defect-aware nest across the EU 36–46 size curve. Leather yield came in at 87.4% against a 79.1% plant baseline.

    inputs · outputs

    in: 64 hides, 1,842 defects · out: yield 87.4%, 12.3 s solve, 22 nested part groups

  3. 03 stitch.tension_control SUCCEEDED 11m 08s

    Held stitch tension inside ±4% across 14 stitching stations, caught three skipped stitches on the lateral overlay and re-ran those uppers before they reached lasting.

    inputs · outputs

    in: seam map v3, thread lot TL-77 · out: 3 skips corrected, 0 escapes

  4. 04 mold.density_control SUCCEEDED 9m 51s

    Trimmed press temperature and dwell on P-3 as the EVA lot ran slightly high in blowing agent, holding midsole density at 0.24 g/cm³ within ±1.8% and cure at 214 seconds.

    inputs · outputs

    in: EVA lot EV-1180, recipe R-08 · out: density 0.24 g/cm³, cure 214 s

  5. 05 last.fit_control SUCCEEDED 6m 33s

    Lasted the upper over last set L-42 under a heat and tension profile tuned to this leather lot. Toe-lasting deviation held at 0.6 mm; heel seat measured true on every pair sampled.

    inputs · outputs

    in: last set L-42, 78 °C, profile T-2 · out: deviation 0.6 mm

  6. 06 bond.window_control APPROVAL 3m 04s

    Humidity climbed to 71% mid-shift, pushing the primer flash-off outside the validated window. The agent proposed +2 °C activation and +0.3 bar press pressure, and paused for the quality engineer to approve — the change touches a validated bonding parameter.

    inputs · outputs

    in: primer PR-9, adhesive AD-22, RH 71% · out: proposed activation 66 °C, awaiting sign-off

  7. 07 quality.inspect SUCCEEDED 5m 47s

    Bond-line vision, 3D fit metrology and cosmetic scan on every pair. First-pass yield 98.2%; 21 pairs routed to rework for adhesive squeeze-out, none for bond gaps.

    inputs · outputs

    in: 1,200 pairs · out: FPY 98.2%, 21 rework, 0 bond-gap escapes

  8. 08 robot.finish_pack SUCCEEDED 18m 22s

    Robotic cleaning, lacing, insole insertion and carton packing to spec C-4, with per-pair traceability written back to the MES and the brand’s quality portal.

    inputs · outputs

    in: carton spec C-4 · out: 1,179 pairs packed, 21 held

Where the platform sits

Steps 01–08 run on the factory-edge runtime. The orchestrator holds the lot goal, the twin supplies the plan, the agents own the stations and the audit log records every write-back.

Decision latency
50–100 ms at the station
Plan solve
12.3 s for the full lot
Writes to controls
guardrailed + schema-validated
Approval gates
1 (bond window)

Autonomy is graduated: shadow, then advise, then act inside a bounded window.

The platform speaks to machines, not to dashboards

Every platform capability lands as a tool call against a real cutter, stitcher, press, oven or vision rig — and every call and result is logged.

RUN-4417 · platform tool-call streamline-b/agent-runtime
  1. 09:14:02 hide.scan(batch="LH-2291", hides=64, sensor="line-b/cam-01")
  2. 09:14:19 64 hides mapped · 1,842 defects marked · mean usable area 91.2%
  3. 09:14:20 nest.optimize(pattern="velo-trainer-2", size_curve="EU36-46", objective="yield")
  4. 09:14:33 nest v7 · leather yield 87.4% (+8.3 pts vs baseline) · solve 12.3 s
  5. 09:18:44 cutter.execute(table="CT-2", nest="v7", blade_profile="leather-1.8mm")
  6. 09:22:56 2,412 parts cut · 0 out-of-tolerance · scrap 12.6% (plant baseline 20.9%)
  7. 09:41:10 stitch.monitor(stations=14, thread_lot="TL-77")
  8. 09:52:18 3 skipped stitches detected on lateral overlay · 3 uppers re-run · 0 escapes
  9. 10:03:02 press.set_recipe(press="P-3", recipe="R-08", dwell_s=214)
  10. 10:12:53 midsole density 0.24 g/cm³ ±1.8% · 0 short shots
  11. 10:31:07 bond.check_window(primer="PR-9", adhesive="AD-22", rh=0.71)
  12. 10:31:09 RH 71% pushes primer flash-off outside validated window
  13. 10:31:10 approval.request(param="activation_temp", from=64, to=66, reason="humidity drift")
  14. 10:34:14 approved by t.nguyen@[PLACEHOLDER] · change logged to audit trail A-88213
Every line is real, copyable text and is announced politely to screen readers.

Guardrails on every write

Setpoint writes are bounded by the validated process window for the construction and material lot. Anything outside it becomes an approval request, not a write.

  • RUNNING
  • APPROVAL
  • SUCCEEDED
  • FAILED

How guardrails are enforced →

Perceive, plan, act, verify — pair by pair

The same closed loop runs at every station, from the cutting table to the packing cell.

  1. Perceive

    Cameras, thermal, pressure, humidity, machine current and PLC tags are fused per station into a live picture of the hide, the upper, the sole and the shoe.

    30–60 FPS inspection

  2. Plan

    The twin and the orchestrator produce the nest, the mould recipe, the lasting profile and the bond window for this lot and these material lots.

    solve < 15 s

  3. Act

    Agents write setpoints through your controls with guardrails, schema validation and approval gates on validated parameters.

    50–100 ms decisions

  4. Verify

    Vision and metrology check every pair, results feed yield and energy optimisation, and the run is logged, cited and retained for the liability window.

    100% pairs

What ships into your plant

Factory-edge inference where latency and IP demand it; cloud training and fleet management where scale demands it.

Factory-edge runtime

A sandboxed inference runtime on Jetson-class and edge-server hardware, running vision, time-series and process-reasoning models per line with signed artefacts and one-command version rollback.

  • SUCCEEDED
  • RUNNING
  • WAITING

Model router

Every step is served by the best or cheapest model that meets its bar — fine-tuned open models for high-volume vision, frontier models for root-cause reasoning.

Shoe-and-line twin

Simulates cut, mould, last and bond before the run and auto-optimises the nest and process window.

Knowledge and memory

Retrieval over patterns, BOMs, construction specs and adhesive datasheets, plus per-factory quality history and per-technician craft memory — versioned and tenant-scoped, never shared across customers.

Connectors and write-back

OPC UA, MQTT, Modbus TCP and vendor SDKs into cutting tables, stitchers, presses, lasting machines, ovens and conveyors; write-back into MES, ERP, PLM and historian so the plant keeps one source of truth.

What the platform does that generic MES cannot

MES tells you a lot ran. Footeon decides how it runs, pair by pair, against the materials actually on the table.

Defect-aware nesting

Solves the nest against this hide batch and its scars — not an idealised rectangle — across the live size curve.

Bond-window control

Holds primer, adhesive, activation and pressure inside the validated window as humidity and lot chemistry drift.

Fit and dimensional metrology

3D measurement of lasting fit, sole geometry and finished dimensions on every pair, not a sampled batch.

Changeover sequencing

Solves line balance and changeover order across a high-mix schedule of models and colourways.

Energy per pair

Optimises press, oven and conveyor energy against cycle-time and quality constraints.

Graduated autonomy

Observe, recommend, act-with-approval or act — configured per agent, per line and per parameter.

Stack at a glance

  • edge · Jetson Orin today / Jetson Thor [ASPIRATIONAL]
  • vision · DeepStream + TensorRT · 30–60 FPS
  • fusion · Holoscan [ASPIRATIONAL]
  • serving · Triton + NIM, per-tenant routing
  • optimise · cuOpt nesting, line balancing, scheduling
  • twin · Omniverse shoe-and-line simulation
  • synthetic · Cosmos + Replicator defect generation
  • training · DGX/HGX model families [ASPIRATIONAL]

Items marked [ASPIRATIONAL] are planned capability, not shipped today.

Built on a Physical AI stack, not a dashboard

Footwear autonomy is a real-time perception and control problem: high-speed leather vision, synchronised sensor streams and sub-second station decisions.

Edge vision runs on Jetson-class units with DeepStream and TensorRT; synchronised camera, thermal, pressure, humidity and PLC streams are fused where bond and moulding quality depend on timing; nesting, line balancing and changeover scheduling are solved with GPU optimisation; the twin runs on Omniverse-based simulation; and rare defect and delamination scenarios are synthesised rather than waited for.

  • 4–12 cameras per line across cutting, stitching, moulding, lasting and packing
  • Multi-model serving with per-tenant routing and version rollback
  • Synthetic generation for hide scars, stitch skips, adhesive gaps and warped soles
  • Optional on-prem deployment where pattern and last IP cannot leave the plant

You decide how much the agent is allowed to do

Every agent and every parameter carries an autonomy level. Most plants start in observe and walk up one workflow at a time.

Autonomy levels and what each one permits
LevelAgent behaviourHuman roleTypical use
ObserveSenses and reports; writes nothing.Runs the line as today.Weeks 1–2 baseline
RecommendProposes setpoints and nests for review.Accepts or rejects each proposal.Shadow-mode tuning
Act with approvalExecutes, but pauses on validated parameters.Approves bonding, lasting, mould changes.Steady state for bond and last
ActExecutes inside guardrails and reports.Reviews exceptions and audit log.Nesting, inspection, packing

Connects to the machines and systems you already run

Footeon is an autonomy layer, not a machine purchase. Connectors ship as versioned adapters.

Cutting and nesting

  • Atom cutting tables
  • Comelz CZ series
  • Zünd G3
  • Lectra Versalis
  • Generic CNC / G-code

Stitching and lasting

  • Programmable stitchers
  • Lasting machine PLCs
  • Toe / side / heel lasters
  • Conveyor MES signals
  • Barcode and RFID pair tracking

Moulding and bonding

  • Compression press controllers
  • PU / EVA injection lines
  • Primer and cement stations
  • Activation ovens
  • Press pressure and cure sensors

Plant systems

  • SAP and Oracle ERP
  • PLM (Centric, PTC)
  • MES / MOM systems
  • OPC UA, MQTT, Modbus TCP
  • Historian (PI, Ignition)

Connectors ship as versioned adapters. Anything not listed is reachable through OPC UA, MQTT, Modbus TCP or the Footeon REST/gRPC API.

Latency and reliability the line can trust

Station decisions have to land inside the machine cycle, or they are just analytics.

  • −18% energy per pair across cutting, moulding and lasting
  • 0.14% defect escape rate to the brand’s QC gate
  • 24 agent tool families shipped across the platform
  • 99.95% factory-edge runtime availability target

Latency and availability targets are engineering commitments for supported hardware profiles; contractual SLAs are set per deployment.

Isolation, approvals and an immutable trail

Pattern, last and construction IP is the most valuable thing in the building. It stays yours.

Certifications and posture

  • In progress SOC 2 Type II — audit window open [ASPIRATIONAL]
  • In progress ISO 27001 — controls implemented, certification pending [ASPIRATIONAL]
  • Supported GDPR — EU data residency and DPA available
  • Supported Per-tenant isolation for patterns, lasts and construction IP
  • Supported On-premises deployment with zero telemetry egress

Agent-specific assurances

  • Graduated autonomy: observe → recommend → act-with-approval → act, set per agent and per parameter
  • Human-in-the-loop approval gates on every validated bonding, lasting and moulding parameter
  • Immutable, quality-grade audit log of every agent action, approval and setpoint write
  • Guardrails and schema validation on tool calls; fail-closed to last known-good recipe
  • Sandboxed factory-edge runtime with signed model artefacts and version rollback
  • Scoped permissions per line, per station and per operator role

Read the security overview →

Platform questions

  • No. Footeon is a software and factory-edge autonomy layer that sits on top of the machines you already run. We connect over OPC UA, MQTT, Modbus TCP and vendor SDKs, read sensors and vision, and write setpoints back through the controls you already trust — with approval gates on any validated parameter.

See the platform on your line

A two-week baseline on one line tells you exactly where your yield, bond and changeover money is.

Book a line assessmentSee pricing