In-vehicle intelligence.
No network dependency.

Battery management, in-cabin safety monitoring, UWB gesture interaction, and touchless HMI — running on automotive ECUs and MCUs at automotive-grade latency. Custom-developed for your platform, your sensor configuration, your OEM requirements.

Kinds of automotive edge AI we build.

Improve in-cabin safety with edge AI vision.

Monitor equipment, sensors, and production systems continuously using anomaly detection models running at the edge. Identify unusual patterns in machine behavior, process data, or visual inspections before they escalate into costly failures. Only alerts and insights are shared — operational data stays local.

Deliver intuitive touchless HMI control.

Use camera-based gesture recognition models to enable contactless interaction with navigation, media, climate, and vehicle controls. Gesture recognition runs locally on embedded compute, providing responsive user experiences without relying on cloud services.

In-vehicle by design.

Automotive edge AI is not edge AI with extra latency budget. It’s a different architectural commitment from the start — about where inference runs, how fast it responds, and where your engineering data lives.

Inference runs in the vehicle

Models execute on automotive ECUs, MCUs, and NPU silicon in the vehicle itself. No connectivity required in the inference path — works the same whether the vehicle has 5G, cellular, or no connection at all.

Automotive-grade latency

Sub-100ms response for safety-critical paths. Meets NCAP active safety latency requirements and CPD regulatory timing. Deterministic, not best-effort.

OEM data sovereignty

Vehicle architecture, sensor configurations, and proprietary training data stay within your engineering perimeter. All engagements are SOW-based with structured sign-off and NDA before technical discussion.

Engagement model: scoping at no charge. NDA before technical discussion. SOW-based delivery. Typical first engagement: 8–20 weeks.

Shipped: NXP Trimension NCJ29D6.

ModelNova developed a 215 KB UWB gesture and motion recognition model for NXP’s Trimension NCJ29D6 — a combined UWB ranging and radar IC for automotive edge sensing.
The model enables touch-free vehicle interactions running entirely on-device at real-time speed: hands-free trunk opening, gesture-based parking assist, and proximity-aware vehicle features. No connectivity dependency. No cloud round-trip. The full inference pipeline fits in 215 KB.

Active production reference. Custom-developed in partnership with NXP.

Go deeper.

Custom Model Development

Our full automotive engagement model — scope, deliverables, timeline, and the engineering depth behind every custom build.

Hardware Compatibility

Which automotive MCUs, NPUs, and ECU platforms we’ve validated, what toolchains we support, and what we can target.

Driver Distraction Blueprint

A worked starting framework that combines vision and sensing models for driver inattention detection. A starting point, not a production system — your custom work begins where this leaves off.

Talk to an automotive AI engineer

Tell us about your program. We’ll come back within two business days with a scoping conversation and an engagement model recommendation.

In-vehicle intelligence. No network dependency.