Service Overview

In-Cabin AI Agent

FPT Automotive's cockpit-ai combines predictive safety from CAN-bus analysis and AI-driven voice interaction with a production-ready orchestration layer for heterogeneous hardware. Deployed on Android Automotive and QNX, it runs on-edge VAD, ASR, semantic search, and TTS on resource-constrained SoCs, resolving simple commands in 1.2s and escalating to Azure OpenAI for complex queries. Anticipates high-risk driving 3–15 seconds before escalation without cameras, ensuring privacy.

Voice AIDigital CockpitDriver MonitoringNLP

1,241ms

Edge command path

4,596ms

Cloud path latency

3.06×

Faster than Tier-1 system

3-15s

CAN hazard pre-warning

≥0.85

Decider confidence gate

Capabilities

Key capabilities

Driver Attention Monitoring

Real-time tracking of driver attention levels, fatigue detection, and alertness assessment.

Occupant & Gesture Recognition

Non-camera occupant detection and gesture-based vehicle control — preserving driver privacy.

Natural Language Voice Commands

Context-aware voice assistant that understands driver intent and controls vehicle functions.

Personalized User Profiles

System learns individual driver preferences and auto-adjusts climate, seat position, and entertainment.

Safety-Critical Design

ASIL-compliant system design meeting automotive industry safety standards.

Technology

Technology stack

In-Cabin AI Agent architecture diagram
Component Technology
Monitoring CAN-bus Analysis, Machine Learning
Voice Processing NLP, Speech Recognition
Integration IVI Platform, Cluster, HUD
Personalization Machine Learning
Safety ISO 26262 Alignment

Use cases

Real-world applications

Documented outcomes from actual deployments.

1

Hybrid Voice AI — 7.5s to 1.2s Latency Cut

Hybrid edge-cloud voice pipeline on Telechips Dolphin-5 SoC. Simple commands resolved on-device in 1,241ms; complex conversational queries escalated to Azure OpenAI GPT-4o-mini and returned in under 5s. No raw audio leaves the vehicle.

Before

Cloud-heavy multi-agent routing stacked 7.5s total latency — real-time interaction and the 4s KPI were mathematically impossible

After

Edge path at 1,241ms; cloud path at 4,596ms meeting the 5s target; 3.06× faster than prior Tier-1 optimized system

1,241ms edge path latency
4,596ms cloud path latency
3.06× faster than prior Tier-1
≥0.85 confidence gate threshold

How we work

Implementation approach

1

Phase 1: Requirements & Integration Planning

  • Define cockpit integration requirements
  • Analyze existing IVI/Cluster/HUD systems
  • Plan DMS/OMS sensor placement and calibration
2

Phase 2: System Development

  • Develop driver monitoring algorithms
  • Train occupant detection models
  • Build voice assistant with automotive context
3

Phase 3: Integration & Validation

  • Integrate with vehicle cockpit systems
  • Conduct safety-critical testing (ASIL compliance)
  • Validate across different driver profiles
4

Phase 4: Deployment & Optimization

  • Deploy to vehicle fleet
  • Collect user feedback and performance data
  • Continuously improve based on real-world usage

Explore more

Related services