Reimagining service delivery in the agentic era with Google Public Sector
cloud.google.com·21h ago
AI Research
arXiv CS.AI·Jun 12·5 min read
PersonaDrive: Human-Style Retrieval-Augmented VLA Agents for Closed-Loop Driving Simulation
TL;DR
Closed-loop driving simulators often use uniform traffic agents, lacking human-like variability in behavior. PersonaDrive is a pipeline that enhances a vision-language-action (VLA) driving agent by conditioning it on human driving demonstrations categorized by style.
✦ Why It Matters
Engineers can leverage PersonaDrive to create more realistic driving simulations that adapt to various human driving styles.
Key Takeaways
- Utilize human-style data for more realistic driving simulations.
- Implement triplet mining for effective data retrieval.
- Train lightweight models to integrate visual and control features.
- Switch between driving styles without retraining the model.
How It Works
PersonaDrive operates through a three-stage pipeline: it first mines human driving data to create a diverse dataset based on driving styles. Then, it trains a retrieval head that combines visual features with control commands, allowing the model to access relevant driving demonstrations.
Finally, it fine-tunes a VLA backbone to use these demonstrations for predicting waypoints, enabling the model to adapt to different driving styles seamlessly.