Show HN: DrivingBench – Frontier LLMs Driving a Real Toyota Corolla
Category: ai-ml
Tags: llm, autonomous-driving, mcp, openpilot, benchmark
Score: 6.8/10 (Innovation: 8, Technical: 8, Documentation: 8, Utility: 3)
DrivingBench is a research harness that lets frontier LLMs control a real Toyota Corolla at parking-lot speeds via MCP tools, running on a comma device with openpilot. It records every action as replayable traces, providing a controlled benchmark for language models in physical-world tasks. The project is technically ambitious but narrowly applicable due to hardware and safety constraints.
Target audience: AI researchers, autonomous driving engineers
Repository: https://drivingbench.com/ · Python · NOASSERTION · 60 stars
View on Hacker News