Original Summary

Hi all, wanted to share a cool project me and two buddies have been working on. Essentially we took a Toyota Corolla, installed a Comma 4, hooked it up to GPT-6 Astra&#x2F;Claude Fable&#x2F;Grok 4.6&#x2F;Sol, gave them MCP tools to control the Toyota, and then we let them try driving a cone course (in an empty parking lot at low speed). They can observe the cameras&#x2F;telemetry, request steering&#x2F;speed, and stop the car. The goal was to see how good frontier LLMs might be just out-of-the-box at a task like driving a real physical car (inspired by recent such demos in robotics).<p>The best model was GPT-6 Astra which fully finished the course on its second attempt (which was pretty shocking to us). We gave each model up to 3 attempts on the course in the same chat conversation (to incentivize in-context learning). Fable also improved quite a bit from 9% to 45% by its 3rd attempt.<p>We also have a trace viewer at our website (for example see <a href="https:&#x2F;&#x2F;drivingbench.com&#x2F;trace&#x2F;gpt-6-astra&#x2F;2&#x2F;" rel="nofollow">https:&#x2F;&#x2F;drivingbench.com&#x2F;trace&#x2F;gpt-6-astra&#x2F;2&#x2F;</a> for Astra&#x27;s successful attempt), and everything is open source (traces, harness&#x2F;code at <a href="https:&#x2F;&#x2F;github.com&#x2F;aditya-ramabadran&#x2F;drivingbench_harness_v1" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;aditya-ramabadran&#x2F;drivingbench_harness_v1</a>, we have a report at <a href="https:&#x2F;&#x2F;drivingbench.com&#x2F;report&#x2F;" rel="nofollow">https:&#x2F;&#x2F;drivingbench.com&#x2F;report&#x2F;</a> on how we did everything, some of the things that went wrong or that we could do better, etc).<p>Obvious disclaimer: the models clearly aren&#x27;t good enough to drive on an actual road yet. Also we did this at super low speeds in an open parking lot with a human always ready to brake at any time. Please use the harness&#x2F;code&#x2F;etc at your own risk. Would love to answer any questions or take any feedback for v2!

中文概览

中文标题: Show HN:DrivingBench——前沿大模型驾驶真实丰田卡罗拉

团队在丰田卡罗拉上装 Comma 4,接入多个前沿大模型,并用 MCP 工具让其控制车辆,在空旷停车场低速跑锥桶赛道。最佳模型第二次尝试完成全程,各模型最多三次机会。轨迹、代码与报告均已开源,作者提醒模型还不能上真实道路。


  • 情报分类:开源项目与落地
  • 分类依据:作者展示开源测试平台与代码、轨迹及报告,属落地项目
  • 信息来源:Hacker News 新项目
  • 发布时间:2026/9/22 22:29:43