GPT-6 Astra Navigated Cone Course in Robot Trial

The OpenAI model successfully completed the entire driving course while utilizing third-party hardware for vehicle control.

Updated on Sept. 23, 2026 in Robotics

Isometric editorial illustration showing a robotic vehicle frame navigating around orange traffic cones on a grey test grid. — OpenAI's GPT-6 Astra successfully navigated a 130-meter driving course during a controlled robotics trial, demonstrating the ability to translate digital commands into vehicle operation. AI Illustration. Upload story photo >

Live Poll

Do you trust artificial intelligence models to operate physical machinery and vehicles safely?

GPT-6 Astra finished a 130-meter driving course in a parking lot using comma.ai hardware for steering and braking. The model successfully completed 100 percent of the test on its second attempt, navigating while maintaining a speed cap of 3.5 meters per second.

Why it matters

The trial tested the ability of advanced AI models to translate chat-based commands into real-world vehicle operation. While the model performed well in the controlled environment, developers noted that it will not be deployed on public roads in the near future.

The Astra model steered a Toyota Corolla through a 130-meter course in 5 minutes and 22 seconds. Models were required to stay within four meters of the centerline while traveling at a maximum speed of 3.5 meters per second.

The players

OpenAI

The research organization responsible for the development of the GPT-6 Astra and GPT-5.6 Sol artificial intelligence models.

comma.ai

A company that produces open-source automotive software and hardware components used for vehicle automation.

The details

The models were evaluated by sending individual commands through a chat session to control the steering, accelerator, and brake systems. During the trial, multiple other models failed to complete the course, with some refusing to participate entirely due to safety protocols.

September 23, 2026: The article reporting the test results was published.

The Tech Race

The use of comma.ai hardware in this trial marks a transition from human-operated driver assistance to end-to-end command-based AI navigation. This test demonstrates how existing open-source vehicle control hardware can be interfaced with large language models to enable autonomous functionality.

Current drivers should note that these command-based AI systems remain confined to controlled environments and are not yet permitted for public road use. The technology currently requires dedicated hardware installations to bridge the gap between AI processing and vehicle mechanics.

The takeaway

While AI models are showing increased proficiency in navigating complex physical tasks, safety concerns remain a primary barrier to real-world integration. Users should continue to monitor these developments for updates regarding safety standards and potential regulatory shifts.