Andon Labs Demonstrates GPT-6 Astra Controlling Drone to Find and Track Person Autonomously

Andon Labs has demonstrated that its GPT-6 Astra model can autonomously control a drone to locate and track a specific person in an indoor environment, marking a significant advance in AI-driven autonomous systems. The company tested Astra across five Drone-Bench benchmarks covering navigation, object search, person tracking, and 3D spatial modeling.

In a key trial conducted in an office setting, the model received a natural-language command — "find this person and track them" — and executed the full mission without human intervention. Astra processed visual data from the drone's onboard sensors, built a real-time 3D map of the space, identified the target individual, and maintained continuous tracking as the person moved.

The test showcases the integration of large language models with robotic control loops for complex, open-ended tasks. While Andon Labs has not disclosed the underlying architecture or computational requirements, the demonstration suggests that multimodal foundation models can now bridge high-level reasoning and low-level actuator control in physical domains.

The development raises questions about privacy, surveillance applications, and the governance of AI systems capable of autonomous physical-world interaction.