UPenn lab shows GPT-6 Astra can pilot FPV drones from raw images, no mapping stack needed
antoniloq · x · 2026-10-15
UPenn's Janus Intelligent Robots Lab (Antonio Loquercio's team) published a blog testing whether GPT-6 Astra can skip the traditional harness — 3D maps, exploration strategies, motion planners — and directly output low-level control commands (body-frame velocities) for a mobile robot from a single RGB camera.
- Prior demos (driving a real car, tabletop pick-and-place) were confined to small spaces and didn't probe large-scale 3D spatial understanding, which LLMs are known to be weak at.
- The answer is yes: Astra predicts FPV drone control commands directly from images and handled every task thrown at it, in both single-agent and multi-agent settings.
- The authors note practical significance: companies lacking mapping/navigation expertise could replace brittle, task-specific 3D pipelines with a general-purpose model.
More from Embodied
- Sipeed shows $50 stereo depth camera with NPU outputting depth maps and point clouds — AIFlow_ML · 2026-10-15
- IRVL is evaluating frontier AI models for manipulation, report coming soon — YuXiang_IRVL · 2026-10-15
- NVIDIA report: agentic setup lets GPT-6 Astra hit 74.8% on RoboLab-120 robot manipulation — YuXiang_IRVL · 2026-10-15
- Rhoda unveils fingertip-like tactile sensor letting robots identify objects while blindfolded — CSProfKGD · 2026-10-15
- Infinite Machine's Olto e-scooter spotted roaming NYC streets twice in 4 hours — soleio · 2026-10-15
- Vals AI Puts Frontier Models in Real-Time MuJoCo Robot Simulations — scaling01 · 2026-10-15