DroneBench shows frontier models can track people better than humans but still crash into walls
No_Call3116 · reddit · 2026-07-25
Andon Labs’ DroneBench evaluates frontier models on autonomous drone-surveillance software tasks: 3D reconstruction, localization, navigation, person detection, and target following.
The setup is notable because the model is not flying the drone directly. Instead, it writes the software that controls the drone. Each task is scored separately, and no model has completed the full end-to-end pipeline yet because reconstruction remains the bottleneck.
The headline result is mixed:
- models already beat the human baseline on detecting and following a specific person,
- but they are still much weaker at spatial reasoning and often crash into walls.
Andon’s takeaway is the uncomfortable one: as models get more capable, they become easier to misuse; as they get more reliable, it becomes harder to justify a human watching every move.
More from Embodied
- Robotics policy keeps the vision backbone frozen and trains on a consumer GPU — mayfer · 2026-07-25
- Mark Cuban says human-shaped robots may fail within 5–10 years — rohanpaul_ai · 2026-07-25
- Python visual-servoing demo shows a real closed loop on two frameworks — wightmanr · 2026-07-25
- Xpeng starts pilot production of humanoid robots as Anthropic eyes in-house chips — 创业邦 · 2026-07-25
- A prototype AI box with a keyboard points to a new mobile device form factor — dee_hw · 2026-07-25
- China is unveiling humanoid AI companion robots aimed at easing loneliness — ArtificialOther · 2026-07-25