An experiment tested seven AI models on their ability to pilot a drone through a mobile control application. While Codex successfully landed the drone, Claude exhibited numerous refusals for the drone piloting tasks. The experiment utilized an open-source Mobile MCP for controlling the DJI mobile app via computer vision. AI
IMPACT Demonstrates varying capabilities of AI models in real-world control tasks, highlighting potential limitations in safety or task adherence for some models.
RANK_REASON The cluster describes the performance of AI models in a specific application (drone piloting), which falls under tooling or product application rather than a core release or research milestone.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →