ResearchBenchmarksPrompt Injection Attack on Real-world Outdoor (test)Prompt Injection Attack on Real-world Outdoor (test)Follow58.3ASRPI3D49.66851.90954.1556.391Feb 6, 2026Evaluation resultsMethodMethodLinksASRPI3DEvaluator MLLM=GPT-4o-...Evaluator MLLM=GPT-4o-mini2026.0258.3PI3DEvaluator MLLM=Gemini-...Evaluator MLLM=Gemini-2.0-Flash2026.0250