EviSeek Framework Improves Embodied Question Answering
Original title: EviSeek: An Evidence-Centric Agentic Framework for Embodied Active Information Seeking
EviSeek makes evidence demands explicit to boost accuracy on 3,436 ESI-Bench questions.
EviSeek is a training-free framework that links observation-conditioned actions to question-derived evidence demands. It improves accuracy by 7.1 points over a single-model agent with Qwen3-VL-32B and 3.7 points with Gemini 3.1 Pro. It also improves visual-support-weighted accuracy while reducing mean token usage.
- Improves accuracy 7.1 points with Qwen3-VL-32B
- Improves accuracy 3.7 points with Gemini 3.1 Pro
- Reduces mean token usage
- Makes informational purpose of interaction explicit
From the original article
1 min read at arxiv.orgActive embodied question answering requires agents to acquire evidence through viewpoint changes and physical interaction. The central challenge is to avoid undirected interaction: taking plausible actions without specifying what evidence they should reveal for answering the question. Action plans and accumulated histories alone can leave the informational purpose of an interaction implicit.
By Jintao Li · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.