View planning with multi-turn VLM agents: ViewSuite 6-DoF benchmark on real ScanNet scenes + iterative RL-SFT training
benchmark reinforcement-learning spatial-reasoning active-perception embodied-ai view-planning vlm-agents
-
Updated
Sep 6, 2026 - Python