RoboBench: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models as Embodied Brain
robo-bench.github.io
RoboBench: a comprehensive benchmark evaluating multimodal large language models (MLLMs) as embodied brains across 5 dimensions, 14 capabilities, 25 tasks, and 6092 QA pairs. Accepted to ECCV 2026.
关键词：Embodied AI, Multimodal Large Language Models, Vision-Language-Action Models, Benchmark, Robotics, ECCV 2026