MM-EQA Benchmark
收藏资源简介:
MM-EQA Benchmark是由加州大学河滨分校可信自主系统实验室构建的多智能体多任务具身问答基准数据集,基于真实场景的Habitat-Matterport 3D数据集开发。该数据集包含丰富的照片级真实家庭环境场景,用于评估智能体在自然语言指令下的协作探索与问答能力。数据集的创建过程涉及对3D场景的语义标注和任务设计,旨在解决多异构机器人协同完成复杂家庭任务时的信息共享与决策优化问题,推动具身智能在真实环境中的实际应用。
The MM-EQA Benchmark is a multi-agent, multi-task embodied question answering benchmark dataset constructed by the Trusted Autonomous Systems Laboratory at the University of California, Riverside. It is developed based on the real-scene-oriented Habitat-Matterport 3D dataset. This dataset includes abundant photorealistic home environment scenes, and is used to evaluate the collaborative exploration and question answering capabilities of agents under natural language instructions. The creation process of the dataset involves semantic annotation of 3D scenes and task design, aiming to solve the problems of information sharing and decision optimization when multiple heterogeneous robots collaborate to complete complex household tasks, so as to promote the practical application of embodied intelligence in real-world environments.

- 1CommCP: Efficient Multi-Agent Coordination via LLM-Based Communication with Conformal Prediction加州大学河滨分校·可信自主系统实验室 · 2026年



