HoME (Household Multimodal Environment)

HoME（家庭多模态环境）是一个多模态环境，供人工代理学习视觉、音频、语义、物理以及与对象和其他代理的交互，所有这些都在现实环境中进行。 HoME 基于 SUNCG 数据集集成了超过 45,000 种不同的 3D 房屋布局，该规模可以促进学习、泛化和迁移。 HoME 是一个开源的、与 OpenAI Gym 兼容的平台，可扩展到强化学习、语言基础、基于声音的导航、机器人技术、多智能体学习等方面的任务。

HoME (Home Multimodal Environment) is a multimodal environment designed for AI agents to learn visual, audio, semantic, and physical knowledge, as well as interact with objects and other agents, all within realistic physical environments. HoME integrates over 45,000 distinct 3D house layouts based on the SUNCG dataset, and this scale facilitates learning, generalization and transfer. As an open-source, OpenAI Gym-compatible platform, it supports a wide range of tasks including reinforcement learning, language grounding, audio-based navigation, robotics, multi-agent learning and more.

搜集汇总

背景与挑战

背景概述

HoME是一个多模态家庭环境数据集，包含超过45,000种3D房屋布局，支持视觉、音频等多种数据类型。它是一个开源的、兼容OpenAI Gym的平台，适用于强化学习、语言基础等多种AI任务。