CHAIC

该数据集名为“受限人机协作”（CHAIC）基准，通过涉及人类限制的任务来评估具身代理在社会感知和协作方面的能力。该数据集涵盖了采用不同规划策略的各种代理，以及用于评估它们在辅助受限人类时表现的一系列性能指标。其规模包括四种代理和八项长期视野的任务，旨在对具身代理在协作和社会感知方面进行基准测试。

This benchmark dataset, named Constrained Human-Agent Collaboration (CHAIC), evaluates the capabilities of embodied agents in social perception and collaboration via tasks involving constraints on human participants. It encompasses a variety of agents adopting different planning strategies, along with a suite of performance metrics used to assess their performance when assisting constrained humans. The benchmark includes four types of agents and eight long-horizon tasks, aiming to benchmark the collaborative and social perception capabilities of embodied agents.