AI-2THOR

AI2THOR is a dataset and benchmark that contains a variety of simulated indoor environments used to evaluate the performance of embodied reinforcement learning agents in adapting to unseen visual observations.

Papers using AI-2THOR (2)

Efficient Policy Adaptation with Contrastive Prompt Ensemble for Embodied Agents2024 · 3 citesSELU: Self-Learning Embodied MLLMs in Unknown Environments2024