AI2-THOR

AI2-THOR is a benchmark dataset that contains interactive 3D environments used to evaluate the performance of embodied agents in tasks involving navigation and interaction based on natural language instructions.

Papers using AI2-THOR (4)

ESearch-R1: Learning Cost-Aware MLLM Agents for Interactive Embodied Search via Reinforcement Learning2025Integrating Vision Foundation Models with Reinforcement Learning for Enhanced Object Interaction2025Planning with affordances: Integrating learned affordance models and symbolic planning2025Efficient Policy Adaptation with Contrastive Prompt Ensemble for Embodied Agents2024 · 3 cites