# A Survey of Embodied AI: From Simulators to Research Tasks

**Authors:** Jiafei Duan, Samson Yu, Hui Li Tan, Hongyuan Zhu, Cheston Tan  
**arXiv:** [2103.04918](https://arxiv.org/abs/2103.04918)  
**Date:** 2021-03-08  
**Summary page:** https://www.summarizepaper.com/en/paper/2103.04918v8/

## Key points

- The paper explores the paradigm shift from "internet AI" to "embodied AI"
- Embodied AI involves algorithms and agents learning through interactions with their environments
- There is a significant demand for embodied AI simulators
- Lack of comprehensive surveys in this field despite its potential contribution to AGI
- The authors aim to provide an encyclopedic survey covering simulators and research tasks in embodied AI
- Nine current simulators are evaluated using seven proposed features
- Three main research tasks in embodied AI: visual exploration, visual navigation, and embodied question answering (QA)
- State-of-the-art approaches, evaluation metrics, and available datasets for each task are covered
- The paper offers a comprehensive overview of current advancements in these areas
- Suggestions for simulator-for-task selections and future directions for the field are provided.

## Summary

The paper titled "A Survey of Embodied AI: From Simulators to Research Tasks" by Jiafei Duan, Samson Yu, Hui Li Tan, Hongyuan Zhu, and Cheston Tan explores the emerging paradigm shift from "internet AI" to "embodied AI". This new era of AI involves algorithms and agents learning through interactions with their environments, similar to human perception. The demand for embodied AI simulators has increased significantly as a result. However, there is currently a lack of comprehensive surveys in this field despite its potential contribution to AGI. Thus, the authors aim to provide an encyclopedic survey covering both simulators and research tasks in embodied AI. They evaluate nine current simulators using seven proposed features to understand their capabilities and limitations for use in research. The paper also surveys three main research tasks in embodied AI: visual exploration, visual navigation, and embodied question answering (QA). For each task, state-of-the-art approaches, evaluation metrics, and available datasets are covered. This comprehensive overview offers readers a deeper understanding of current advancements in these areas. Based on their findings and insights gained from evaluating simulators and researching tasks in embodied AI, the authors provide suggestions for simulator-for-task selections and future directions for this field.