Reproduce “Ella: Embodied Social Agents with Lifelong Memory” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hongxin Zhang, Zheyuan Zhang, Zeyuan Wang, Zunzhe Zhang, Lixing Fang, Qinhong Zhou, Chuang Gan
Ella, an Embodied Lifelong Learning Agent from UMass Amherst, Johns Hopkins, and Tsinghua, addresses the lack of long-term memory in embodied AI by introducing a dual-form structured memory system integrated with foundation models. The agent demonstrates enhanced social influence and leadership in a 3D open-world simulation, completing four times more goals than baselines in coordination tasks and achieving higher show-up rates in social persuasion challenges.
Key insights
- A structured dual-form long-term memory, inspired by human cognition, is essential for embodied agents to achieve lifelong learning and complex social behaviors in open 3D worlds.
- Integrating such a memory system with foundation models enables sophisticated cognitive functions like robust daily planning, social reasoning, and adaptive behavior by providing rich, contextually grounded information.
- Robust perception capabilities are crucial for embodied social agents to accurately identify environmental features and other agents, leading to more effective social interactions and navigation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.