Reproduce “ComBodied Agents: a New Paradigm of Human-Centric Agentic AI” 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
Qianggang Ding, Xingyao Wang, Rui Feng, Zhibin Wang, Feixiang Yao, Kelong Mao, Hao Sun, Zhiyao Luo and 14 more
The paper proposes Combodied Agents, which judge success by effects on a person’s evolving state and agency rather than immediate task completion. Its framework combines multimodal perception, longitudinal memory, a Personal World Model, and a separate policy for selecting permitted support under consent and safety constraints. This is a research program, not a demonstrated system, and its performance remains unestablished.
Key insights
- A structural gap exists in current Agentic AI paradigms, which predominantly focus on external task completion, neglecting the evolving state and agency of the human as the primary object of interaction.
- A closed-loop architecture is essential for human-centric agents, requiring event-based multimodal perception, longitudinal and correctable memory, and personal world models for predicting human states under interventions.
- Preserving human agency, privacy, and control is paramount, necessitating edge-native deployment architectures and new evaluation metrics beyond mere task completion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.