[Paper Review] Embodied Science: Closing the Discovery Loop with Agentic Embodied AI

Xiang Zhuang, Chenyi Zhou+11 more|arXiv (Cornell University)|Mar 20, 2026

Embodied and Extended Cognition0 citations

TL;DR

This paper argues for embodied science and introduces the PLAD (Perception–Language–Action–Discovery) framework to enable long-horizon autonomous scientific discovery through closed-loop agentic embodied AI.

ABSTRACT

Artificial intelligence has demonstrated remarkable capability in predicting scientific properties, yet scientific discovery remains an inherently physical, long-horizon pursuit governed by experimental cycles. Most current computational approaches are misaligned with this reality, framing discovery as isolated, task-specific predictions rather than continuous interaction with the physical world. Here, we argue for embodied science, a paradigm that reframes scientific discovery as a closed loop tightly coupling agentic reasoning with physical execution. We propose a unified Perception-Language-Action-Discovery (PLAD) framework, wherein embodied agents perceive experimental environments, reason over scientific knowledge, execute physical interventions, and internalize outcomes to drive subsequent exploration. By grounding computational reasoning in robust physical feedback, this approach bridges the gap between digital prediction and empirical validation, offering a roadmap for autonomous discovery systems in the life and chemical sciences.

Motivation & Objective

• Define Embodied Science as a closed-loop, long-horizon discovery paradigm grounded in real experimental interaction.
• Introduce Agentic Embodied AI as a persistent cyber–physical scientific agent with autonomy, embodiment, and long-horizon persistence.
• Argue why current reasoning-centric and execution-centric AI4S approaches are structurally insufficient for sustained discovery.
• Propose the PLAD loop (Perception–Language–Action–Discovery) as an end-to-end framework to couple instrument signals, reasoning, action, and discovery.

Proposed method

• Present a unified PLAD framework linking perception of instrument signals, language-based reasoning with knowledge and tools, embodied laboratory action, and discovery of new scientific insights.
• Detail how perception uses instrument data and experimental state to ground reasoning.
• Describe the role of language with foundation models, knowledge graphs, retrieval-augmented generation, and tools for evidence gathering and hypothesis testing.
• Explain action as embodied execution across spatially constrained or unconstrained laboratory environments, including humanoid and mobile embodiments.
• Define discovery as internalizing execution outcomes to refine hypotheses, models, and future exploration cycles.

Experimental results

Research questions

• RQ1 How can scientific discovery be framed as a closed-loop process that sustainably couples perception, reasoning, and action?
• RQ2 What are the essential components and properties of Agentic Embodied AI needed to achieve long-horizon autonomous discovery?
• RQ3 Why are existing disembodied or execution-centric approaches insufficient for enduring scientific exploration?
• RQ4 How can PLAD be instantiated across diverse experimental settings to enable continuous cycles of hypothesis, design, execution, and revision?

Key findings

• Current AI4S approaches are largely split into reasoning-centric or execution-centric paradigms and fail to sustain long-horizon autonomous discovery.
• Embodied Science reframes discovery as a closed-loop process that integrates perception, language-based reasoning, embodied action, and discovery.
• PLAD provides a modular framework where instruments extend perception, LLMs and knowledge/tools enable grounded reasoning, and embodied action grounds interventions in the lab.
• Long-horizon autonomy requires grounding cognition in instrument-derived signals and linking plan execution to real laboratory feedback.
• Examples show how PLAD can be instantiated in enzyme design and chemical reaction optimization to cycle perception, reasoning, action, and discovery.