Kavindie Katuwandeniya Madhawa Perera Samith  Ashan Hashini Senaratne

Brandon Matthews Alexander Krumpholz Dana Kulic Leimin Tian

The integration of Large Language Model (LLM)-powered agentic AI architectures, systems capable of autonomous reasoning, tool use, and iterative decision-making, into robotics is driving rapid advances in embodied agentic AI. These systems enable robots to interpret natural language instructions, perform semantic reasoning, generate multistep plans, and execute long-horizon tasks [1,2]. By leveraging the extensive knowledge, reasoning and planning capabilities of LLMs, embodied agentic AI can generate high-level policies that interface with robotic middleware or simulation environments while utilising established low‑level controllers [3, 4].   

Despite this promise, the rapid pace of progress in embodied agentic AI makes it challenging for researchers and practitioners to design safe, interpretable, and reliable systems—particularly when defining tools, separating agent and robot responsibilities, representing world state, and constructing closed-loop reasoning workflows. Effectively deploying these capabilities requires practical understanding of these foundational concepts. This tutorial provides structured, hands-on experience with the core ideas and implementation patterns of embodied agentic AI. 

 
This half-day hands-on tutorial introduces participants to the emerging field of embodied agentic AI by first introducing the fundamentals of AI agents and the practical frameworks used to build them, such as LangChain and LangGraph, before extending these capabilities to embodied contexts through robot simulators. The first session is tailored for newcomers and focuses on core concepts including tool abstractions, agent–environment boundaries, and safe execution patterns. Participants will learn how to construct basic agent workflows, observe agent behaviour, and experiment with controlling simulated robots, establishing a solid foundation that they can later extend to more advanced robot behaviour manipulation. After the coffee break, the next session builds on the foundations introduced in the first session and explores enhanced capabilities such as context-aware planning over world state, agent memory, and looping or iterative decision-making. The final hour features a panel discussion with spotlight talks from researchers and practitioners in embodied agentic AI, offering insights into current challenges, emerging techniques, and real-world applications. 

The tutorial is intended for a broad audience, from beginners with foundational knowledge of AI and robotics to more experienced participants who wish to deepen their understanding. Its progressive structure enables learners to engage at their current level and build incrementally—newcomers establish essential conceptual grounding in Session 1, while more advanced participants refine their approaches and explore contemporary implementation patterns in Session 2. This tutorial is highly relevant to both research and practitioner communities, as it addresses the growing need to connect deterministic robot control with flexible, high-level reasoning enabled by modern agentic AI systems. Through structured, hands-on exercises and expert discussion, the tutorial equips attendees with practical skills, conceptual clarity, and awareness of emerging trends, while fostering dialogue across communities in AI, robotics, and human–robot interaction. 

The Future of Agentic Robotics: Current Challenges, Research Directions and Practical Applications