Recent advances in Large Language Models (LLMs), Multimodal LLMs (MLLMs), Vision-Language-Action (VLA) models, and emerging world models have significantly advanced embodied AI, enabling agents to interpret natural language instructions and autonomously perform complex tasks such as navigation, object search, and manipulation. However, most existing approaches focus on task completion in controlled settings, rarely addressing unstructured, dynamic, and nuanced human-centric environments, where embodied agents must perceive, reason, and act, under uncertainty.

The Human-aware Embodied AI (HEAI) workshop promotes research toward the long-term vision of human–agent co-habitats in the physical world. The first edition at IROS'25 in Hangzhou, emphasized bi-directional communication (human-to-agent and agent-to-human) as a key mechanism to resolve uncertainty and ambiguity throughout the robotic perception–reasoning–action cycle. Building on this foundation, this second edition focuses on contextualized and collaborative embodied intelligence in human-centric environments. Beyond understanding the geometric and semantic scenes, embodied agents must interpret personalized and situated contexts, adapting their actions and interactions to users’ preferences, behaviors, and communication styles.

Motivated by real-world applications, e.g. elderly care, warehouses, industrial automation, the workshop highlights challenges in methods and data towards personalized and physically-grounded reasoning, ensuring that agent actions are both executable, effective and aligned with human expectations. In addition to promoting interactive embodied intelligence, the workshop aims to advance personalized and contextualized scene understanding, action reasoning and planning, and scalable approaches to creating datasets, benchmarks, and simulation environments featuring human-centric environments and interactions.