Kitakyushu, Fukuoka, Japan| August 24 to 28, 2026| Realizing Human-Robot Symbiosis with AI

Specific room and location TBA.

This workshop aims to focus on the ethical and responsibility considerations for development of robotics foundation models, large-scale machine learning models capable of generalising across a wide range of embodiments and tasks. Building on work and discussions from three other successful workshops (one in 2024, two in 2025) we now continue this workshop series with a specific focus on evaluation and benchmarking of foundation models used for social robotics, in particular for use cases involving vulnerable populations, such as people with disabilities, elderly, and children.

Foundation models have the potential to rapidly expand the capabilities and deployment contexts of robotic systems, unlocking new applications and accelerating adoption. Alongside the development of specialised vision-language-action (VLA) robotics foundation models, standard large language models (LLMs) are increasingly integrated into social robotic systems and studied for their impact on human–robot interaction (HRI) including for the generation of socio-affective behaviours [6], [7] and for deployment in social and care contexts. Key capabilities of these models, such as multimodal understanding, structured reasoning and actions, instruction adherence, natural language and non-verbal cues generation, and action prediction for robotic control are frequently cited as indicators of progress. However, there remains a lack of critical consensus on how such capabilities should be evaluated in the context of physical embodiment use cases, what constitutes meaningful performance in real-world social contexts, and how benchmarking practices should account for the situated, relational, and ethically-sensitive nature of social robotics applications. This is particularly salient for contexts involving vulnerable populations, where evaluation practices directly shape perceptions of safety, appropriateness, and legitimacy.

However, the increasing adoption of robotics foundation models also raises urgent questions about how their impacts, limitations, and risks are assessed. The physical embodiment of these models introduces the possibility of errors that may cause harm, as well as concerns surrounding the collection and commodification of behavioural and psychological data. Their performance across diverse environments similarly demands careful scrutiny, as current evaluation approaches risk-obscuring performance disparities and overestimating generalisability, thereby privileging certain contexts and populations over others. 

In HRI contexts, tendencies toward sycophantic or overly agreeable behaviours may undermine trust, distort user expectations, or compromise meaningful engagement, especially in real-world settings requiring sensitivity, care, or critical judgement. These issues highlight the need not only for ethical reflection, but for robust, context-aware evaluation and benchmarking practices that take seriously questions of harm, fairness, accountability, epistemic diversity, and situated interaction, rather than treating performance as a purely technical metric.

While previous workshops in the series established important groundwork on responsible development, the 2026 workshop will focus on how foundation models for social interaction in social robotics use cases are evaluated and benchmarked, and what ethical and societal assumptions are embedded within these practices. A key objective of this workshop is therefore to make questions of evaluation, benchmarking, and measurement legible and accessible to a broad audience, including technical researchers, while encouraging more critical reflection on what is being measured, for whom, and with what consequences. The key topics the workshop aims to address include:

• Embodiment and Foundation Models

• Safety Evaluations and Benchmarking

• Vulnerable Populations

• Interdisciplinary Methods and Approaches
  
  

Overall, we aim to create a space for learning and open discussion, with the goal of influencing future leaders working in industry, academia, and policy on robotics projects whose awareness and attention for these considerations are critical.