Workshop: Biomechanical RL as a Methodological Direction for Computational Interaction?

Organisers Florian Fischer (U Cambridge), Arthur Fleig (Leipzig U), Patrick Ebel (HPI), Miroslav Bachinski (Associate Professor, U Bergen), Roderick Murray-Smith (U Glasgow), Antti Oulasvirta (Aalto U), and Per Ola Kristensson (U Cambridge)

Motivation Imagine testing hundreds of interaction designs for an XR application overnight—taking into account criteria that you deem important, such as physical toll, accuracy, and speed. Future biomechanical simulations could offer these possibil ities: acting as “virtual crash test dummies,” reinforcement learning (RL) agents controlling musculoskeletal models in physics-based interaction environments can already predict interaction behaviour and movement strategies to guide interaction design. “Simulated users,” i.e., biomechanical models of the upper or full body with muscles controlled via a pre-trained sensorimotor policy, have achieved remarkable performance across core HCI tasks, including mid-air pointing, keyboard typing, and mobile touch.

However, several challenges limit the practical applicability of biomechanical RL as a research and prototyping method for interaction design. Biomechanical RL depends on computationally expensive training, with typically hand crafted rewards and learning curricula, and does not generalise well across tasks. Moreover, biomechanical models are often perceived technically complex and immature, requiring substantial domain expertise. Overall, biomechanical RL incurs significant computational costs and mental effort.

As a community, we must therefore ask: What role should biomechanical RL play in Computational Interaction? Which of its current limitations are we positioned and willing to address, and where to start? What qualities of biomechanical RL do we deem essential, and how to communicate this to the broader HCI community? This workshop seeks a critical, strategic reflection on biomechanical reinforcement learning as a method in Computational Interaction, and to define a roadmap for biomechanical RL in HCI.

Call for Participation: Imagine testing hundreds of interaction designs for an XR application overnight—taking into account criteria that you deem important, such as physical toll, accuracy, and speed. While future biomechanical simulations trained via reinforce ment learning (RL) could offer these possibilities, acting as “virtual crash test dummies” at early stages of the design process, several challenges, including methodological com plexity, steep entry barriers, and a lack of interpretability and adaptability, currently limit this approach for HCI research and interaction design. This S3CIX 2026 workshop seeks a critical, strategic reflection on biomechanical reinforcement learning as a method in Computational Interaction, and to define a roadmap for biomechanical RL in HCI. It invites all interested HCI researchers to discuss which limitations we as a community are positioned and willing to address, how to tackle them, and where to start best.

To let us know about your interest in this workshop, please fill the form on our website https://biomechanicalrl.github.io, where you are asked to share your previous experiences with biomechanical RL (if any) and share your expectations for this workshop. All participants will be invited to contribute to an Interactions article, where we will share our insights on the potential of biomechanical user simulations for computational interaction and HCI research.