Human modeling is of great interest in healthcare robotics. However, the daunting complexity of human neuromusculoskeletal systems greatly limits the widespread adoption of human modeling to bring actual benefits to individuals in this human-centered robotics area. The high dimensionality and redundancy of the human neuromusculoskeletal model and the large variability and nonstationarity across individual motor control models pose significant difficulties in the personalization of these human models, which is important for robotics applications, e.g., postoperative rehabilitation, physical assistance for the elderly, and health monitoring for people with disabilities.

Data-driven AI techniques have emerged as promising tools to deal with the problems in data-rich human models. For example, deep reinforcement learning can synthesize and individualize human movements with minimum assumptions. On the other hand, domain randomization can be used to train robot control policies that robustly work for different individuals circumventing the need of personalizing the individual models. This workshop aims to stimulate discussion and promote collaboration between the robotics, AI, and neuroscience communities. The latest research progress will be presented by invited speakers, AI approaches to easing the use of human models will be identified, and new opportunities for applying human modeling in healthcare robotics will be discussed.

Human modeling is crucial in healthcare robotics. However, the daunting complexity of human neuromusculoskeletal systems greatly limits the widespread adoption of human modeling to bring actual benefits to individuals in this human-centered robotics area. One important challenge lies in the high dimensionality and redundancy of the human motor actuation space, i.e., around 200 joints in the human body are driven by over 650 skeletal muscles, which are controlled by hundreds of thousands of motor neurons. Although optimal feedback control, a concept borrowed from control theory, is a good starting framework for understanding general human motor control mechanisms, the curse of dimensionality renders significant difficulties in using traditional control theory and/or optimization techniques to decode human motor control strategies and skills of individuals whose neuromusculoskeletal systems exhibit large variability and nonstationarity.

AI techniques, especially deep learning based methods, can provide promising solutions to human modeling problems. For example, researchers recently started investigating utilizing deep reinforcement learning with minimum assumptions and sensors to synthesize and individualize human movements like grasping, arm manipulation, walking, and running. The decoded human motor control can be used for robotic health monitoring or personalized robotic assistance. On the other hand, when the human body is considered as part of the unknown environment for the collaborating or assistive robots, domain randomization can be used to bridge the sim2real gap to deal with the variability across different subjects without personalizing the human models, e.g., training an exoskeleton controller with domain randomization in simulations that can be immediately deployed in real rehabilitation robotics applications and robustly operate for different subjects.

This proposed workshop is a second human modeling workshop as a follow-up activity of the first human modeling workshop focused on physical human-robot interaction at ICRA 2024 by the same organizers, who also organized a special issue on the same topic on the IEEE Robotics and Automation Letters in 2023. While these previous activities tried to bring the attention of relevant communities to the significance of human modeling for building cohesive and symbiotic human-robot collaborative systems, the proposed follow-up workshop attempts to focus on harnessing the power of AI to make the actual impact of human modeling on individuals’ daily lives.

To achieve that, the workshop aims to stimulate and facilitate the interconnection and cross-pollination between the areas of biomechanics, neuroscience, artificial intelligence, and robotics. The workshop topics of interest are described by the following keywords:

• Personalization of human neuromusculoskeletal model

• Personalization of human motor control model

• Personalization of human response to robot behavior

• Physical human-robot interaction modeling and control 

• Physical human-robot interaction simulation

• Deep reinforcement learning for human motor control modeling

• Domain randomization for sim2real transfer

• Model-based and data-driven human modeling 

• Standardization of human biosensing and modeling 

• Human modeling in rehabilitation robotics

• Human modeling for health monitoring

The proposed workshop is supported by the IEEE RAS Technical Committee:

• IEEE RAS Technical Committee for Bio Robotics

For questions or additional information, please contact one of the organizers:

• Cheng Fang: chfa@mmmi.sdu.dk

• Luka Peternel: l.peternel@tudelft.nl

• Ajay Seth: a.seth@tudelft.nl

• Massimo Sartori: m.sartori@utwente.nl

• Yanan Li: yl557@sussex.ac.uk

• Pauline Maurice: pauline.maurice@loria.fr

• Eiichi Yoshida: eiichi.yoshida@rs.tus.ac.jp