Biomechanical User Models for Two-Handed Typing on Physical Keyboards using Reinforcement Learning

Status: open / Type of Theses: Master theses / Location: Leipzig

Physical keyboards remain central to productivity tasks, yet typing behavior varies significantly across individuals depending on hand anatomy, skill level, and learned motor patterns. Understanding how users coordinate both hands during typing can significantly influence typing speed, comfort, and error rates.

The primary goal of this work is to develop biomechanical user models that simulate two-handed typing on physical keyboards within realistic human motor constraints. The project will use the existing BimanualMuscle body to interact with a virtual keyboard environment. Based on computational rationality, the models will assume that users adopt typing strategies that optimize speed and accuracy while minimizing physical effort. Reinforcement learning will be used to train agents to perform typing tasks using coordinated finger movements across both hands. The resulting models can help evaluate keyboard layouts, typing techniques, and assistive technologies, supporting more inclusive and efficient input device design.

One hand typing:
https://arxiv.org/abs/2601.21043

BimanualMuscle body:

Computational Rationality as a Theory of Interaction:

Breathing Life into Biomechnical User Models:

Adapting User Interfaces with Model-based Reinforcement Learning:

Touchscreen Typing As Optimal Supervisory Control:

Simulating Interaction Movements via Model Predictive Control:

Biomechnial