Digital twins are increasingly used in robotics to simulate interactions, predict outcomes, and support planning. According to a new arXiv preprint, building accurate twins depends on precise estimates of physical properties like mass, which are often uncertain in real-world settings.

The paper introduces a probabilistic physics digital twin for human-robot interaction. By framing the twin probabilistically, it aims to account for uncertainty in physical property estimates rather than relying on single, potentially inaccurate values.

The approach could improve prediction and planning in human-robot interaction, where safety and adaptability are critical. The preprint is available on arXiv under reference 2610.19336.