Data-Driven Approach to Reproduce Individual Motor Signatures for Avatars and Robots

A recent paper on arXiv (2503.15225) introduces a novel, fully data-driven technique for creating unique one-dimensional motions that reflect the distinct motor signatures of individuals, characterized by specific velocity patterns. This research, presented as a replace-cross update, highlights the necessity for authentic human motion models in autonomous virtual avatars and robots engaged in collaborative activities such as rehabilitation therapy, sports, and manufacturing. The authors initially illustrate how motion amplitude serves as an effective means to define individual motor signatures, enhancing existing descriptors. They then unveil a data-driven method to generate motions that embody these signatures. This advancement is crucial for the evolution of cognitive architectures and control strategies for agents interacting with humans, as prior models provided only simplified representations of motor behavior. The paper can be accessed on arXiv with the identifier 2503.15225.