AnyMo: Geometry-Aware Framework for Setup-Agnostic Human Motion Modeling
A new framework named AnyMo has been developed by researchers for modeling human motion using wearable inertial sensors, independent of the setup. This innovative method tackles the issue of inertial signals, which are significantly influenced by various factors such as body location, sensor orientation, mounting position, device hardware, and sampling protocols. AnyMo employs physics-based IMU simulations across multiple body-surface placements to create a range of synthetic signals. It also pre-trains a graph encoder utilizing paired synthetic placement views and masked partial observations, converting multi-position IMU data into comprehensive motion tokens. These tokens are then aligned with a large language model (LLM) to enhance motion-language comprehension, facilitating the transfer of motion representations across different devices and datasets.
Key facts
- AnyMo is a geometry-aware framework for setup-agnostic human motion modeling.
- It uses physics-grounded IMU simulation over dense body-surface placements.
- The framework pre-trains a graph encoder from paired synthetic placement views and masked partial observations.
- It tokenizes multi-position IMU into full-body motion tokens.
- Tokens are aligned with an LLM for motion-language understanding.
- Inertial signals are highly dependent on sensing setup: body location, mounting position, sensor orientation, device hardware, and sampling protocol.
- The goal is to learn motion representations that transfer across devices and datasets.
- The research is published on arXiv with ID 2605.22715.
Entities
Institutions
- arXiv