LMM Technologies is building foundation models that treat motion as a first-class sequence domain across rehabilitation, sports, robotics, safety, and other embodied AI applications.
Sequence modeling for motion
Biomechanically grounded
Cross-domain
What is a Large Movement Model?
A Large Movement Model treats motion itself as the primary data type. Instead of learning from words and sentences, it learns from sequences of movement—joint positions, trajectories, timing, and interactions.
Like language models for text, LMMs model how movement unfolds over time—supporting recognition, prediction, and evaluation of motion across real-world environments.
Domains
Rehabilitation
Gait, balance, and motion-quality analytics.
Sports
Biomechanics, fatigue signals, and performance patterns.
Robotics & Safety
Human intent modeling and motion-aware systems.
Pipeline
- Capture → video or sensor streams
- Pose extraction → structured keypoints
- Normalization → clean, aligned motion sequences
- Tokenization → motion as sequence tokens
- Modeling → transformer / diffusion inference
Current Focus
- Motion data pipelines and normalization
- Tokenization strategies
- Foundation model feasibility
- Real-time motion systems