Embedded learning foundation
We invent a lightweight training and inference framework running in constrained embedded system computing.
Self-Learning Intelligence
Intelligence that grows through interactions.
A robot’s deployment is the beginning of its learning journey.
MAI owns foundational IP in an embedded training and inference framework. Learning from its users, tasks, and surroundings, a deployed system can develop greater familiarity with its field of use over time.
Explore our foundation →01 / Our foundation
Mahcines learn in the field. Focus intelligence where it matters.
Field specialization
Field learning develops expertise around the conditions repeatedly encounters: its users, recurring tasks, interaction patterns, and operating environment. Generally outperform the broadly trained model.
Compact intelligence
The model is trained to specialize in the relevant field instead of pursuing broad capability across every possible situation, resulting more compact and lightweight model with greater performance and accuracy.
02 / Our vision
Human–robot collaboration
Learn from confirmed interaction
We build embedded self-learning capability so people and robots can become more familiar with one another through use. Confirmed interaction gives the system an opportunity to refine its understanding of individual intent and preferences.
Explore precision gaze interaction →Application example / Gaze-assisted HMI
Learn from the choices a person confirms.
- Observe Gaze and interaction context
- Anticipate Propose the likely interface target
- Confirm The user accepts or corrects the choice
- Refine Use feedback to improve future interaction
A development direction for non-safety-critical UX and personalization.