Modality Framework
This section is informative and explains the Modality Framework in a human-friendly way.
What Is a Modality?
A Modality is a structured type of data — like:
- a skeletal pose
- a contact state
- an EEG window
- a depth frame
- a robot joint command
Think of it like a “class” of data with:
- a name
- a family/domain
- a version
- some metadata requirements
- a schema defining its fields
Example:
MIND.pose/SkeletalPose@1.0.0
Why namespaces?
Namespaces prevent collisions and allow ecosystems to coexist:
MIND.pose/SkeletalPose
MIND_EXT.pose/HandPose
ACME.cv/HandKeypoints
LAB_X.bio/EEGHighDensity
This is exactly how:
- OpenXR uses
XR_EXT_* - glTF uses
KHR_* - ROS uses
sensor_msgs/*,geometry_msgs/* - USD uses
UsdGeom*,UsdSkel*
What are Modality Families?
Families help organize the ecosystem.
Examples:
pose/→ joint rotations, full-body trackingbio/→ EEG, EMGcv/→ 2D/3D landmarks, depth imagesagent/→ commands to a robot or virtual body
This keeps things clean and scalable.
ASCII diagram:
MIND namespace
├── pose/
│ ├── SkeletalPose
│ ├── HandPose
│ └── FullBodyPose
├── bio/
│ ├── EEG
│ └── EMG
├── cv/
│ ├── Pose2D
│ ├── DepthFrame
│ └── ObjectDetections
└── agent/
├── JointCommand
└── GazeCommand
Versioning Explained
Every modality evolves:
- New fields
- Adjusted constraints
- Better precision
Semantic versioning makes this predictable:
1.x.x→ stable and compatible2.0.0→ breaking changes
Required Metadata: Why Important?
A modality cannot stand alone.
Examples:
- A SkeletalPose without a SkeletonProfile is meaningless
- EEG without a sensor layout is unusable
- A robot joint command without a RobotModelProfile can’t be applied
So modality specs define:
- which metadata profiles are required
- how they link together
Unknown Modalities — Safe and Future-Proof
If a new lab invents:
LAB_X.bio/GalvanicSkinResponse
Tools should:
- keep the data
- ignore it unless needed
- not crash
This makes MIND future-proof.
Registry Overview
The Modality Registry is the authoritative list of:
- core modalities
- extensions
- vendor additions
It tracks:
- names
- versions
- stability
- family
- required metadata
It acts like the “package index” for the MIND ecosystem.
Summary
Modality design in MIND follows the same patterns used in:
- OpenXR
- glTF
- USD
- ROS
- BIDS
It is:
- scalable
- extensible
- future-proof
- versioned
- namespaced
- metadata-aware
This makes MIND suitable for:
- mocap
- XR
- biosensing
- CV pipelines
- robotics
- embodied AI