Skip to main content

Event Architecture

This section explains the event system in simple language.

What Are Events?​

Events are “things that happened.”
They attach meaning to time in your dataset.

Examples:

  • “The user pressed button A”
  • “A grasp started”
  • “A robot arm reached waypoint 3”
  • “EEG channel spiked”
  • “The agent completed a task phase”

Events turn raw pose/bio/CV data into semantic structure.

Event Hierarchy Explained​

Events come in three main forms:

EventPoint     → a moment in time
IntervalEvent  → a span of time
TypedEvent     → a domain-specific event

Think of it like inheritance in object-oriented programming.

ASCII:

Event
 ├── EventPoint
 ├── IntervalEvent
 └── TypedEvent
       ├── GraspEvent
       ├── AttentionShiftEvent
       ├── TrackingLostEvent
       └── ObjectPlacedEvent

Why Multiple Timestamps?​

Different sensors use different timebases.

Examples:

  • EEG uses sample indices
  • Video uses frame numbers
  • XR uses monotonic clocks
  • Mocap uses system clocks

Events may include ANY of these.

This is what allows cross-device alignment.

Event Relationships (actor, target)​

Typed events can describe who did what to what.

Example:
A GraspEvent:

  • actor = left hand
  • target = object_42

A CognitiveEvent:

  • actor = “EEG montage”
  • context = “working memory trial 7”

An AgentEvent:

  • actor = robot arm
  • target = null

Compound Events​

Complex tasks break down into smaller events.

Example:

PickAndPlace
   Reach
   Grasp
   Transport
   Place

This structure is essential for:

  • imitation learning
  • hierarchical RL
  • multimodal analysis
  • robotics task decomposition

Referencing Streams and Samples​

Events may link to:

  • specific pose samples,
  • specific EEG windows,
  • specific frames,
  • specific objects or metadata elements.

This allows learning systems to reconstruct the exact context of an event.

Summary​

Events are:

  • semantic,
  • hierarchical,
  • multimodal,
  • timestamp-aware,
  • cross-device friendly,
  • essential for training embodied AI.