Miraxis Physical AI Data Onboarding

Transition from digital agents to the physical world by mastering Miraxis-specific data primitives, robotic model architectures, and structured evaluation frameworks for real-world embodiment. * **Map viewpoint taxonomy** to robotics data structures * **Master data primitives** like episodes and trajectories * **Differentiate model architectures** from VLA to JEPA * **Execute targeted annotation** for physical world models

9 sections ยท 51 lessons

Course outline

Foundations

  1. Physical AI
  2. LLM Differences
  3. Agentic Comparison
  4. Embodied Reasoning
  5. Foundations

Data Primitives

  1. Episode Anatomy
  2. Trajectory Dynamics
  3. Sensor Modalities
  4. Temporal Synchronization
  5. Spatial Calibration
  6. Data Primitives

Miraxis Taxonomy

  1. Viewpoint Framework
  2. Egocentric Priors
  3. Exocentric Observation
  4. Allocentric Mapping
  5. End-Effector Contact
  6. Robot-Centric State
  7. Cross-View Alignment
  8. Miraxis Taxonomy

Model Architectures

  1. VLA Models
  2. Diffusion Policies
  3. World Models
  4. JEPA Architectures
  5. Action Chunking
  6. Model Architectures

Training Lifecycle

  1. Pre-training Diversity
  2. Post-training Hardening
  3. Deployment Adaptation
  4. Bridge Demonstrations
  5. Training Lifecycle

Advanced Enrichment

  1. Targeted Annotation
  2. Contact Dynamics
  3. Failure Mining
  4. Reward Modeling
  5. Geometric Refinement
  6. Advanced Enrichment

Simulation Strategy

  1. Synthetic Coverage
  2. Digital Twins
  3. Sim-to-Real Gaps
  4. Scenario Injection
  5. Simulation Strategy

Evaluation Governance

  1. Release Gates
  2. Safety Hardening
  3. Regression Testing
  4. Acceptance Packs
  5. Evaluation Governance

Practical Application

  1. Request Triage
  2. Case Studies
  3. Workflow Design
  4. Customer Solutions
  5. Practical Application

Start learning with Wondering