Partnerships

Technology

Tianzhi — Embodied AI Development and Runtime Platform

A unified base for data, models, simulation, evaluation, deployment and operations in public-safety settings — carrying robots from understanding a scene to acting in it.

The closed loop

  1. Field data capture
  2. scene construction
  3. multimodal understanding
  4. task planning
  5. skill execution
  6. on-robot feedback
  7. simulation iteration

Platform modules

  1. Data and scene foundation

    • Field duty data capture
    • Multimodal data intake and annotation
    • Scene, task and incident libraries

  2. Multimodal understanding

    • VLM scene understanding
    • Recognition of people, vehicles, objects and events
    • Spatial relationships and 3D scene modelling

  3. Task planning and decision-making

    • Instruction understanding and task decomposition
    • Temporal reasoning and route planning
    • Multi-robot coordination and scheduling

  4. Execution control

    • VLA and controller action execution
    • Chassis, PTZ and manipulator coordination
    • Human takeover and safety fallback

  5. Training, evaluation and simulation

    • Model training, fine-tuning and policy optimisation
    • Simulation training, scenario replay and evaluation
    • Sim2Real transfer to physical robots

  6. Deployment and operations

    • Edge–cloud inference deployment
    • Model versioning and staged rollout
    • Live monitoring, alerting and data return

A closed data–training–planning–execution–feedback loop that keeps improving how public-safety robots understand, decide and act.

Related