NovaX
Full-stack embodied intelligence

From understanding the world
to acting within it.

NovaX is a company focused on full-stack embodied intelligence. The company brings together embodied AI models, integrated robotic systems and learning from real-world execution, developing robots that can perceive their surroundings, coordinate actions and carry out multi-step tasks in physical spaces.

Watch NovaX in 60 seconds →Explore the full stack
NovaX in action

One connected system. From models to motion.

60 seconds · English · 1080p

The NovaX approach

Build the model. Integrate the robot. Learn from execution.

Physical intelligence depends on how the entire system works together. Our development spans multimodal understanding, action generation, robotic integration and the feedback that connects execution to learning.

01 / EMBODIED AI MODELS

World–Action Models

Our WAM direction connects perception, spatial understanding and action. We are developing a multimodal foundation for reasoning about physical tasks and translating understanding into robotic behavior.

02 / ROBOTIC SYSTEMS

Intelligence meets hardware

We connect model development with robotic execution, including manipulation and dexterous-hand demonstrations. Whole-robot and first-person views show how the system interacts with objects and its surroundings.

03 / COMPLETE TASKS

Beyond a single action

Kitchen and manipulation scenarios provide a setting for developing coordinated, multi-step behavior. Our focus includes task continuity, interactions across a workspace and feedback during execution.

04 / RSI SELF-EVOLUTION

Turn execution experience into the next capability

RSI self-evolution is a key research direction at NovaX. We explore how experience and feedback from robot tasks can feed training and evaluation, guide policy improvement and validate new capabilities. The focus is the complete path from task execution to the next iteration, connecting model development, robotic practice and continuous learning.

RSI self-evolution · An integrated learning loop

Execute. Learn from feedback. Validate again.

We explore continuous iteration grounded in real tasks: gather experience from robotic execution, use feedback analysis, training and evaluation to improve policies, then return to physical environments for validation. This loop connects progress across the full stack to practical tasks.

Perceive & understand→Plan & act→Execute on the robot→Evaluate & learn

The film presents real demonstrations, product-design imagery and explanatory animations. Design concepts and learning mechanisms are shown alongside observed robot actions.