DexVerse™ Embodied Intelligence Engine

Environmental Assets

Robot Assets

Sensing Paradigms

Vision Sensor

Force/Torque Sensor

Tactile Sensor

Online Parallel Data Stream

DexGen

Generates 3D assets from descriptions of physical-aware objects or scenes

DexSim

Generates data

DexAI

Trains models

Differentiable Generative Simulation

Model Assessment

Model Deployment

Fully In-House Designed Embodied Intelligence Engine DexVerse™

Powers an automated generative data-model pipeline scalable for diverse robots, tasks, and scenarios

Cost-Effective Generation of Massive 3D Scenes and Datasets

DexVerse™ efficiently creates realistic, physics-accurate 3D scenes from multimodal inputs, integrating high-precision simulation and rendering to generate large volumes of high-quality synthetic data with detailed lighting, friction, and rigid/soft/fluid dynamics.

Engine-Driven Streaming Data and Online Model Training Paradigm

DexVerse™ uses high-concurrency simulation and flexible automated data generation to stream large-scale training data, enabling continuous model learning. With domain randomization and differentiable simulation, it supports self-adaptive evolution and reduces the sim-to-real gap.

Superior Robustness and Generalization

Rich datasets, advanced training paradigms, and novel architectures enable highly robust, generalizable models achieving up to 99.9% accuracy.

Task Instructions

Robot serves fruit

Robot Organizes Desktop

Robots stack cups

Robot serves dishes