Embodied Intelligence Platform overview
Embodied intelligence is transitioning from laboratories to industrial deployment. During this process, robot companies face three core challenges:
-
Fragmented data pipelines: Data collection, annotation, review, and format conversion rely on different tools. Data is repeatedly transferred between multiple systems, making quality assurance difficult.
-
Sim-to-Real gap: The lack of a unified data pipeline and evaluation standard between simulation environments and real robot deployment leads to high model transfer costs.
-
High engineering barriers: The journey from model training to on-device inference involves multiple technology stacks such as distributed computing, containerization, and device communication, making it difficult to quickly convert research results into products.
The AnalyticDB Embodied Intelligence Platform is built on a cloud-native architecture. It integrates data collection, annotation, model training, simulation evaluation, and real robot deployment into a unified workflow, helping robot companies, data service providers, and research institutions reduce the development and deployment costs of embodied intelligence applications.
Use cases
The platform provides differentiated capability support for the following three types of users:
|
User type |
Typical roles |
Core requirements |
|
Robot manufacturers |
Embedded engineers, AI engineers |
Zero-modification device access, operational data collection, model training, and real robot deployment |
|
Data service providers |
Annotation project managers, annotators |
Annotation task management, automated annotation for improved efficiency, multi-role collaboration, and quality control |
|
Universities and research institutions |
AI graduate students, postdoctoral researchers |
Custom model development, distributed training, simulation evaluation, and experiment reproduction for papers |
Core capabilities
The platform covers the complete lifecycle of embodied intelligence applications:
-
Data collection: Supports multi-device, multi-modal data collection. Robot operational data is synchronized to cloud storage in real time.
-
Data annotation: Integrates Label Studio with 7 built-in annotation templates and 6 auto-annotation operators. Supports image and video annotation.
-
Model training: Includes mainstream VLA models such as GR00T and PI 0.5. Supports one-click training for official models and distributed training for custom models (based on Ray).
-
Simulation evaluation: Integrates NVIDIA Isaac Sim for managing simulation assets, building simulation scenes, and running automated evaluations.
-
Real robot deployment: Trained models can be deployed as cloud-based inference services with one click, or you can use the Python Client SDK for cloud-edge collaborative inference.
-
Cloud-edge collaboration: Coordinates scheduling between cloud computing resources and edge devices. Supports remote collection task dispatch and automatic data upload.
Feature overview
|
Feature module |
Key features |
|
Device management |
Device registration and status monitoring, remote task dispatch (collection/inference), and MQTT communication |
|
User management |
Four roles (Administrator, Collector, Annotator, and Reviewer) and permission assignment |
|
Data management |
Dataset registration (supports sources such as local files, OSS, and development workspaces), data collection, upload, format conversion, and review |
|
Annotation tasks |
Annotation task creation and assignment, 7 built-in Label Studio templates, and 6 auto-annotation operators |
|
Model management |
Model library browsing, model training (official/custom), one-click inference service deployment, and development machine environments |
|
Simulation management |
Simulation asset management (robots/scenes/objects), simulation development machines, and simulation evaluation tasks |
|
Image management |
Unified management of official and custom images |
Platform architecture
The features of the AnalyticDB Embodied Intelligence Platform are distributed across two layers:
AnalyticDB console (adb.console.aliyun.com)
Manages the platform infrastructure, including:
-
Creating and managing Embodied Intelligence Platform instances
-
Configuring computing resources for platform services, inference services, and development training
-
Managing the underlying resource groups for the simulation platform (Isaac Sim & Lab)
-
Viewing metric analytics
Embodied Intelligence Platform (standalone web application)
The primary operational interface, accessed by logging on to the platform service URL. It includes:
-
Device management: Robot registration, status monitoring, and remote task dispatch
-
User management: Role creation and permission assignment
-
Data management: Data collection, upload, processing, review, and annotation
-
Model management: Model training, deployment, and development machine environments
-
Simulation management: Simulation assets, simulation development, and simulation evaluation
Recommended reading path
Based on your role and use case, we recommend reading in the following order:
Robot manufacturers (Device access → Data collection → Model training → Deployment)
Data service providers (Permission configuration → Data management → Annotation)
Universities and research institutions (Model development → Simulation evaluation → End-to-end pipeline)
What to do next
Create and log on to the Embodied Intelligence Platform: Learn how to create a platform instance, complete the prerequisites, and log on to the platform.