Embodied Intelligence Platform overview

更新时间:
复制 MD 格式

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)

  1. Create and log on to the Embodied Intelligence Platform

  2. Manage devices and robot access

  3. Collect and manage data

  4. Train and deploy built-in models

  5. Best practice: end-to-end from model development to simulation evaluation

Data service providers (Permission configuration → Data management → Annotation)

  1. Create and log on to the Embodied Intelligence Platform

  2. Manage users and permissions

  3. Collect and manage data

  4. Annotate data (Human-in-the-Loop)

Universities and research institutions (Model development → Simulation evaluation → End-to-end pipeline)

  1. Create and log on to the Embodied Intelligence Platform

  2. Train and deploy built-in models

  3. Develop custom models

  4. Manage simulation environment and evaluation

  5. Best practice: end-to-end from model development to simulation evaluation

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.