Autonomous Driving Solutions

Scenario Analysis

01.

Model development

Improve resource utilization and reduce TCO.
02.

Data acquisition

Efficient and compliant data collection requires an efficient, reliable, and multi-protocol storage environment.
03.

ERWE

Efficient and compliant data collection requires an efficient, reliable, and multi-protocol storage environment.
Efficient and compliant data collection requires an efficient, reliable, and multi-protocol storage environment.
Efficient and compliant data collection requires an efficient, reliable, and multi-protocol storage environment.
04.

Model training

The efficient construction and deployment of fusion models requires a large number of resources for large-scale training of fusion models.
05.

data processing

An efficient data management, cleaning, and annotation platform requires a seamless connection between the data processing platform and the model development and training platform.

Solution

The autonomous driving solution jointly created by Inspur Information and ecological partners provides the core source power for autonomous driving model training through leading computing infrastructure, data management, data labeling and model training solutions. Efficiently build a full-stack cluster solution from "data collection, storage, and processing" to "model training" and "simulation verification" to solve the problems of insufficient computing power, insufficient storage resources, and difficult software for autonomous driving customers.

The solution mainly includes AI server, CPU server, AI pod networking, and AIStation artificial intelligence development platform.

1. The AI server adopts a leading technical architecture to provide strong computing power for the algorithm development of autonomous driving single-task model and fusion model.

2. The CPU server provides stable and reliable computing power support for data management, data annotation and big data management;

3. The AIStation artificial intelligence development platform manages AI applications in a unified manner, and provides users with algorithm platforms and application optimization services;

4. AI Pod networking is designed for convergence models, providing effective network communication guarantee for strong AI computing power.

Application Scenarios

The solution provides basic computing power and software platform support for autonomous driving model development, and provides full-stack computing power and software platform support from data collection, data annotation, data management, model training and simulation verification.

The Value of The Solution

Perfect meta-brain ecology

MetaBrain Ecosystem unites left-hand partners with algorithm development capabilities and right-hand partners with rich industry landing experience to jointly provide platform support for data management and data processing for autonomous driving users.
Powerful resource management platform

It provides a reliable, easy-to-use, and flexible service deployment and computing resource management platform for users of autonomous driving algorithm development, helping users quickly launch their services and improving the utilization efficiency of computing resources.
AI pod networking

The leading AI pod networking method is designed for massive models, effectively helping autonomous driving customers iteratively upgrade from single-task models to fusion models.
Platform optimization for fusion models

Based on the autonomous driving solution, a multi-camera-based spatio-temporal fusion model architecture was developed, which greatly optimized the prediction of target object monitoring speed and displacement direction. Topped the list of nuScenes pure vision 3D object detection tasks in the autonomous driving dataset, and increased the nuScenes Detection Score (NDS), a key indicator, to 62.4%.

Related Cases