Towards BEV+Transformer Based Autonomous Driving — Case Study On Baidu-Geely Collaboration (Part 1)

4 min readNov 22, 2023

--

Leveraging years of expertise in the autonomous driving sector and serving a global clientele, maadaa.ai proudly presents our new case study series, called “Towards BEV+Transformer based Autonomous Driving”.

This series aims to explore and discuss the latest advancements in the autonomous driving industry, sharing insights and case studies that reflect our experiences.

As a premier provider of AI data services, maadaa.ai serves a wide range of autonomous driving scenarios. Our auto-driving data services have spanned from 2D to 3D and point cloud technologies. Recently, maadaa.ai announced our MaidX Auto-4D auto-annotation engine supporting BEV multi-sensor fusion data.

In the first case study of our series, we look at the AI technologies that power the recently launched Baidu-Geely collaboration, Jiyue Auto, an auto-driving startup from China.

Stay tuned with maadaa.ai as we journey through the exciting world of autonomous driving.

Accordingly, Jiyue Auto is using AI technologies like BEV+Transformer, Pure Vision, and End-to-End Large Language Model (LLM) technologies for its own production of a recently launched model — Jiyue 01.

Although Jiyue Auto is backed by two Chinese giants, Baidu and Geely, and has received more attention since its inception, maadaa.ai focuses more on its performance and technologies.

Let’s dive in and find out!

1. What is Jiyue01?

With the incorporation of Baidu Apollo Advanced Intelligent Driving ANP3.0 and Geely SEA Haohan architecture, the entire lineup comes standard with Qualcomm 8295 and dual NVIDIA Orin chips, 28 Smart Driving Sensors, as well as BEV + Transformer Pure Visual Perception, and OCC Occupancy Network.

The support Baidu provides in this car is its ANP3.0 technology developed by Baidu Apollo and Baidu’s Ernie Bot LLM capability.

Baidu Apollo is an autonomous driving platform. ANP3.0 is an L2+ level intelligent driving software and hardware-integrated product solution. It includes the self-developed intelligent driving domain controller ACU and the full-domain intelligent driving software system with urban areas.

Geely’s SEA Haohan architecture is an open-source pure electric original architecture focusing on the user’s driving experience. SEA can provide customers with excellent connectivity, shared vehicle functions, e-motor capabilities, and more.

2. Test drive review

The main content of the following review is taken from a test ride video in the urban area of Shanghai, which includes high-frequency urban use scenarios, such as the CBD in Shanghai’s Pudong District.

In addition, we have also collected various test drive videos of Jiyue to provide you with a comprehensive and objective review.

The car completed all the road sections, without human intervention. These included complex intersections with mixed traffic, on- and off-ramps, continuous pilot assistance in tunnels, and other challenging scenarios.

2.1 Advantages

Driving in tunnels: When the GPS signal is weak or unavailable, it can still accurately locate and overcome the challenges of poor lighting and many obstacles. Able to autonomously detect detours.

Driving on urban roads: It can autonomously plan turns based on navigation information and actively detour when encountering disabled vehicles. Can actively avoid pedestrians and two-wheeled vehicles occupying the road.

High-speed driving: It can autonomously enter and exit ramps, merge in and out of traffic, and actively avoid congested vehicles.

Productivity: Prioritize faster lanes and improve traffic efficiency while ensuring safety.

2.2 Weaknesses

U-turn scenarios and automatic entry into straight-ahead pending stop zones are not supported and require human intervention.
Unable to identify construction pylons and barrels. On both highway and city roads, the test car had problems with not recognizing the barrels. It was finally solved with human intervention.
Others. Some test drivers found the car’s driving strategy to be slightly conservative, with weak right-of-way awareness, not decisive enough to change lanes, and so on. But as a self-driving car, safety must be the primary consideration.

(In the next article on this topic, we will explore the AI technologies behind the test drive, as well as how to deal with massive autonomous driving data. Please Stay tuned!)

For more information, please visit:

Autonomous Driving - maadaa.ai

Autonomous Driving Autonomous vehicle technology has made a breakthrough in the past decade, with the development of 3D…

maadaa.ai