Tencent Autonomous Driving Overview ft. Simulation

The four giant companies in China (Baidu, Alibaba, Tencent and Huawei) have all joined the race for autonomous driving technologies. In an earlier post, I dived into Huawei’s plan and strategies. This article is about the Tencent Autonomous Driving (TAD) layout featuring simulation.

Tencent’s auto ambitions date back to March 2015 when the company signed strategic agreement with Foxconn and Hexie Auto aiming to develop and produce what we call today “intelligent connected electric vehicles”.

In five years, Tencent has identified its role as the connector and toolbox for the auto industry. It focuses on software and services to OEMs, government, test zones, industry alliance and research institutes through three basic competencies:

• Data cloud (DevCloud)
• Simulation (TAD Sim)
• HD map

Tencent believes data is key throughout the R&D cycle of autonomous driving technologies. The data cloud platform (DevCloud) with its tools comes in handy for AV companies to do data crunching efficiently in a closed loop. Last year, Tencent announced that it would build a high-performance computing data center for BMW to handle data from test trials in China.

The simulation platform (TAD Sim) is capable of conducting a closed-loop simulation of all modules including perception, planning and control under different driving conditions and scenarios like rain, snow, low-light conditions or a crowded three-lane road.

TAD Sim Product Overview:

Tencent used to be the user of 3rd party simulation software hence has a good understanding of the pain points. Combining expertise in game engine, virtual reality and cloud technologies, the company has build TAD Sim offering 3 core features:

• Scenario and sensor simulation: Tencent uses 3D reconstruction technologies to recreate road traffic and city scenes with less than 3cm precision as well as virtual engine to reproduce various weather and light conditions to ensure the maximum representation of the physical world.

• Scene generator supporting multiple sources: besides the traditional methods (defining a scene with the creator tool or playback of actual road data), Tencent is training traffic AI for highly interactive scenes by applying authentic surrounding objects’ behaviors. This is powered by Agent AI technologies, commonly used in gaming.

• Local and cloud based: simulation can be conducted on the cloud as scenario based or in virtual city mode. Virtual city mode is about loading a city-level HD map to the system and deploying multiple vehicles in it running 24x7. It helps to discover scenarios where the algorithm fails and then, the scenario-based simulation can perform dedicated tests to eventually iterate the algorithm.

By the end of 2019, Tencent claimed the ability to run the equivalent of 100 autonomous vehicles driving 100,000 km a day in the simulation. They also won the project to create digital model of the National Intelligent Connected Vehicle (Changsha) Test Zone for simulation purposes.

As one of the map vendors, Tencent is approaching HD map by using retrofitted surveying vehicles out in the field to collect data (15 1st generation vehicles and 2nd generation vehicles to join soon) as well as lab team to process data and produce HD map. (Reference: China HD Map Evolvement — Ultimate Data Ownership) Up till this point, Tencent has finished the base map for all the highways across China.

Tencent HD Map Solutions:

Tencent HD Map Coverage (till now):

Disclaimer: all the exhibits in this article are taken from Tencent public presentations with Chinese words modified to English if not originally in English