Inventory of well-known self-driving simulation platforms, Waymo, Tencent on the list

How far is driverless driving from us?

Before sending out this question, first of all, we should know that if self-driving cars want to really go on the road, they must go through multiple safety tests, and road testing is an essential and important part.

According to an estimate by the RAND think tank, a system of self-driving needs to be validated for at least 110 miles (about 17-18 billion kilometers) before it can reach the conditions for mass production. That means that even a fleet of 100 test cars, tested 24/7 at an average speed of 25 miles (40 kilometers) per hour, would take about 500 years.

Meanwhile, a full-fledged self-driving car would also need to be sensitive enough to recognize objects on the road and react correctly and quickly in extreme conditions such as blizzards, rainstorms, and intense sunlight exposure. However, in the natural environment, these extreme conditions do not occur often, and many extreme situations are very dangerous and must be estimated in the test. Therefore, it is inefficient, costly and not comprehensive enough to rely solely on the field to carry out self-driving road tests. In such a situation, simulation platforms that can provide a perfect virtual reality road test environment become a cost-effective option for self-driving cars to conduct road tests. Today we will take a look together, those domestic and foreign well-known automatic driving simulation system is how to carry out virtual road test.

NVIDIA: Drive?Constellation - a computing platform based on two different servers

NVIDIA, the world leader in programmable graphics processing technology, officially announced last March that it had listed its cloud-based automated driving simulation platform Drive?Constellation. Constellation.

Drive?Constellation, a cloud-based self-driving car test system that uses photo-realistic simulations, is a computing platform based on two different servers. The first server runs NVIDIA?DRIVE?Sim software, which simulates self-driving car sensors such as cameras, LIDAR, and radar.DRIVE?Sim software generates photo-realistic data streams to create a large number of different test environments, for example, it is able to simulate different weather conditions such as torrential rains and blizzards, changes in light at different times of the day, and all different types of road conditions. changes, and all different types of road surfaces and terrain. The second server is powered by the NVIDIA DRIVE Pegasus AI automotive computing platform, which runs the complete self-driving car software stack and can process simulated data like that from sensors on the road-going car.

NVIDIA Drive?Constellation

Tencent: TAD?Sim - a simulation platform like a large-scale RPG

It's well known that many of the mainstream simulation systems for self-driving cars are currently based on game engines, and Tencent, which specializes in game Tencent, which specializes in game development and management, has also applied its professional game engine, industrial-grade vehicle dynamics model, and virtual-reality traffic flow technology to TAD?Sim (Tencent?Autonomous?Driving?Simulator), which is a unique advantage for Tencent to do the automatic driving simulation platform.

In TAD?Sim, a scenario that is infinitely close to the real world, not only can it satisfy the constant iteration testing needs of self-driving cars, but also improve the efficiency of the research and development of self-driving technology.

The high-precision maps built into TAD?Sim can complete the closed-loop simulation and verification of all the modules of the real car, such as perception, decision-making, and control algorithms, and the geometrical simulation of the environments of different weather and lighting conditions, as well as the sense of the test vehicle, and the geometric simulation of the test vehicle. Geometric simulation of different weather and lighting conditions, as well as testing the vehicle's sensing ability, decision-making ability, and vehicle control simulation. Combined with the collected traffic flow data and the simulation of extreme traffic scenarios, TAD?Sim is capable of holding all kinds of aggressive driving and extreme conditions of autonomous driving tests, and completing the tests that cannot be conducted in the real world in a more efficient and safer way.

Last year, Tencent and BMW entered into a strategic cooperation in the field of autonomous driving, analyzing and processing complex and changing driving scenarios in China, and helping BMW develop autonomous driving technologies and products that meet the Chinese market. In addition, TAD?Sim can also provide policy-making departments and traffic management departments with tests on traffic scheduling management, road and traffic planning, and research on automated driving regulations, etc. TAD?Sim can also help BMW to develop automated driving technologies and products that meet the needs of the Chinese market. TAD?Sim can also assist policy makers in traffic management, road and traffic planning, and automatic driving regulation research.

Tencent's TAD?Sim simulation of complex road conditions

Waymo: Carcraft - the "trailblazer" that turns reality into virtual reality

From the initial " Google drone car" to the later Google Waymo, Google can be said to be the "big brother" in the field of self-driving research. Not long ago, Google announced that Carcraft, Waymo's self-developed simulation test software, has simulated 10 billion miles of road scenarios, and supports Waymo models for large-scale testing.

Since self-driving cars need to use network nerves and algorithms for continuous learning, Carcraft sets up a wide variety of road conditions from which the vehicle can learn more. many of the situations Waymo?test cars encounter on the road can be blurred directly in the simulator, while programmers can also overlay multiple situations to create a variety of extreme situations. The data from Carcraft's simulations can then be fed back to the real-world test vehicles, and the cycle continues as the vehicles become more and more powerful. For Waymo, which was previously using "scenario playback" for testing, Carcraft plays an unprecedentedly important role in self-driving research, and even for the world's self-driving technology, Carcraft is of great significance.

Late in 2019, Waymo announced the acquisition of Latent Logic, a UK-based simulation technology company, to help Waymo realize more realistic simulation technology for better self-driving prediction and planning.

Currently, the domestic automatic driving simulation system is still in its infancy. Even international companies with more mature simulation technology are still not mature enough in the development of Chinese road scenarios and the simulation of Chinese driver behavior. From another perspective, this actually releases a good opportunity for local Chinese companies developing autonomous driving simulation technology. We expect more and more local technology companies like Tencent to join in the construction of automatic driving simulation systems, build simulation test environments with Chinese characteristics, and help mature driverless technology to be realized as soon as possible.

This article comes from the author of AutoNavi, and does not represent the viewpoint of AutoNavi.