Digital Twin Helps Autonomous Driving Land, Smart Transportation Runs at a "New Speed"

Reprint: ITS intelligent transportation

Autonomous driving is the "crown" of artificial intelligence technology, which is considered to be a weapon to change the way humans travel in the future, and global technology giants are competing in this field. Today, the digital twin has also become the new peak of simulation technology, virtual and real interaction, deduction of decision-making, in the field of transportation, cities, architecture and shine. When digital twins meet autonomous driving, it will greatly accelerate the arrival of the era of unmanned driving.

In the recent Tencent X-Talk scientist class, Dr. Su Kuifeng, head of Tencent's autonomous driving laboratory, mentioned that the simulation test of automatic driving can help accelerate the application of autonomous driving technology, and digital twin technology is the key support. He said that Tencent's digital twin technology will build a closed loop for rapid research and development, verification and upgrading of autonomous driving, helping to solve pain points such as cost, efficiency and safety in the development of intelligent networking, autonomous driving testing, and intelligent transportation.

At present, with the emergence of automatic driving in the fields of park logistics, mines, ports, urban sanitation and other fields, the technology of automatic driving is constantly moving towards the landing from the beginning of the "sense of distance". However, there has always been a consensus in the industry that mass production applications must run through the "11 billion miles".

How far is 11 billion miles? There are more than 50 round trips between the sun and the earth. Google's Waymo, which entered the market earlier, has only completed 0.2 billion miles of self-driving test miles. In order to complete 11 billion miles of actual road testing, the time is long, the cost is high, and the testing is difficult and dangerous in extreme scenarios, which is "unbearable" for many entrants.

The amount of test volume that is difficult to complete in real-world testing is handed over to the virtual platform to complete, which is already a common practice in the industry. In view of the demand for virtual simulation testing of autonomous driving and the pain points of the industry, and with the goal of providing a safer and more efficient development and testing method for autonomous driving, Tencent has created a built-in high-precision map, virtual and real combination, online and offline virtual simulation platform TAD Sim, and at the same time uses digital twin technology to predict traffic scenarios and help management departments achieve more scientific and effective urban management.

01

Virtual Simulation Technology "Hard Power"

Building a "Driving School" on the Cloud for Autonomous Driving

Tencent began to lay out autonomous driving in 2016 and made virtual simulation technology one of its key businesses. In 2017, we developed core capabilities such as 3D scenario and sensor simulation of virtual simulation platform, rich test scenarios of data-driven traffic flow simulation, scenario-based cloud simulation and virtual city cloud simulation, which made Tencent's automatic driving virtual simulation platform in a leading position in China.

In 2018, Tencent released the TAD Sim, an autonomous driving simulation system, and released an upgraded version of the TAD Sim 2.0 in the first half of 2020. As the industry's first virtual simulation platform driven by game technology and real data dual engine, TAD Sim is leading the industry in terms of authenticity, comprehensiveness, visualization, lightweight, standardization, etc., so that autonomous driving can be tested and learned in the virtual world, and the cost has been greatly reduced.

In terms of integrity, TAD Sim provides a full-stack toolchain that can be deployed in Windows, Linux and Web environments, and is an autonomous driving simulation system with a complete test and verification process of "data acquisition, scene modeling, automatic scene generation, model in the loop, software in the loop, cloud acceleration evaluation, hardware in the loop, and vehicle in the loop".

In terms of efficiency, TAD Sim cloud simulation technology supports parallel computing of more than 10,000 scenarios, and the running time is greatly reduced. At the same time, Tencent uses the traffic flow data of virtual reality to build a virtual simulation world at the city level, supporting thousands of autonomous vehicles and hundreds of thousands of traffic flow vehicles running at the same time, and the information between vehicles in the virtual city is synchronized in real time, ensuring the effectiveness and completeness of the test while being efficient, and opening up a new pure machine test method.

In terms of authenticity, TAD Sim has developed a full set of models including vehicle dynamics, sensors, traffic flow, map and scene editing, and supports the OpenX series and OSI international simulation standards, solving the problem that some core models of traditional simulation software rely too much on third parties. At the same time, tencent relies on its in-depth accumulation in high-precision maps, game technology, AI and other fields, and uses artificial intelligence algorithms to achieve high-precision and automated three-dimensional reconstruction of scenes. In addition, the simulation results of the dynamic model in the system are more than 95% consistent with the field experiment data to ensure the true validity of the industry-leading test.

TaD Sim has reached cooperation with the National Intelligent Connected Vehicle (Changsha) Test Zone, China Automotive Technology and Research Center, National Automotive Quality Supervision and Inspection Center (Hubei), Traffic Management Science Research Institute of the Ministry of Public Security, Shenzhen 5G+ Intelligent Network Test Demonstration Platform and other testing institutions, as well as a number of leading automobile enterprises at home and abroad, and has launched real-time digital twin exploration and construction of traffic in scenarios such as smart high-speed and open urban roads. Previously, the special "automatic driving simulation and digital twin test evaluation tool chain" project for new energy vehicles declared by Chang'an University and Tencent and other units was approved and successfully selected into the national "14th Five-Year Plan" key research and development plan.

02

Digital twin technology "runs" autonomous driving

The landing ability is further improved

The superposition of the new generation of information technology has long had the effect of 1+1>2.

With the combination of virtual simulation technology and artificial intelligence, cloud computing, big data, high-precision maps and other technologies, virtual simulation of autonomous driving has been continuously applied and popularized, and continues to enter the field of traffic flow simulation and urban traffic simulation that combine virtual and real to achieve prediction of traffic scenarios.

Digital twin technology emphasizes the use of massive real data as the "raw material", builds a model of the physical world in the virtual space, and realizes the mapping of the digital twin environment and the real environment.

Su Kuifeng said that the application of digital twin technology to autonomous driving research and development testing can reduce the risk of road testing, improve the safety of the testing process, and greatly reduce the time and capital cost of testing, and improve the efficiency of development iteration. In addition, at the regression test level of the autonomous vehicle software OTA, digital twin simulation can effectively improve the test verification efficiency of intelligent driving software upgrades.

It is understood that the Tencent Digital Twin Autonomous Driving Simulation Test Platform has four core capabilities, including geometric restoration, physical restoration, logic restoration and high concurrency capabilities for traffic scenarios.

The platform emphasizes the high fidelity restoration of the scene, and realizes the calculation results of the high-precision sensor simulation through the three-dimensional scene geometry restoration and the high-precision sensor simulation model. In terms of scene logic restoration, the platform's data-driven traffic flow simulation capability can describe the operation logic of all dynamic elements inside the scene, such as pedestrians, motor vehicles, non-motorized vehicles and other traffic flows. At the same time, the platform also has simulations for control algorithms and vehicle dynamics, which can complete simulation calculations under physical collision, friction, wind and other conditions.

In addition, based on the technical advantages of Tencent Cloud, the platform can quickly complete the rapid regression test of hundreds of thousands of physical scenarios by running tens of thousands of simulation system instances in the cloud at a high concurrency, and improve the iteration efficiency of the automatic driving algorithm.

03

Digital twins are fully enabling the evolution of smart transportation

At present, Tencent has built a real-time digital twin system based on digital twin simulation technology, combined with The Radar Fusion or Radar-Only roadside real-time perception system, which integrates the road environment and traffic flow information around the driver in real time and presents it in real time on the intelligent terminal in the car, providing more accurate navigation, prediction and guidance for high-speed driving. Provide efficient and safe traffic guidance in cases of poor visual visibility such as dark nights, rain, snow and fog.

In addition, at the cross-river bridge in Huizhou, Guangdong, Tencent is trying to use digital twin technology to "control congestion", through three-dimensional reconstruction technology, digital twin technology and other traffic situations to deduce, to provide traffic police departments with diversion schemes to facilitate their decision-making; at an airport expressway in Jiangsu, Tencent helped customers create a "digital twin toll station", which has become an innovative application under the high-speed toll collection link.

It can be seen that in addition to supporting the research and development and testing of automatic driving, in a broader dimension, Tencent's digital twin technology can also provide a scientific decision-making basis and a free cloud drill site for urban traffic scheduling management and future transportation solutions, and continue to land more cases.

The physical and digital worlds have broken the boundaries, interacted and evolved, and served humanity. With the blessing of technologies such as virtual simulation and digital twins of autonomous driving, the future of smart transportation and smart cities is still full of imagination.