CANGO Auto View: Dawn of the Autonomous Driving Era

The auto industry embraces new technologies 

“The most important milestones happening right now are autonomous driving and changes in AI,” said Tesla founder Elon Musk at the latest earnings call. “Once relevant laws allow, millions of vehicles will increase, perhaps overnight, five-fold their performance. Nothing can compare against this.” 

Dramatic reform behind autonomous driving

`Elon Musk wasn’t delusional when he said that. As the auto industry upgrades itself, autonomous driving has become the most important battlefield for vying capital and entrepreneurs. Also called “intelligent connected vehicle”, autonomous driving (intelligent + connected) enjoys the same status as electrification and shared mobility and has long been recognized by the auto industry as one of the four new trends leading future development of the industry.   

 This technology is likely to trigger a future revolution, because from a business perspective, in the era of autonomous driving, vehicles will no longer be vehicles but will become the third space for users. High-level autonomous driving means that hands, feet, eyes and attention will gradually be liberated, and the transition from “machine-assisted human driving” (L2) to “human-assisted machine driving” (L3) and further to “machine driving” (L4/L5) means the freeing up of vehicle owners’ productivity and time. As a result, vehicles will no longer be a means of transportation. Instead, they are likely to evolve into the third living space outside home and office, since the users can carry out both entertainment and work inside their vehicles.

 Autonomous driving vehicle has redefined its own product, and its commercial value will expand into a larger number of dimensions. (Autonomous driving has created a new consumer economy and productivity market, i.e., passenger economy in which passengers carry out consumption or work or entertainment on the road, so that any vehicle can become a piece of mobile commercial real estate.) 

Once autonomous driving technology reaches a highly advanced level, drivers will have more time for other things. Autonomous driving can save users fifty minutes a day to be used for work, relaxation and entertainment while on the road. And because autonomous driving frees up drivers’ attention, the original driving scenario will become a different, and important, scenario for shopping, information delivery, relaxation and entertainment.

Moreover, Tesla and automakers worldwide are expected to define and adjust their strategic positions in the autonomous driving field. Baidu plans to realize commercial production of autonomous vehicles in 2020; BMW plans to launch fully autonomous vehicles in 2021; Ford plans to introduce autonomous vehicles in 2021; and Daimler states that trucks capable of autonomous driving on regular roads will be successfully developed in 2020.

Technology giants are focusing on “transferable” consumer groups so as to quickly capture traffic and maintain matching business models. On the other hand, with their key technology and commercial capital, OEMs will become quick followers and invest in autonomous driving research. Then, while their penetration into high-end fields deepens, OEMs will wait for the cost of core technology and of entire vehicles to drop.

In addition, more innovative products are expected to derive from autonomous driving. Over recent years, a lot of mobile application innovations have taken place, and the majority of them come with a corresponding payment model. Carpooling, sharing the use of one vehicle, “e-hailing” taxi substitute and point-to-point vehicle rental---all of these are attracting investment and on an apparent upward curve. In particular, the “e-hailing” model is reporting tremendous growth in terms of annual investment funds and market penetration.

Autonomous driving has also accelerated the development of robots available to users. It’s highly possible for the wide penetration of autonomous driving to accelerate the development of robots (including humanoid robots) available to users, because these two can share many technologies including advanced remote sensing, ultra-precision positioning/GPS, image recognition and advanced AI. Besides technology sharing, autonomous driving and robots can benefit from using the same infrastructure including shared charging stations, service centers and M2M (machine-to-machine) communication networks. These commonalities will likely prompt participants to invest in the application of both, which has in fact been proven by the generous investment made in robots by selected automakers and high-tech participants. 

5G+AI present two keys to autonomous driving

With the wide application of AI, autonomous driving technology has become more reliant on the internet, as seen in the obtaining of high-precision maps, precise navigation and other data from the cloud, and its security feature has become even more prominent.

Currently, common autonomous vehicles populating the market use L2 autonomous driving technology, and Tesla’s AutoPilot is a typical example. Autonomous driving technology at L2 and below is still assisted driving technology. Even though hands can be freed up to a certain extent (as in HandsOff), environmental perception and takeover still need to be executed by humans. In other words, humans will have to observe the driving environment and directly take over in an emergency. Therefore, how can a breakthrough of this technical limitation be achieved?

An autonomous vehicle works as follows: the surrounding environment is perceived through on-board sensors including cameras, lidars, millimeter-wave radars and ultrasonic devices, judgments and decisions are made based on the obtained information, and corresponding strategies are formulated by modeling such as predicting how this vehicle and other vehicles as well as pedestrians are going to move in the immediate future and making collision avoidance plans. 

The key advance in L3 is real-time monitoring of the environment and the ability to react to it, and the main challenge is whether the machine’s perception capacity meets requirements. Driving an L3 vehicle, the driver will only need to take over control of the vehicle or make judgments when prompted by the system, while normal acceleration, deceleration, turning and other operations are handled by the system.

The current industry consensus is that 5G/V2X technology opens up the external “brain” for autonomous driving. Vehicle-to-everything communication V2X connects vehicles to networks or connect vehicles into networks and it includes vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), vehicle-to-network (V2N) and vehicle-to-pedestrian (V2P). Through V2X communication, the external “brain” is opened up for autonomous driving, resulting in the input of rich and timely “external information” to effectively compensate for sensory blind spots in individual vehicles’ built-in intelligence.

In other words, V2X is an accelerator for autonomous driving as it can effectively supplement individual vehicles’ intelligence and enhance response efficiency. At present, using 5G networks, known for their low latency, high throughput and high reliability, can improve the richness and timeliness of information transmitted by V2X as well as the technical value of V2X sensors.  

On the other hand, NVIDIA CEO Jensen Huang is of the opinion that the essence of autonomous driving is AI computation and the required computing power depends on the desired function. For him, since autonomous vehicles need to make judgments, and then decisions, about the surrounding environment, and what actions to take is essentially an issue of AI computation, the vehicles must be equipped with an AI super processor so that recognition, reasoning and driving can be carried out based on AI algorithms. According to Horizon, a startup designing autonomous driving chips, realizing L3 autonomous driving requires at least 20 teraflops of computing power and requirements for computing power will increase exponentially for L4 and L5.  

The rise of AI machine vision as represented by deep learning has successfully broken through the technical bottleneck of L3. Take, for example, the L2 assisted driving from Mobileye. It still employs traditional machine vision based on back-end rule base and matches back-end rule base with data input by front-end cameras to recognize and track objects. The biggest issue with traditional machine vision is that the rule base is limited whereas the environment faced by vehicles is unlimited. With the introduction and development of the deep learning framework, however, AI’s ability to process and classify images has dramatically improved and the percentage of errors has dropped directly. Take, for example, ImageNet machine vision competition. Machine vision within the technical framework of deep learning shows apparent quantitative improvement over traditional machine vision. Increasingly mature AI machine vision supplemented by high-precision maps has laid thick padding for safety, thus playing a key role in breaking through the technical bottleneck of L3. 

L4 autonomous driving scenarios can be realized by introducing AI technology as represented by reinforcement learning and opening up the external “brain” through 5G. Traditional driving decision-making systems based on search or rule engines often adopt only very conservative driving strategies such as immediately braking for obstacles. However, changing lanes, overtaking other vehicles and cutting off another vehicle are situations often faced in daily driving, and the current systems handle them by manually designing various intricate strategies. Oversight in strategy design, could led to crashes and human fatalities. How to make machines drive truly like humans and learn to make autonomous decisions is the key to L4.

The success of Google’s AlphaGo in the game of Go is an important landmark, as AlphaGo innovatively introduced a brand-new AI leaning framework including reinforcement learning and simulated the way people think, thus marking a key breakthrough in machine intelligence. With the introduction of a augmented learning framework, autonomous vehicles can think and learn like AlphaGo and make autonomous decisions. In addition, the introduction of V2X as represented by 5G is equivalent to opening up the external “brain” for autonomous driving. As a result, more real-time and comprehensive external information can be provided to autonomous vehicles, and multi-vehicle coordination and interaction can be better realized, thus leading to a breakthrough of the technical bottleneck in individual vehicles’ built-in intelligence and the realization of L4 autonomous driving scenarios.

Gradual opening up of policies, and co-existence of opportunities and challenges

While autonomous vehicles are not recognized by law in most countries and regions, the US, EU, Japan and other developed countries and regions believe that autonomous vehicles are the development trend for transportation and should be encouraged by legislation.

For example, certain states (districts) in the US, UK, Sweden, Germany, Japan and other countries and regions already have relevant legislation in place, but it generally targets assisted driving through autonomous systems on board L1-L3 vehicles. It stipulates that autonomous vehicles drive on the road only for test purposes and normally sets strict conditions and requirements for such road tests. It also stipulates driver supervision so that in an emergency, human driving mode can be switched to immediately. And relevant traffic rules and liabilities are still confined to the existing legal frameworks.

In China, the year 2020 has seen the accelerated application of AI, 5G and other cutting-edge technologies and consequently the continued boom of autonomous vehicles. With the current status and future trend of the autonomous driving industry in mind, the state and local governments have issued a series of supporting policies to help the autonomous driving industry chain grow in a healthy, rapid and sustainable manner.

On February 24^th, the Ministry of Industry and Information Technology and ten other ministries and commissions jointly released Strategy for the Innovation and Development of Intelligent Vehicles. Per the strategy, by 2025, technical innovation, industrial ecology, infrastructure, regulations and standards, product supervision and network security systems for China’s standard intelligent vehicles will have been completed on the whole. At the same time, intelligent vehicles capable of conditional autonomous driving will have been in mass production, and intelligent vehicles capable of highly autonomous driving will have been commercialized in specified environments. 

 On April 16^th, the Ministry of Industry and Information Technology also released Key Points for the Standardization of Intelligent Connected Vehicles in 2020. The key points are clear. The Ministry of Industry and Information Technology will accelerate the building and completion of a standard system for intelligent connected vehicles, achieve Phase I goals under Guidelines for Building a Standard System for the National Vehicle Network Industry (Intelligent Connected Vehicles), and set up a standard system for intelligent connected vehicles that can support driving assistance and lower-level autonomous driving.

As of now, national-level autonomous driving test sites authorized and regulated by the state are officially open to the public in Shanghai, Beijing, Hebei, Jiangsu, Changsha and other places, and the test roads are of various types including rural, in-town, urban, high-speed and off-road.

Once autonomous vehicles are on the road, however, accidents are bound to happen due to certain reasons. How to assign liabilities and make fair rulings are issues that require in-depth discussion and verification. At present, no country has formulated sound policies or regulations specifically for autonomous vehicles, and to truly popularize autonomous vehicles, this is a problem that will have to be solved with extra care.

Nonetheless, the general belief is that as technology matures, autonomous driving policies will become more specific, and guidance and management of the development of the autonomous driving vehicle industry will become clearer. In the future, with the accelerated implementation of various policies and gradual improvement of new policies and new regulations, autonomous driving technology will propel the auto industry into a new era.