Self-driving Technology and the Role of the KICT
▲ Research Specialist Kim Ji-soo, Department of Future Technology and Convergence Research
Experts agree that the advent of the self-driving vehicle will revolutionize the concept of mobility in our everyday lives. Tesla recently revealed its FSD (Full Self-Driving) beta program, which naturally piqued the interest of the masses in self-driving vehicles. The term "self-driving" conjures up a number of images, perhaps of a driver watching a movie instead of the road, or passengers engaged in face-to-face conversation in a car without a driver's seat. A video released on the internet depicting a driver activating the auto pilot function and going to sleep caused quite a stir. In this sense, a self-driving vehicle can be called a "driverless vehicle." However, a driverless vehicle is just one type of self-driving vehicle, as not all self-driving vehicles can do without a driver. This report provides a definition of self-driving and the self-driving vehicle, outlines the current level of self-driving technology, and examines how the Korea Institute of Civil Engineering and Building Technology (KICT) can contribute to the self-driving field.
Definition of a Self-driving Vehicle
In English-speaking parts of the world, self-driving is also referred to as "autonomous driving" or "automated driving." But these terms lack a concrete definition or classification. In the automotive industry, "automated driving" most closely resembles "self-driving" in meaning, and the US Department of Transportation and most of its transportation and automotive-related subsidiaries are using it. "Autonomous driving," which encompasses a wider range of meanings including "the means to travel without spatial restrictions," is generally used to represent the concept of self-driving.
The definition of "self-driving" that is observed globally was announced by the US Society of Automotive Engineers (SAE) in January 2014 and is in compliance with the SAE's official standard J3016, "Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles," revised in June 2018 and in effect to this day. SAE J3016 defines driving through a 6-level structure that defines the roles of the driver—i.e., the person seated in the driver's seat—and the automated driving system (ADS), as well as the instances of driver-ADS interaction at each level. In the field of self-driving, a vehicle controlled and driven by an ADS as defined in SAE J3016 is commonly referred to as a self-driving vehicle
The Levels of Self-driving
In the structure of driving defined in SAE J3016, at levels 0 to 2 the driver is in control, and at levels 3 to 5 the ADS is in control. The role of the ADS when the driver is in control is as follows. At level 0, the ADS provides driver alerts and safety support—e.g., blind spot alerts, collision alerts, etc.—at crucial moments. At level 1, the ADS assists with either steering or acceleration/braking—e.g., lane departure prevention and adaptive cruise control (either forward/backward or sideways). At level 2, the ADS assists with both steering and acceleration/braking (forward/backward and sideways), with the driver remaining vigilant and in control of the vehicle while receiving such ADS support. Advanced driving assistance systems (ADAS) featured in vehicles available on the market today provide such level 2 driver assistance functions. At level 3, the driver can only assume direct control of the vehicle by taking over the ADS. At level 4 and above, the person seated in the driver's seat is not a driver, with the difference between level 4 and level 5 being whether the operational design domain (ODD) of an ADS is limited or unlimited.
It is commonly thought that a linear progression through these levels of self-driving must take place—that is, that one level must be perfected before the next can be applicable; however, this is not necessarily the case. Let us hypothesize a self-driving shuttle operating on the grounds of the KICT along a route between the entrance of the main building and the entrance of the Innovation Center, with set stops along the way. This shuttle, as long as no traffic control takes place on the roads within the KICT, will be able to perform level-4 self-driving—i.e., driverless driving—at a speed of 20 km to 30 km per hour without causing accidents. Indeed, with very limiting ODD conditions in place, the technology available today is sufficient for level-4 self-driving. However, because ODD limits are removed at level 5—i.e., all limits, including road and weather conditions, are removed—the vehicle in question must be able to self-drive to any destination, at least in a single country. As such, level 5 cannot be attempted until level 4 is perfected.
Self-driving Technology: The Present
Globally, the development of self-driving technology is taking place in two directions. First, automakers and parts suppliers are working to create an ADAS that can achieve self-driving; second, IT-driven companies such as Waymo (Google), Uber, and Cruise are working to enable full self-driving—i.e., level 4 and over—to create future mobility services.
With the former, level-3 self-driving is pursued by improving level-2 ADASs, or ODD expansion at level 2 is sought. That is, automakers are seeking to maintain their existing market share by continuously improving and expanding the commercial technology they already have. One exception is Tesla, which competes with traditional automakers by taking an IT- and electronics-oriented approach rather than an automotive-oriented one. In the field of ADASs, traditional automakers and Tesla share the ultimate goal of realizing level 4 self-driving or higher, but as far as the release of technology is concerned, the two are diametrically opposed. While traditional automakers remain conservative, in that they do not venture to commercialize technology that has not yet been guaranteed as safe, Tesla, as with its FSD beta, takes the risk of releasing incomplete technology. Automakers, Tesla included, have developed ADASs, e.g. HDA-2 of Hyundai/Kia, FSD of Tesla, Super Cruise of GM/Cadillac, Co-pilot 360 of Ford/Lincoln, and Driving Assistant of BMW, but none of those are yet capable of level-3 self-driving.
With the latter, the dominant trend is not the development of self-driving technology for impending commercialization, but the creation of a complete self-driving technology to be applied to the creation of different kinds of mobility services. Some prime examples of such mobility services include the robo-taxis developed by Waymo and Uber and the self-driving trucks of TuSimple, services that are still in their infancy, as well as the self-driving-car-sharing platform developed by the GM-owned Cruise, which began as an aftermarket self-driving platform. As such, a comparison with the former may give the impression that these companies have not much to show in the way of breakthroughs, but the fact of the matter is that because their goal is not the progressive development of self-driving technology but to cut straight to level 4, they actually have obtained a vast pool of data through testing that is so large in scale that it dwarfs that of the former. In actuality, these companies have already achieved level-3 self-driving, but their biggest limitation is their still-very-limited ODD. Most of their testing is carried out in perennially arid regions like California and Arizona. Indeed, ODD expansion is such a massive barrier that even Waymo, which has made the greatest progress, began testing in rainy Florida only in the fall of 2019.
Self-driving Technology: The Limit
Self-driving can be defined as the execution of part or the entirety of driving, which used to be performed by a human driver, by a vehicle or, to be exact, a computer built into a vehicle. The act of driving takes place through the stages of perception, judgment, and control. At the stage of perception, the functions of the human eyes and ears are performed by sensors, and the functions of the human brain are performed by a cognitive algorithm that connects sensors to a self-driving program. At the stage of judgment, a self-driving program takes over the functions of the human brain. At the stage of control, the functions of the human arms and legs are performed by a self-driving module (actuator). Technology involved in judgment and control is primarily developed by automakers and electronics suppliers. Because a solution to every single scenario that could occur during driving is required and self-driving must resemble human-driving as closely as possible, the functions of judgment and control can be fulfilled without perception. As such, the perception stage presents the biggest challenge in achieving the technological integrity of self-driving, or in expanding an ODD.
The situation that poses the greatest danger to a human driver is when they are not able to perceive their surroundings accurately. The occurrence of an accident can depend on the speed of reaction (judgment or control), which is the product of driving experience or individual constitution; however, if heavy rainfall severely impairs the driver's vision or lanes cannot be discerned because of snow on the road, even a driver with competent judgment and control will experience difficulty. Self-driving must overcome such diverse driving difficulties through technology, but the sensors commonly used in self-driving are all fundamentally limited in function. Moreover, any unprecedented driving situation will not be recognized, which can lead to a critical safety issue, as has been demonstrated by accidents involving Tesla vehicles. In particular, overcoming limits in ODD expansion related to weather conditions or non-standardized situations, e.g. roadworks in progress, is a huge task for industry players pursuing self-driving through ADAS or IT. Perceptive technology based on sensing technology and artificial intelligence is improving at an accelerated rate, but the fundamental limits on the function of sensors cannot be completely eliminated.
Efforts to Overcome Limitations and the Role of the KICT
The information provided in this report thus far pertains to the "standalone" automated vehicle (AV), which drives based on information fed by sensors and a precision map. To overcome the limits of the AV, research into connected automated vehicles (CAV), which combine the AV with the concept of the connected vehicle (CV) , a separate area of development, is taking place around the world. In 2015, the KICT formed plans for the creation of an HD map, the most basic information required in self-driving, and succeeded in creating one for the first time in Korea. It is also developing a range of technologies for communicating road safety information, such as road surface temperature and the presence of potholes, to vehicles through infrastructure. Another project underway is the development and testing of a local dynamic map (LDM), a dynamic information platform that detects, generates, and supplies real-time environmental information to automated vehicles.
Because not every AV is a CAV, ways to overcome the limits of an AV that does not use communication must be researched. An AV is activated by a human to be self-driven—this process takes place on the road. As such, the difficulties experienced by the human and vehicle must be resolved on the road. From this standpoint, the Smart Mobility Research Center is working on finding the causes of sensor perception reduction and addressing them through road infrastructure improvement. "Self-driving sensor data collection equipment" for understanding how a sensor perceives a road environment has been created and is being used to research sensor perception with the goal of developing road infrastructure that will remove reductions in sensor perception by 2022.
In this endeavor, only a limited range of road infrastructures will be developed; however, as sensor data research will take place on a greater scale in the future, allowing the setting of the road infrastructure standards required for self-driving, the KICT is expected to play a core function in the self-driving field.