Two autonomous driving trends: Level 3 for highways and Level 4 for community roads

From Autonomous Vehicle & ADAS Japan 2017



Autonomous driving for highways and public roads is coming soon
Autonomous driving for highways and public roads is coming soon

  In May 2017 the Autonomous Vehicle & ADAS Japan 2017 exposition was held in Tokyo. This report will outline a presentation titled "Two major overseas trends that will create a drastic revolution in autonomous vehicles in the future" given at the conference by Tsuguo Nobe, Architect and Director Government Policy & Business Development, Intel K.K. and Visiting Associate Professor of Nagoya University.

  According to professor Nobe, there will be two major trends in the development of autonomous driving. First, level 3 autonomous driving on highways will be introduced around 2020, which will be the outcome of continuous innovation on the current development track for Advanced Driver Assistance Systems (ADAS) by the existing automobile industry.

  Meanwhile, the technology required for level 4 autonomous driving on community roads has rapidly progressed over the past two years, and is expected to be introduced around the same period in 2020. The response to the demand for vehicles to complete the final leg of a commute, ride sharing, and car sharing is included in this type. Although it will require considerable effort to tackle various obstacles compared with autonomous driving on highways, driving on public roads is becoming increasingly feasible with AI issuing driving inputs thanks to advances in deep learning. Since 3D maps are created for limited areas, less investment is required. Professor Nobe went on to emphasize that rather than legacy automakers, it is IT companies that will generate disruptive innovations.

  In the future, it is expected that the pyramid structures of the automotive industry's manufacturing and data driven areas will be united, but service providers and IT companies could achieve prominence over OEMs. E.U. and U.S. OEMs are increasingly moving to become service providers themselves.

Related reports:
U.S. NHTSA: Announcement of guidance for autonomous vehicles (November 2016)
Ford: Aiming to achieve volume production of driverless vehicles in 2021 for ride-sharing (December 2016)

NHTSA's "Federal Automated Vehicle Policy": Establishing a course for the production and marketing of autonomous vehicles

  At the beginning of the presentation, professor Nobe emphasized the importance of understanding the "Federal Automated Vehicle Policy" announced by the NHTSA in September 2016. This policy is important not only due to the NHTSA's position as authority over transportation in the U.S., but also as it provides guidance based on a sufficient understanding of the state of autonomous driving around the world. He mentioned that particular attention must be paid to the following two points.

  The first is that the policy now includes production and marketing, whereas the "Preliminary Statement of Policy Concerning Automated Vehicles" released by the NHTSA in May 2013 only touched on areas through driving tests on public roads. This policy clarified that vehicles driven by computers can be produced and marketed if the vehicle conforms to the Federal Motor Vehicle Safety Standards (FMVSS). However, since current autonomous vehicles do not conform with the FMVSS as is, automobile manufacturers planning to produce and market autonomous vehicles must file a petition to the NHTSA for changes of the interpretation or exemptions from FMVSS provisions to acquire approval.

  The NHTSA announced that the guidance will be revised every year.

Changes to the NHTSA's level 3 definition

  The second point worth noting in this guidance is that the definition of level 3, which includes both autonomous driving and human operation, was changed.

  The "Preliminary Statement" in May 2013 stated that the vehicle "can determine when the system is no longer able to support automation, such as from an oncoming construction area, and then signals to the driver to reengage in the driving task, providing the driver with an appropriate amount of transition time to safely regain manual control. " in level 3. However, in the September 2016 guidance, the subject of the sentence changed from the system to the driver, with the text stating "the human driver will respond appropriately to a request to intervene."

  Regarding the response of level 3 autonomous vehicles during an emergency, the guidance also indicates that; "Fallback strategies should take into account that despite laws and regulations to the contrary, human drivers may be inattentive. Such fallback actions should also minimize the effects of errors in human driver recognition and decision-making during and after transitions to manual control."

  In short, while the operation and peripheral monitoring can be entrusted to the system in level 3 autonomous driving, the driver cannot be fully relaxed, and instead must only rest to the extent that they are able to take over operation again immediately.

Differing approaches to achieving Level 3 autonomous driving

  Policies for level 3 autonomous driving differ by automobile manufacturer. Ford and Volvo will skip Level 3 and progress towards levels 4 and 5.

  • The policy of Ford is to skip Level 3 because it is difficult for drivers to maintain awareness once the operation is entrusted to the automated system and develop level 4 autonomous vehicles without pedals or even a steering wheel by 2021 for use in ride sharing.
  • Volvo also believes that safety cannot be maintained in level 3. The company announced that they will assume full responsibility as a manufacturer as long as autonomous driving is used appropriately in fully autonomous modes (Levels 4 and 5).

  However, Audi, BMW, Daimler, Toyota, and Honda plan to develop level 3 autonomous driving systems.

  • Audi will offer "Traffic Jam Control" in the new European spec A8 released in July 2017. This function performs autonomous driving in the same lane where the driver has no monitoring responsibility during congestion of 60 km/h or less under certain conditions. Although this model will be the world's first level 3 autonomous vehicle, the company deliberately set the speed to 60 km/h or less and limited operation to one lane. It has also been reported that versions of the vehicle with level 3 autonomy will not be offered when the A8 is released, because it is taking too long establish relevant traffic regulations in Germany.
  • Toyota is aiming to realize level 3 autonomous driving through the establishment of a cooperative and reliable relationship between humans and vehicles, and to ultimately develop Level 4 autonomous driving.
  • In June 2017, Honda announced that the company will "introduce an autonomous driving system in which peripheral monitoring by the driver is not required during congestion, including automatic lane changing on highways by 2020." This likely means that the company is aiming for level 3 autonomy.

Technology required for level 4 autonomous driving on community roads progresses rapidly

  The NHTSA defines level 3 to 5 autonomous vehicles as Highly Automated Vehicles (HAV) in the latest guidance, and is calling for OEMs aiming to produce and market HAVs to clarify the scope of their function as "Operational Design Domains" (ODDs) (specifically, 1) geographical location, 2) roadway types, 3) speed range, 4) lighting conditions (day/night), 5) weather conditions, and other operational domain constraints.)

  Professor Nobe presented the image to the bottom left, which illustrates the logic of ODD. The X-axis indicates speed and the Y-axis the difficulty of predicting peripheral constraint conditions (difficulty of autonomous driving). Level 3 autonomous vehicles will be introduced around 2020.

  Until around 2014, the successful development of working technology for Level 4 autonomous driving on local roads was thought to be far from being achieved (only Google was concentrating on this field from an early stage). However, the required technology has rapidly progressed over the past few years. It seems that E.U. and U.S. manufacturers have been concentrating on the development of Level 4 autonomous systems rather than level 3 particularly, after an accidental death occurred in a Tesla Model S in May 2016. (Note) Although the Model S that caused the accident was a level 2 vehicle, the potential for a problem to occur if a driver overestimates the autonomous driving system is common to levels 2 and 3 vehicles.

  The figure to the bottom right shows a typical image of level 4 autonomous driving on local roads. The bold line indicates a railroad network. This type of autonomous driving connects the various railroad station (red circles) and homes with driverless taxis and microbuses (so-called last mile mobility solutions). The market is expanding from commuters to also include areas like car sharing for vulnerable road users, and ride sharing.

  In the latest guidance the NHTSA indicated that the Authority's policy it to promote the popularization of autonomous driving as it can be expected to transform personal mobility, while at the same time, car sharing and ride sharing with autonomous vehicles will bring benefits to vulnerable road users, such as those who can't afford a vehicle and the elderly.

  Since the geographical zones targeted for level 4 autonomy are limited, the scope of the 3D maps created can also be reduced, which will minimize initial investments. The maximum speed can be reduced to 40 km/h, further reducing potential risks.

  Autonomous driving on highways is the result of the continuous innovation by OEMs, which have been working to improve existing ADAS technology. Although a response to various obstacles (pedestrians, bicycles, motorcycles, etc.) is required for autonomous driving on local roads, the feasibility of driving with 3D maps and AI commands that have grown in sophistication through deep learning is in sight. Professor Nobe emphasized that this will bring about disruptive innovations from IT companies.

ODD mobility conceptual image
Autonomous driving for highways and local roads will be introduced around 2020, and systems for major arterial roads are expected to be introduced around 2025. A conceptual image diagramming the use of ride-sharing for the last leg of a commute from railroad stations (red circles)

Example of the growth of level 4 systems: nuTonomy and DeNA

< nuTonomy >

  nuTonomy, a company developing software for autonomous driving, was established in the U.S. in 2013 by MIT researchers. In August 2016, nuTonomy began trials of autonomous taxis on public roads in cooperation with the Land Transport Authority in Singapore. The company has a plan to start providing autonomous vehicles on a commercial basis in Singapore in 2018.

  In addition to Singapore, nuTonomy also began trials of autonomous vehicles in Michigan, U.S. and the U.K. In November 2016 the company began trials of autonomous vehicles on public roads in Boston, Massachusetts.

  In May 2017 the company formed a partnership with PSA. They will start trials of Peugeot 3008s installed with autonomous driving software on public roads in Singapore from September 2017. The partners aim to run Peugeot vehicles equipped with autonomous driving technology in cities around the world.

  In June 2017 the company formed another partnership with the major ride sharing company Lyft. The companies will jointly develop means for customers to reach their destinations with autonomous vehicles. Work will be carried out in Boston.

nuTonomy autonomous
nuTonomy autonomous vehicle
(a Renault Zoe EV)

< DeNA >

  In August 2016 DeNA started joint operation with AEON Mall Corporation of a 12 passenger EV autonomous driving "EZ10" minibus (called the Robot Shuttle), developed by France-based EasyMile, in Chiba City, Japan (the Robot Shuttle runs on sites owned by Chiba City because it currently cannot be operated on public roads.) This minibus is a fully autonomous vehicle with no steering wheel and a maximum speed of 40 km/h, although it runs between 10 to 20 km/h to ensure safety.

  DeNA is also running tests on the lakeside of Lake Tazawa in Senboku City, Akita Prefecture, and the Ito Campus of Kyushu University. On April 27 and 28, 2017, DeNA held a test drive event at the Kanazawa Zoo in Yokohama City.

  Nissan will start conducting joint demonstration tests with DeNA to develop self-driving vehicles. For the first phase the companies will focus development activities for self-driving technology in areas in Japan designated as special strategic zones in 2017. The companies announced that they will expand their scope and perform demonstration tests, including utilization of technology with mobility services, within the Tokyo metropolitan area by 2020 (announced at CES in the U.S. in January 2017).

Robot Shuttle
Autonomous driving minibus Robot Shuttle
made by EasyMile of France operated by DeNA

Progress of deep learning utilization

  Hypothetically speaking, when creating programs for driving on roads as shown in the photo on the right with conventional algorithms, creating complicated structures while classifying cases such as "confirming the nearest signal, and making a decision to stop if the signal is red, locate the stop line, or keep driving if the light is green" involve a high degree of difficulty. However, utilizing AI developed with deep learning in conjunction with 3D maps may make it possible for a computer to make decisions on its own as to "how to drive."

  There is a growing understanding that autonomous driving will be realized by AI-operated vehicles. The software will be updated even after vehicles are sold via over the air updates in order to upgrade their function. The development of 3D object recognition via deep learning at data centers (Ex. AI technology that makes a system recognize an object 3 dimensionally like how humans do when a vehicle is located 10 meters ahead), creation of driving algorithms, and an optimal vehicle allocation system will be keys to success.

Evolved AI and 3D maps
Evolved AI and 3D maps enable autonomous driving on such roads

The likelihood web service providers will reach the top of the industry

  According to professor Nobe, the key for the future expansion of autonomous driving is the development of software that instructs a vehicle on how to drive. He believes that the pyramid structure of the manufacturing industry, and the pyramid structure of data driven industries will be integrated in the future. The integrated automotive industry will compete over how speedily necessary mobility services can be provided, so it is possible that companies offering services like ride sharing and those that develop autonomous driving technology could gain prominence over OEMs that currently lead in production.

  Among automobile manufacturers in the E.U. and U.S., there has been noticeable activity on the part of various companies to become service providers themselves.

Key words

Autonomous driving, level 3, level 4, NHTSA, ADAS, highway type, public road type

<Automobile Industry Portal MarkLines>