As it is expected that 14.5 million autonomous vehicles will be on the road worldwide in 2025*, an era of autonomous driving is just around the corner. Amid the rapid development of specific systems for practical use, "TierIV Academy (Cram School to Learn How to Construct Autonomous Driving Systems)" was held at TechShop Tokyo in Akasaka. The academy offers students opportunities to create autonomous driving systems on their own using open source technology and experience autonomous vehicles first hand. We asked Shinpei Kato, general producer of the program and associate professor at the University of Tokyo, how the program works and the present and future prospects of autonomous driving.
*: BCG analysis, just-auto.com July 2014 market analysis, LMC Automotive, HIS Global Standards
Autonomous Driving System Development Heating Up Around the World
As the practical application of autonomous driving approaches, a revolution that is shaking the foundation of the automobile industry has arrived. Its impact will extend beyond car manufacturers to the energy, insurance, logistics, advertising and many other industries. In addition, new players who are boldly trying to take on the challenge of this new trend with products and services that we have never seen are emerging one after another. Some such products and services include unmanned passenger services, on-vehicle information content, and autonomous driving insurance services.
Participants in this TierIV Academy also include people in charge of management strategies, in addition to engineers and researchers who are aiming to put autonomous driving into practical use. In the 5-day course, which includes a full range of activities from seminars to practicing the skills learned, more than 30 participants earnestly worked on the program in the hope of finding tips for tomorrow's business.
DAY 1: Commentary on the implementation of autonomous driving systems
DAY 2: ROS (Robot Operating System) seminar
DAY 3 and 4: Autoware (open source software) seminar
DAY 5: Practical training on operating autonomous vehicle (at a driving school)
Commentary by the Academy began with Shinpei Kato, Associate Professor at the University of Tokyo, explaining world trends related to the development of autonomous driving systems.
Kato said, "Today, many of the autonomous driving systems introduced in the world are laser-based and millimeter wave-based systems as well as camera-based systems." He then started to explain the features of each of these systems.
Google has also adopted a laser-based system. In this system, a device called LiDAR, which is attached to the vehicle, emits laser (invisible rays) to measure the distance to objects surrounding the vehicle. Kato says the laser- and millimeter wave (radio wave)-based systems both have advantages and disadvantages.
"The millimeter wave-based system adopted by Tesla uses less expensive equipment, reducing the initial costs. However, it can only be used in places where there is not much information, such as on expressways, because the system's ability to identify objects is insufficient. Meanwhile, the laser-based system is highly capable of identifying objects and can therefore handle urban areas that have complex topography. However, the initial cost of this system is high, with LiDAR costing 1 to 5 million yen per unit, although the price of the equipment is expected to drop in the future."
Although laser and millimeter waves each have strong and weak points, students of the session learned about systems using LiDAR for the purpose of driving in urban areas. In this system, the distance between objects in the periphery of the vehicle is measured by LiDAR installed on the vehicle. The information is then converted to high-resolution 3D map data, and the route for autonomous driving is set. Finally, a vehicle control program that enables the vehicle to follow the specified route is built to complete an autonomous driving system.
At the Academy, Kato and other lecturers held seminars where they constructed the foundation of those autonomous driving systems by using "Autoware," open source software jointly researched and developed by eight universities.
Collecting Knowledge of University and Autonomous Driving Researchers to Share Their Know-how with Participants
Autonomous driving is cross-cutting complex technology. The University of Tokyo specializes in super computers and operating systems, Nagoya University in on-vehicle systems and data analysis, and each of the other universities has its own specialties. Researchers and students at these universities conducting advanced research and development served as the lecturers of TierIV Academy, sharing detailed knowledge based on each university's specialty.
Seminars Providing Basic Features of Autonomous Driving, Including Recognition, Judgment and Operation
From Day 2 onward, students who belong to each university presented seminars as lecturers in lieu of Kato. Seminar participants carried out exercises using computers. The exercises started with using ROS (Robot Operating System). ROS is open source middleware originally developed by Open Source Robotics Foundation (OSRF) for controlling robot movements. Due to its versatility, it is recently widely used among engineers as a platform for developing autonomous driving systems.
On Days 3 and 4, participants learned the basic functionality of applications installed on the software. The autonomous driving system they learned about in the session requires hardware, such as cameras and GNSS (GPS), in addition to LiDAR. Vehicle control becomes possible by processing information obtained from such hardware. These applications consist mainly of seven applications, and the following introduces some exercises using them.
What is most vital in autonomous driving is high-resolution 3D map data. In order to set the location of the vehicle and the route it will be taking based on map data, information obtained from LiDAR is used to visualize surrounding topography using RViz, software with visualization functionality. On the display, road width and building shapes are represented by collections of points in 3D. As the vehicle drives on the road, the groups of points move accordingly. In this exercise, students experienced trouble adjusting the number of points. The denser the point groups are, the easier it is to recognize the topography. However, higher processing load affects operation. With advice from the lecturer, participants used trial and error in adjusting parameters to find the optimal figures.
Detecting objects is considered to be another difficult exercise. In this process, clustering is performed using point groups obtained from LiDAR and sensor fusion of camera data to identify whether the object is a vehicle or pedestrian in front of the vehicle or on the opposite lane. "At the current stage, the system determines the object based on candidate obstacles that are set in advance. However, a mechanism for automatically distinguishing objects based on machine learning will be available in the future," explained Kato.
These tasks continued to be performed each day for many hours. However, participants did not lose concentration and they frequently asked questions to resolve their problems one by one during the exercises.
Finally, the Test Drive! The Compilation of Four-day Session: What Are the Results?
On the last day of the Academy, a test drive was conducted for autonomous driving systems at a driving school in Tokyo. A vehicle manufactured by ZMP Inc. with LiDAR and GNSS equipment installed for autonomous driving was used for the test drive. Participants leaned forward to observe the mechanism of the vehicle placed in front of them and asked many questions.
Once the preparation is complete, the test drive started finally. When a participant sit on the back seat and gave instructions from the on-vehicle computer, the vehicle slowly started to roll without him touching the wheel or accelerator pedal.
One of the participants was surprised and said, "I understand it theoretically, but it's still weird when I actually see it happening." LiDAR with a height of only 20 centimeters placed on top of the roof of the vehicle rotated at a high speed to collect data on surrounding topography. In this way, LiDAR operates even when the vehicle is moving to accurately take the route. Despite driving in the rain, the vehicle operated trouble-free without being affected by the weather.
We heard from some of the participants who finished the experiment, and many of them rated TierIV Academy highly. One of them commented, "I did not expect my development to take shape to this level in only four days. I will take advantage of this experience in my research going forward." Another commented, "Although our development of autonomous driving has yet to start, I gained a lot of wisdom."
"TechShop" Co-creation Space Stimulates Creativity
At last, the five-day workshop finished. We heard from Kato at the end of the workshop. Expecting future cooperation with TechShop, he said, "We usually hold this program in a bare meeting room, but doing it in a creative space like TechShop was stimulating for us, too. TechShop creates the hardware, we create the software. Each of us has different output, but they are finally combined and deployed as a single product. We would like to use TechShop as the base of our co-creation activity."
What course of development will autonomous driving systems take in the future? In Part 2 of this article, we will discuss with Kato two major barriers that must be overcome to spread autonomous driving in the real world and future prospects.