Toshiba's Autonomous Driving System Takes to the Road
Collaboration with Nagoya University to test system based on Toshiba's image recognition processor

Toshiba Corporation has brought its autonomous driving system off the test track and onto public roads in Aichi prefecture. Based on Toshiba's image recognition processor for advanced driver assistance systems and newly developed technologies for obstacle map generation and route creation, the system is expected to establish practical technologies for autonomous driving.

The first test was conducted successfully in September 26 in cooperation with Nagoya University, using a high-precision 3D map developed by Aisan Technology Co., Ltd., and a control system developed by Dr. Shinpei Kato, Visiting Associate Professor at Nagoya University, and Tier IV Inc.

Autonomous driving systems must perform many and complex functions at once, including real-time monitoring and recognition of the road and its environs from video feeds and LIDAR^{(Note 1)} information, and plan an appropriate course. Many autonomous vehicles now under development handle this computational complexity with high-end PCs^{(Note 2)}. Replacing such PCs with much more cost-effective dedicated processors that consume less power is vital for progress in autonomous driving.

Toshiba's system employs the TMPV7608XBG^{(Note 3)} image recognition processor, which carries out computationally intensive processing to measure 3D points^{(Note 4)} from camera images, functions usually carried out by a high-end PC. The system also uses Toshiba-developed technologies to create obstacle maps of the vehicle's surroundings from camera and LIDAR input, and to determine a course that avoids such obstacles, and has demonstrated that this is computationally feasible for vehicle-mounted processors. Toshiba has implemented these technologies on a PC-based autonomous driving software platform , developed at Nagoya University, and has now successfully brought the system onto public roads for testing.

Toshiba has identified "perception, recognition, decision-making, and control" as the core requirements that must be met by autonomous driving technologies, and through verification tests aims to realizing a practical recognition and decision-making technology by 2020.