Toyota has announced it has been developing a high-precision map generation system that will use data from on-board cameras and GNSS (global navigation satellite systems) installed in production vehicles. The mapping system, which appears to have no name as yet, aims to assist the safe implementation of automated driving by using data collected from sensors on consumer cars to create automatically generated and updated maps.
This is a fundamentally different system to that currently used by the likes of HERE maps, which relies on a fleet of their own customised mapping vehicles. Toyota unveiled the proposed system publicly at this week’s Consumer Electronics Show (CES 2016) in Las Vegas, among other plans to support automated vehicle technology.
Toyota’s proposed mapping system uses camera-equipped production vehicles to gather road images and vehicle positional information. This information is sent to data centres, where it is automatically pieced together, corrected and updated to generate high precision road maps that cover a wide area.
In a media statement, the giant auto-manufacturing company stated, “Until now, map data for automated driving purposes has been generated using specially-built vehicles equipped with three-dimensional laser scanners.”
“The vehicles are driven through urban areas and on highways, and data is collected and manually edited to incorporate information such as dividing lines and road signs,” the statement read. “Due to the infrequent nature of data collection, maps generated in this manner are seldom updated, limiting their usefulness.”
Mapping is essential for emerging driverless cars technology as an understanding of road layouts and traffic rules is critical for a successful, safe and efficient integration of the technology into widespread use. Also required is the high precision and timely measurement of positional information of a range of road features, including dividing lines, curbs, and other road characteristics.
While mapping of this type is usually performed using expensive 3D laser scanners, Toyota plans to use just satellite positioning and photos collected en-masse to generate high precision road image data from designated user vehicles. Toyota has admitted that while a system relying on cameras and GPS in this manner has a higher probability of error than a system using three-dimensional laser scanners, positional errors can be mitigated using image matching technologies that integrate and correct road image data collected from multiple vehicles. Toyota has quoted a resulting margin of error to a maximum of 5cm on straight roads. One of the major benefits of the system is that widespread data acquisition could equate to maps that are updated in real time at a relatively low cost.
Toyota plans to include this system as a core element in automated driving vehicles that will be made available in production vehicles by around 2020. While initial use of the system is expected to be limited to expressways, Toyota plans to eventually cover ordinary roads and assist in hazard avoidance. Toyota has also indicated that they will seek to collaborate with mapmakers, with the goal of encouraging the use of high precision map data.
The news comes as Toyota revealed at CES 2016 a billion-dollar investment into the new Toyota Research Institute (TRI), which includes four key development goals: to build an autonomous car that is incapable of crashing, to increase access for those who can’t drive currently, to extend research findings into indoor mobility, and to accelerate development of AI and machine learning.
CEO of TRI Gill Pratt, announced that Toyota will open two TRI facilities with several high-profile new appointments and advisors. The two facilities, one at Stanford and another at MIT, will have unique project aims towards developing driverless technologies.
The Stanford project, entitled ‘Uncertainty on Uncertainty,’ will focus on teaching cars how to deal with novel situations with machine learning rather than specific programming. This would assist decision making ahead of unpredictable situations that might result in traffic accidents, and it could ultimately help make automated cars a lot more practical. “It’s one thing to teach cars to respond to events you expect to occur,” Pratt said. “But the really challenging thing is, how do we teach the car to respond safely to events we do not expect, that we don’t anticipate.”
The MIT project, called ‘The Car Can Explain,’ seeks to devise ways for vehicles to account for their actions, making it possible to identify logical problems that caused a car to behave incorrectly.
The below video from Toyota explains how the concept aims to work.
HERE last week announced the release of a similar a similar system, the HD Live Map, which seeks to create a highly detailed and near-real time representation of the road environment, based on data collected from road network sensors, fleets and probes and sent in real time to the HERE cloud to make real-time network warnings and condition reports so that connected vehicles can “see around corners.” HERE states that the data generated would be compatible regardless of vehicle manufacturer.