Fully-automated vehicles will only be as smart as the datasets they use to determine their driving pathways. Jakarto Cartographie 3D, a young company based in Montreal, is working on AI-powered, high-definition maps that it claims offer 2-3 cm (.787 to 1.2-inch) absolute precision and relative precision measured in millimeters. In other words, better maps that will allow for better automated vehicles (AVs).
Started two years ago, Jakarto currently has three Nissan NV200 mapping vans equipped with a data collection unit that contains five cameras, a GPS unit and two ZF-supplied lidar sensors that each generate one million points per second. Jakarto calls the resulting maps RoadSkeleton, claiming they'll be beneficial for smart cities and AVs. As numerous Jakarto representatives said during the company's participation at the recent 2019 Michelin Movin'On sustainable-mobility summit in Montreal, RoadSkeleton is like "Google Street View on steroids."
RoadSkeleton can identify objects such as lampposts, manhole covers, traffic lights, signs and bridge height, but not to 100% accuracy just yet. That's because the in-house software still is in development phase. Jakarto developer Antoine Chaux said the HD maps will give the company a lot of advantages when it comes to supporting autonomous driving. A passenger car might not need to know much about height, but if you are in an automated cargo truck or recreational vehicle under a bridge, it's vital information, Chaux said.
Jakarto has mapped about 5000 km (3107 miles) of the Trans-Canada Highway between Montreal and Vancouver and another 1,000-2,000 km in Quebec, Chaux said. Of the company’s ten employees, about half are coders, three are drivers and the rest are executive staff. Chaux said the next six months will be dedicated to mapping more roads in Canada, as well as starting service in France. “To have a 3D map of the cities will be an interesting market,” he said.
Some AVs may already have maps generated by lidar, GPS and cameras, but Chaux said they may not be precise enough for some applications. Finding a parking space in a city, for example, might require knowing exactly how long and wide that space is and the size of your vehicle. “With our map, you can measure precisely the size of the parking space and you will know where your car can park,” he said, adding that a second service that identifies if the space is currently available obviously also will be required, since Jakarto would pass by the space and measure it only once or twice a year.
Chaux said that Jakarto's mapping sensors can precisely measure spaces even if they are visually blocked by, say, a parked vehicle. This is because the system measures the lines painted on the ground and gets multiple views of the space as the truck drives by. Currently, one of Jakarto's vans can in a day map around 200 kilometers (124 miles) on the highway or around 30 to 50 km (19 to 31 miles) in a city, because it needs to drive slower there.
Once collected, a day's worth of driving data requires three days to process, Chaux said. This will speed up as the company grows, he promised. "In the three days of processing, we have one whole day which is someone doing it manually," he said. "We're in the process of adding more AI and are targeting a fully-automatic system."Continue reading »