Road images collected through AI-powered dash cams are turned into mapping data. Credit: Nexar
Data is the foundation of autonomous driving. Nexar, an AI computer vision company, analyzes street information through a network supported by its dash cams. Integrating real-time data with human driving behaviors, the Israel-based company aims for a solution that would upgrade self-driving cars' training.
Nexar was founded in 2015 by Eran Shir, a serial entrepreneur, and Bruno Fernandez-Ruiz, a former senior fellow at Yahoo. The company has secured a foothold in the US and Japan, and is now opening its activity to the UK, Singapore and Australia as well. Its AI-powered dash cams collect road images that are turned into mapping data for various uses, including car insurance, construction zone monitoring, street parking, and autonomous driving.
Shir, Nexar's CEO, said customers who buy the dash cam can upload as much footage as they want to the company's cloud. In return, Nexar can access and analyze a large amount of anonymized data.
For example, the company would use videos capturing car accidents to do 3D collision reconstruction scenarios that can be plugged into a simulation. Using the reconstruction, one can teach a self-driving car to avoid a similar collision or react more safely.
Crowdsourced data enables real-time mapping
The founders started Nexar with the idea of building a vehicle network that can help improve real-time situational awareness and alert drivers, whether they are humans or computers.
While the industry strives for "zero collision" with advanced driver assistance systems (ADAS) and autonomous driving technology, Shir said there is a gap between what the industry has promised and what it can deliver now. Aiming to fill the gap, the founders began by deploying dash cams, collecting data through customers and building a real-time network with AI.
The crowdsourced approach gives Nexar leverage in multiple aspects. First, it enables the company to gather data on a large scale. According to Shir, Nexar accumulates videos capturing 200 million miles monthly across the US. In comparison, an autonomous vehicle (AV) company usually collects three to four million miles of data from one or two cities.
In addition, collecting street images from customers enables Nexar's real-time mapping data. To elaborate, Shir showed DIGITIMES Asia a digital map identifying construction and road work zones ongoing in New York City. When a user clicks on the red areas representing a construction, images of the scene will pop up.
The CEO said if a company wants to send an AV to the city, it needs the most updated information so that the car would not get into work zones.
Eran Shir, co-founder and CEO of Nexar Photo: Company
He added that Nexar gathers visual information from the cameras with AI, creating the live mapping data with sub-meter level accuracy. To reach an update frequency of seconds, the company analyzes the visual data early in the edge instead of waiting for it to get to the cloud.
The constantly updating mapping data represents the most recent street information, Shir said, which could be very useful for self-driving cars. On the other hand, Google and several mapping companies focus on creating high-definition (HD) maps. While HD maps offer accuracy down to centimeter-level, Shir said the companies can only update the information once a year or every few years because the technology is costly.
Therefore, HD maps can tell the "history" precisely but are less appealing to autonomous driving, which cares about the current situations.
Human-driving data helps AVs maneuver better
AVs will likely take dozens of years to take over the roads altogether. Before that, they must learn to interact with pedestrians, cyclists, motorcyclists and human-driven cars.
"You (an AV) are basically a guest in their environment. So, you have to understand that environment very well if you don't want to get 'thrown away,'" Shir said.
That is why Nexar recently released the Driver Behavioral Map to humanize robot drivers. According to a press release, the map covers all 50 states in the US and provides insights into different road segments, driver types, weather and road conditions. AVs trained with the data would make more sophisticated decisions regarding turning at intersections, switching lanes and more.
Shir said a human driver would accommodate their behaviors to the environment and how fellow drivers maneuver their cars. For example, one may switch to the far-right lane on a highway to take an exit at a certain point so that they would not get stuck in the traffic.
The concept of a behavioral map is offering self-driving cars human-driving data so they would drive better in various areas. Shir said drivers in Boston do not behave like those in Phoenix. Therefore, the algorithms need to be adapted for an AV to maneuver well in both cities.
The idea of the behavioral map originated from the discussion that Nexar had with an AV company, Shir said. During conversations about issues the AV company encountered when developing the technology, Nexar realized it could build something to help solve the problem.
Expanding businesses in Asia
The company raised US$53 million in a Series D funding round completed in November 2021. It is backed by Aleph, Alibaba, Ibex Investors, Qumra, State Farm, and others.
Nexar has scaled up its business in Asia this year. After entering the Japanese market for over three years, it established a subsidiary in the country in April.
Shir said the company partners with Mitsui Sumitomo Insurance and powers the insurance giant's next-generation collision claim process. Nexar's collision reconstruction service has been deployed in over 250,000 connected dash cams in Japan.
Besides expanding businesses in East Asia, the company opened an Asia-Pacific headquarters in Singapore at the beginning of this year.
"We've seen a lot of opportunities (in Asia)," the CEO said.
He added that the region has adopted a lot of camera-related technologies compared with countries such as the US. Secondly, Nexar looks forward to working with automotive OEMs and AV companies based in Asia.
Shir said Asia is such a strong market that Nexar decided to invest in it before expanding to Europe.