An estimated 30,000–40,000 Americans die each year in car crashes, with 94 percent of accidents attributed to human error.
Today I’m thrilled to elaborate on our funding of a breakthrough startup tackling one of the major barriers to autonomy: mapping.
Mapper is led by an all-star team of Berkeley, Carnegie Mellon and Oxford-educated Ph.D.’s and serial entrepreneurs, including CEO Nikhil Naikal and CTO Alonso Patron. The company bolted out of stealth mode last year with a feature piece in Wired, an op-ed in Fortune, and now, as they launch their globally available on-demand maps product, a meaty feature in MIT Technology Review. We’re proud to have supported this innovative company since its early days.
Mapper ushers forth the next major wave of mapping technology: machine-readable maps
Human drivers need certain types of inputs, like where to turn and what lanes to choose. We want maps to speak to us, as they guide our path one instruction at a time, and we want images to help us identify roads and landmarks. For autonomous driving, the needed inputs are vastly different. Machine-readable maps must be more precise, more granular, and perpetually updated. Autonomous vehicles (AVs) will be safer not just because they can’t get distracted or inebriated (though this is important), but also because they will know and quickly respond to the nuances of each street and its potential hazards, from overly sharp turns to particular traffic patterns and construction zones.
Machine-readable maps will enable AVs to process dangerous road conditions faster and more cost effectively — leading to more lives saved
For AVs to realize their life-saving potential, they need the ability to instantly and continuously analyze multiple inputs, and they need to do this quickly enough to have time to respond to rapidly changing driving conditions. Whether it’s troublesome weather or unexpected road obstructions, it will take significant computing power for real-time perception systems to process static environmental inputs such as parked cars, in parallel with their processing of new, evolving road conditions such as a child running across the street. Any lag in processing time could cost lives, plus the cost of creating such powerful computing embedded within cars could put this life-saving technology well outside the financial viability for most car owners.
Mapper is working to solve this issue. If AVs utilize detailed pre-existing mapping data, they can focus their computing power on dynamic changes and anomalies rather than the static basics of the road and environment. Maps enable a level of safety, repeatability and foresight that perception systems alone cannot do cost effectively or quickly. The combination of Mapper’s machine-readable maps with real-time perceptual computing will deliver technically viable autonomous solutions quicker and at a lower cost.
Cost effective and scalable mapping technology will enable even remote locations to benefit from maps
The promise of safer roads is only possible if mapping technology is not just state-of-the-art but also cost-effective and scalable enough to enable the mapping of large swaths of roads. The key to Mapper’s scalability is their ability to capture 3D maps using inexpensive sensors and a crowd-sourced data acquisition model. Mapper’s hardware contains sensors and cameras that can easily be mounted in any vehicle. To create rich 3D maps, drivers install the device and simply drive around the streets that need to be mapped. The device uses the car’s USB outlet (or cigarette lighter) for power and sends captured street data wirelessly via the driver’s mobile phone.
I personally love the clever way that Mapper is mapping the streets by utilizing cars that are already on the road, such as those driven for Uber and Lyft, trucking companies, and fleets — targeting virtually any driver looking to make extra money during down time. These drivers, deployed in a scalable, targeted way, will help Mapper create “base maps,” in any city or geographic region.
Base maps are updated regularly to reflect the dynamic nature of road conditions. This is another major advantage of Mapper’s approach. Because the cost of their mapping hardware is low, thousands of cars can be mapping the roads at any given time. This is unlike any other approach to 3D mapping, and it enables Mapper to deliver maps that are current with changing road conditions. Mapper creates and updates maps at the lowest cost while yielding the most up-to-date maps available.
We don’t have to wait decades to realize the benefits of machine-readable maps. They can make vehicles safer today.
Mapper’s technology isn’t just beneficial for autonomous vehicles — it will benefit today’s vehicles as well. We’ve already seen features such as lane and parking assistance and adaptive cruise control in some high-end cars. Machine-readable maps will improve automotive safety features for today’s standard, non-autonomous cars.
Here’s an example of how this could work. Using mapping inputs, vehicles will know how many lanes are on the highway, in which lane the car is driving, and how many lanes need to be crossed in order to exit safely. As another example, as a car approaches a busy intersection with a high accident rate, it will know to begin stopping early as it approaches a traffic light in order to mitigate the risk of accidents. Consumers are happy to pay for such features, which in turn will help fund the creation of more accurate maps and other technology to nudge us all towards a safer and more autonomous driving future.
From my perspective, machine-readable mapping can’t take off fast enough. The sooner machine-readable maps are developed, the sooner they can be deployed both to makers of autonomous vehicles and to makers of traditional vehicles — and the sooner lives will be saved. Mapper is paving the way to such a worthy future.