In general, an autonomous vehicle could drive itself around based only on its sensing systems without having any access to maps. Unfortunately, while such a vehicle might be able to avoid collisions, this would severely limit the overall capabilities possible with its autonomy. The type of detailed maps with constant updates required to create a robust autonomous mobility on-demand system require substantial resources. This is something that startup Civil Maps is trying to address through crowd-sourced data collection.
The Albany, California-based company recently announced a $6.6 million seed funding round led by Motus Ventures that also includes money from Ford. Civil Maps was founded in 2014 and began development of its crowd-sourced map-building platform in 2015.
“There are three layers to the navigational ecosystem for autonomous vehicles: strategic, tactical, and the decision engine,” said Civil Maps co-founder and CEO Sravan Puttagunta. “We are focused on the middle tactical layer that includes a more granular level of detail such as lane configuration, traffic signs, and signals.”
The strategic layer of mapping data includes the metadata about street names and directions of the type found in current navigation systems. This data can be used for overall routing to a destination; however, it is often inadequate for the low-level control that happens in the decision engine that sends commands to the vehicle actuators.
The tactical layer helps the decision engine determine which lane the vehicle should be in to make the turn that the strategic layer has asked for in 200 meters. This layer will know if the intersection has a traffic light, a four-way stop, or a roundabout. As a result of constant updates from vehicles in the field, it will also have awareness of lane closures and detours for construction—or just general road reconfiguration. While traditional map-makers such as TomTom, Here, and Google (and now even Apple) have begun to collect this sort of data in specific areas, the update frequency is low.
Civil Maps has developed a software layer that automakers can integrate into vehicles that have depth perception sensors in order to turn them into real-time probes. To collect this type of data, a vehicle needs either a stereoscopic camera—like those used by Subaru and Daimler—or a lidar sensor. The company has also built a cloud platform that aggregates and validates the data by cross-checking it from multiple sources.
The raw sensor data would be processed locally in the vehicle and filtered into vector data for uploading to the Civil Maps platform. To make the amount of data processing and transmission manageable, the company has devised a task management system that would see different vehicles assigned to gather lane markings, traffic signals, and more.
Ford Fusion Autonomous Prototype Testing at Mcity
(Source: Ford Motor Company)
OEMs Key to Developing Revenue
Civil Maps is still working out the details of its revenue model, but Puttagunta acknowledged that it will likely have two components. In the future, when autonomous vehicles are deployed using Civil Maps data, OEMs may pay a license fee per vehicle for the base data set. But before that happens, there will be a credit system for data contributions and use. For every set of data uploaded from a vehicle, the OEM would earn credits that would be spent when updated data is withdrawn and sent to cars.
Navigant Research’s 2015 Autonomous Vehicles report projects that more than 4 million autonomous-capable vehicles could be sold by 2025, and these will all need detailed 3D maps. If automakers adopt the Civil Maps approach in the next few years, they could help build those maps without operating expensive fleets of street-view style vehicles.
Tags: Automotive Mapping, Autonomous Vehicles, Mobility, Transportation Efficiencies
| No Comments »