Autonomous Driving – Independence Day (Part 2)

Working with Stanford University, Ford is exploring how the sensors could see around obstacles. This research would enable the sensors to ‘take a peek ahead’ and make evasive maneuvers if needed: for example, if the truck ahead slammed on its brakes, the vehicle would know whether the area around it is clear to safely change lanes.

“Our goal is to provide the vehicle with common sense,” says Greg Stevens, global manager for driver assistance and active safety, Ford research and innovation. “Drivers are good at using the cues around them to predict what will happen next, and they know that what you can’t see is often as important as what you can see. Our goal in working with MIT and Stanford is to bring a similar type of intuition to the vehicle.”

Smart Car

A stack of five computers in the trunk receives the data collected by the multiple sensor types on the car via a 1Gb Ethernet switch, which is also trunk-mounted. Most prominent among the sensors are the four, third-generation lidar (light detection and ranging) units on the roof, which are the “eyes and ears of the car”, in the words of Ford research scientist (and former University of Michigan doctoral student) Gaurav Pandey. These are made by Velodyne and each contains an array of 32 vertically mounted laser beams in an assembly that spins about the vertical axis.

The ultimate goal is not to rely on a single sensor. We’ll use multiple sensors and use data from all of them to form the navigation [path], so we don’t get stuck down the line with one particular sensor

The lidar units work like this: a laser beam is shot; when it hits an object, it returns back to the detector, which can measure the intervening time and therefore give the distance of the object from the car. Point clouds are built up from each reading and algorithms then run that can tell the car whether the ‘blob’ is, for example, a human or a car. This in turn builds an accurate 3D map of the car’s surroundings that is itself mapped onto a regular satnav map. Since the distance of the objects from the car is known, as well as how fast they and the car are moving, the car can make decisions about how to progress.

The lidar units also provide details of the reflectivity of the surrounding objects, which enables them to recognize features such as lane markings and zebra crossings – a level of detail essential for the vehicle to navigate in complex urban environments but nigh-on impossible to obtain from a GPS-based navigation system in such places – and work independently of daylight.

Lidar has some shortcomings, however, notably its inability to handle foggy or snowy conditions, so further types of sensor are required. These include a roof-mounted, omnidirectional camera (easy to mount and dismount on the research vehicles, so currently only used during map creation), forward-facing, high-resolution cameras for stop-light detection and GPS antennae that detect the speed and orientation of the vehicle via a trunk-mounted Applanix inertial measurement unit (IMU).

3D view of Ford’s stand at MWC in Barcelona generated by the Lidar equipped Ford Fusion

The research vehicles also use the sensors that are already on production vehicles and used by ADAS technologies like ACC and forward collision warning. These include forward-looking radar, a forward-looking camera, two blind-spot radars and six front and six rear ultrasonic sensors.

“The ultimate goal is not to rely on a single sensor,” says Pandey. “We’ll use multiple sensors and use data from all of them to form the navigation [path], so that we don’t get stuck further down the line with one particular sensor. Radar, for example, isn’t upset by fog so it’s helpful in those conditions. A combination of all these sensors helps us to understand the environment better.”

Not surprisingly, considerable computing power is required to bring all the sensor data together and process it to localize the position of the car in its surroundings to the required level of accuracy.

“Since this is a research vehicle, we’re free to have as much computing power as we want!” laughs Pandey, referring to the practicality compromising stack of computers in the trunk. “But computers are becoming smaller day by day, computers are becoming cheaper, and computing power itself is increasing, so who knows? In the future we may need only one computer to do all these things.”

Ford Motor Company’s “Blueprint for Mobility” calls for partnership with telecommunications industry to create an inter-connected transportation network as part of the solution for alleviating “global gridlock”

How long the research program might run is unclear but the Dearborn-based team is in it for the long haul: Ford’s Blueprint for Mobility foresees autonomous technologies perhaps appearing after 2025. As Pandey puts it, “Until we get the answers we need to the questions we have, there’s no end to the research.”