The drive to autonomy: Companies gear up with sensors, strategies

For the past decade, widespread deployment of autonomous vehicles (AV) has been just over the horizon — that imaginary line that recedes as you approach it.

It has been delayed mainly by technical issues, which will eventually be followed by legal and regulatory ones, mainly regarding liability, and by a struggle to gain public acceptance. When they finally reach the mass market, however, AVs will reduce traffic fatalities by at least an order of magnitude because they do not get distracted, drunk, drowsy or enraged and are much better able than humans to gauge distances and speeds.

Image: IGphotography/iStock/Getty Images Plus/Getty Images

Additionally, they will be able to communicate with each other and with the infrastructure, which will not only further improve safety but also reduce congestion and fuel consumption via the adoption of techniques such as convoying.

Logically, even if AVs only somewhat reduced traffic fatalities (about 38,000 per year in the United States), the public should welcome them with open arms. In reality, though, the reaction to even a single death caused by an AV — like the one in Tempe, Arizona, in March 2018 — can set AV deployment back years.

Therefore, car manufacturers are challenged to develop AVs that can navigate extremely safely in a wide range of traffic, road and weather conditions. For more than a century, human drivers have routinely managed sudden obstructions, poor visibility and dangerous behavior by other drivers that still bedevil their new robotic counterparts, despite the sensors, microprocessors and algorithms at their disposal.

The primary technological obstacle to widespread deployment of AVs on roads is “the complexity of the system and the amount of time that it takes to develop a functionally safe autonomous vehicle,” said Steve Ruff, general manager of Trimble’s On-Road Autonomy Division, which develops positioning solutions for autonomous vehicles that operate on public roadways. He cites the time required to develop “a comprehensive, safe, autonomous vehicle technology stack” and points out that “we are on the verge of going from level two to level three, which requires the driver to stay engaged in the driving experience in case the autonomous system has a problem.”

Multiple sensors

While AV developers are exploring different ways of obtaining reliable sub-centimeter positioning accuracy, all generally rely on collecting data from multiple sensors on the vehicle and applying an algorithm to synthesize the data in real time and generate a continuous, accurate position. Computer vision, radar and lidar play important roles in an AV by perceiving its surroundings and localizing it to an a priori map. This functions well in feature-rich urban environments, but can degrade in sparse highway settings.

Radar has good ranging accuracy, but is unable to detect and recognize traffic signs and road markings. Lidar has even greater ranging accuracy but is challenged in featureless areas, such as straight highways and country roads. Digital cameras are good for detecting objects and navigating in tunnels and urban canyons, but, like lidar, are less effective on featureless roads and in low visibility conditions (rain, fog, darkness, snow, sun glare).

Plus, they are challenged by the absence of road markings or the presence of construction. Inertial navigation systems (INS), while excellent at compensating for brief GNSS outages, can only guide vehicles for short stretches due to their inherent drift. (INS are essential on aircraft and vessels, whose attitude is constantly changing, but that is not relevant for vehicles, which travel essentially flat relative to, and at a constant distance from, the road surface.)

GNSS and Corrections

Satellite navigation plays a central role in an AV. At a minimum, it guides it from a trip’s origin to its destination, including stops or waypoints in between, the same way it would advise a human driver. It also continuously alerts the vehicle to upcoming stops, slowdowns, turns, congestion and other challenges that are already mapped—whether long in advance by map makers or moments earlier via crowdsourced updates. Finally, if sufficiently accurate, it can steer the vehicle to keep it in the center of its lane and to make smooth lane changes and turns. Determining on which road a vehicle is requires an accuracy of less than 5 meters; determining in which lane it is requires an accuracy of less than 1 meter; and determining where in the lane it is requires an accuracy of less than 0.5 meters.

Two kinds of GNSS corrections are commonly used for AVs: real-time kinematic (RTK) and precise point positioning (PPP). RTK, which is generally accurate to the centimeter level, relies on ground-based reference stations at fixed, surveyed locations that process and transmit error-corrected signals to receivers within a 10- to 20-kilometer range, typically in real-time via a cellular link. PPP, which is accurate to the tens of centimeters, uses a global network of ground stations to generate an accurate signal, and transmits it to subscribers via the internet or geostationary satellites. However, the receiver in the vehicle needs 20 to 60 minutes to align with the PPP signal before it can rely on it.

Both RTK and PPP are established in industries such as mining, construction and precision agriculture, where vehicles operate in controlled environments with little or no traffic. AVs on public roads present a far greater challenge. A car’s typical range far exceeds that of any RTK base station, and base stations can also have down time, while in-vehicle systems must use multi-frequency receivers to reduce the convergence time of the PPP signal. In case of outage of either the GNSS signal or the correction signal, the vehicle’s system must rely on data from its other sensors and recover swiftly from the error state.

Trimble’s RTX is road ready

The first PPP service in commercial use for passenger vehicles is Trimble’s RTX, which provides real-time, centimeter-level positions via IP/cellular connection or satellite broadcast worldwide. It delivers positioning via satellite to GM’s Super Cruise, a hands-free driver assistance feature for use on limited access freeways.

“We’re GNSS receiver-agnostic,” said Steve Ruff of Trimble’s On-Road Autonomy Division. “We’ll use any receiver that’s preferred by the OEM building the AV.”

Image: Trimble

Trimble, he recalled, became GNSS agnostic with regard to automotive navigation nearly 15 years ago, when it decided to get out of the commercial-grade or consumer-grade GNSS business. “It has worked out quite well, because not only can we meet the quality costs and performance targets of our OEM customers, it also allows us to do what we’re good at. We can take our positioning solution, adapt it to work with any measurement engine, and put together a solution that fits the OEM’s requirements just right.”

Automotive companies, Ruff explained, generally have certain requirements for the GNSS receiver, including certain standards for application-specific integrated circuits (ASIC) and automotive safety integrity level (ASIL), as well as meeting their accuracy requirements. “So, if the receiver has suitable code and carrier phase measurements that can support their accuracy level, then that will be the third requirement for the receiver for the automotive segment.”

For off-road vehicles for agriculture, construction and mining, Trimble only uses its own receivers, said Thomas Utzmeier, general manager of the company’s Off-Road Autonomy Division. Their requirements center on precision, position availability in challenging environments, and integrity of the position. “In the use cases on which we are working,” Utzmeier said, “we certainly see sub-decimeter accuracy. We are targeting probably three, four, sometimes five centimeters.” In more challenging use cases, GNSS plus sensor fusion — including INS and optical data — maximizes position availability and accuracy, he explained.

For the on-road segment, Ruff’s division offers a “positioning stack” that includes corrections, the GNSS position algorithm and inertial fusion. “Then we provide services to help the OEMs take our software and integrate it on the platform of their choice.”