Off-highway engineers are making solid progress in the push to complete truly driverless vehicles, though they are not yet close to the mainstream. Development is being aided by continuing advances in the range of sensors and receivers that provide input.
These sensors are coming down in price while their capabilities are rising. On the control side, the need for computing power continues to benefit from the nonstop advances of the semiconductor industry. As engineering staffs press beyond operator assistance systems toward full autonomy, they’re already targeting applications that seem to offer the greatest potential.
“The largest markets will be for autonomous vehicles running in well-understood, constrained environments such as we currently see with mining vehicles and container-handling systems,” said Dave Reeve, Senior Principal Engineer at Hemisphere GPS. “Agricultural and earth-works applications for this well-mapped, constrained type of operation are huge.”
Product designers in the off-highway market are leveraging advances made in other fields. Many components come from driver assistance systems being used in passenger cars, while some concepts and components come from aerospace, where requirements are similar but deployment is already widespread.
“The control and navigation problem is not that different in ground or air,” said Dave Vos, Control Technologies Senior Director at Rockwell Collins. “Both need light weight and low power.”
Many of the technologies needed for autonomous operation are currently being proven in military applications. Some of the sensors and processing techniques are used in small unmanned vehicles, while others are helping operators by maintaining some control over the vehicle so humans can focus on battlefield tasks.
One of the most basic inputs for autonomous driving comes via the GPS receiver. When design teams move from driver assistance to driverless guidance, these inputs must change. Inexpensive GPS systems claim accuracy tolerances as great as a meter. While that may work for driver assistance and in some autonomous applications, most designers do not want vehicles that continuously drift a few feet.
“The level of detail and frequency of update required of GPS for autonomous driving is considerably higher than when GPS is used as a driver’s aid,” said Francis X. Govers, Chief Engineer of Land Solutions at Elbit Systems of America. “Certainly the designers of unguided vehicles want the highest possible accuracy in centimeters and update rates of at least 10 Hz.”
On large vehicles, there is also a trend to increase the number of GPS receivers on a vehicle, which can help ensure that the control system can rapidly determine where the vehicle is headed and how well trailers are following.
“This may consist of a couple of receivers on the same rigid vehicle body to provide heading information or may involve multiple antenna disposed across a nonrigid assembly, such as a tractor towing an implement,” said Reeve.
Engineers also have to devise ways around the occasional GPS dropout. A range of sensors can be brought to bear, providing solutions that often leverage inputs from other systems such as steering control.
“Dead reckoning can effectively fill the gaps in GPS coverage. We can accurately use odometry, wheel angles, and compass guidance that would allow highly useful DR navigation,” said Govers.
When vehicles move at high speeds, these sensors can keep guidance systems updated so they stay on track. Input from accelerometers and wheel speed sensors have far faster cycle times than GPS updates, making them more useful at high speeds.
“Inertial systems have a much quicker response than a GPS receiver,” said Reeve. “When coupled to a GPS, they can provide a stream of positional data at a rate suitable for the precision control of a vehicle traveling at speed. Inertial systems also allow the stream of positional data to continue after loss of the GPS signal.”
Design teams have devised techniques that allow ground vehicles to maintain precise routes even when GPS signals are blocked for long periods. “We’ve shown a very low-cost sensor that degraded only a few meters when we lost GPS for 15 minutes while moving at 30 mph,” said Vos.
Processing all these inputs is no small task. Video and radar generate data that is high in both volume and frequency. That means that CPUs must process and analyze a lot of complex data in very short time frames.
Simply using the fastest CPUs is too pricey. Instead, many design teams are focusing on software techniques that reduce the processing workload.
A big part of keeping computing power and cost low is developing control algorithms that do not require many CPU cycles. In some environments, that makes it possible to use fairly slow 32-bit processors.
“We can do a plane with a MHz-range CPU, though usually we go over 100 MHz,” said Vos. “We don’t need to go anywhere near dual cores. We still focus on elegance. It’s an important element for success. We don’t waste cycles or watts of power.”
Another critical factor is to make sure that everything is synchronized. For example, data from GPS, cameras, radar, and other sensors must all be tied together precisely so that decisions are accurate. The fact that sensors and systems come from many companies exacerbates the challenge.
“In most projects, everyone works with a lot of subcontractors and a lot of types of data—video, audio, and radar—that all have different data rates,” said Tron Kindseth, Principal Engineer at RTI. “Everything in the system has to be put into a unified model so it can be integrated.”
The unified model makes it easier for hardware developers to develop new parts, since they can use the same interfaces. The need for easy access carries out to other parts of the vehicle. Standard interfaces make it simpler to swap modules out when parts fail. Kindseth noted that middleware is also important when application software is upgraded, also making it possible to change media and move to faster networks.
As processors get faster and systems are able to pull together larger volumes of data, engineering teams are figuring out ways to do more. One technique is to let many different workers throughout a site share information. That could lead to even higher levels of efficiency.
“We’re doing a lot to promote the idea of connected sites, where software ties into what machines are doing and shares it with project managers and others on the site, making the whole site management team more effective,” said Roz Buick, General Manager at Trimble Navigation Ltd.