The Innovators Solving Ag Robotics’ Perception Problems

Autonomous agriculture robots have come a long way. From tractors with autosteer capabilities to self-driving weeding robots, today’s farmers have an abundance of established and emerging technologies at their disposal, with many more on the way.

The race to deliver solutions to the industry’s biggest problems—labour shortages, rising costs and increasingly strict environmental regulations, to name a few—has prompted many startups and manufacturers to use off-the-shelf technologies in their robotics. There are pros and cons to this approach. The resulting machines are often quicker to reach the market but slower to complete their designated tasks.

One area where this is particularly apparent is in perception. Hexagon | NovAtel, global leaders in GNSS positioning and autonomous solutions, have studied the challenges. They are working to ensure the next generation of agriculture robots have what it takes to get the job done—safely and reliably.

“The majority of the sensors that are being used by agriculture companies today have been brought over from other industries,” says James Szabo, Agriculture Autonomy Product Manager at Hexagon | NovAtel. “LiDAR, radars and ultrasonics have all been used in the automotive industry for a long time. Their availability is very high, meaning people can just go and buy them off the shelf, but it is difficult to use those same sensors reliably in agriculture. We face a range of challenges that most other industries don't face."

Szabo gives the example of driverless cars. Although these vehicles operate similarly to autonomous robots, the vehicle is driving on a perfect black surface where the smallest obstacle it must detect and avoid is often a human pedestrian. Ag robots have a whole range of other variables they must be able to perceive and navigate.

“Ultimately, the farm is such a complex and dynamic environment that it’s very hard to take something designed from another industry and just push it into the agriculture space,” Szabo says. “What we're seeing today is the companies trying to solve this problem by adopting these available technologies are finding that they're not proving to be all that reliable. As a result, they have to put such tight constraints around what the robots are doing that it almost becomes unfeasible for the customer to run these machines. They’re not viable and difficult to scale. The robots either operate at really low speeds or with really narrow working widths.”

Updating these solutions requires an innovative approach. Even the transformational technologies of yesteryear are not sufficient to power the fully autonomous farms of the future. Fortunately, NovAtel has operated on the cutting edge of this space for a long time. They are up to the challenge.

“Agriculture has a long and successful history with GPS and GNSS positioning, and the autosteer technologies that have been around for 20 years now and completely revolutionised farming,” Szabo says. “These innovations have led to major efficiencies, which is fantastic for farmers and the environment, but they rely on a human being in the cab in the event of something going wrong. When we start to remove the operators, we need to look very carefully at the technology being used to see if the machine is actually where you think it is.”

NovAtel has focused on GNSS positioning systems as the best option for reliability. This technology not only has high availability, but it also boasts a high level of integrity. GNSS is resilient to outside interference from trees, telephone poles and other potential obstacles.

With the right positioning system in place, the next step toward full autonomation is to improve the robot’s perception capabilities. Here again, the automotive industry’s off-the-shelf sensors frequently leave something to be desired.

“NovAtel is building up perception systems that go beyond these sensors,” Szabo says. “They are part of the jigsaw, but we’re not just blindly taking sensors from one industry and using it in agriculture. We’re adding in an extra layer of vision-based artificial intelligence to ensure these sensors are reliable enough to give our customers a more robust perception system. The advantage is that our systems enable autonomous machines to become faster, wider and scale over time.”

The biggest challenge the NovAtel perception systems overcome involves separating out false positives and negatives in a machine’s field of vision. When five mature corn plants have the same silhouette as a standing human, the perception system needs to be able to accurately identify which objects are plants that can be driven over and which are not. A high level of confidence in an autonomous robot’s ability to know the difference will profoundly impact the industry overall.

“Whether farmers want to use these machines to replace skilled labour, reduce production costs or simply avoid spending 12 to 16 hours per day driving a tractor up and down the rows, these technologies are going to allow farmers to focus their attention on what provides the most value to their operation,” Szabo says.

While these perception systems are still being perfected, NovAtel is working with industry partners to drive them forward. Szabo is confident the forthcoming products will deliver on their promises.

“Hexagon and NovAtel have a long history of providing solutions to the problems that our ag customers face,” he says. “We work with our partners to determine what issues need to be solved and offer customised solutions—something our partners value. We’ve been doing this for a long time on the positioning side. Now, we are approaching perception with the same problem-solving mentality.”