Challenges In Off-Road Vehicle Autonomy
By Steve Caudill, Director, Agriculture Sector, Digital Operations, CNH Industrial [BIT: CNHI]
Steve Caudill, Director, Agriculture Sector, Digital Operations, CNH Industrial [BIT: CNHI]
As an agriculture platform manager for CNH Industrial’s connected vehicle FARM/FLEET products, I’m responsible to deliver off-vehicle software solutions to our dealers and growers that help them manage their operations. For many years, the agricultural fleet has been capable of autonomous— but attended—operation. Tractors and implements outfitted with precision guidance systems operate from prescriptions that describe exactly what the machinery should do with as much as sub-inch/centimeter precision. The vehicle operator need only line the vehicle up with a guidance line on the map, start the task and the vehicle will run the prescription without any intervention.
This hands-free version of vehicle autonomy has been available in the marketplace since the 1990’s and has steadily improved in accuracy, ease of use, and capability such that very complex sets of turns, planting depths, speeds, spraying coverage, and other task particulars are available as fully automated features. In fact, they are handled so smoothly by the equipment that when examining a field, a seasoned observer can tell exactly where an operator took the wheel or pressed the accelerator instead of letting the vehicle control the operation.
I’m frequently asked why, if this level of control is available, most of the equipment they see in the field still has an operator. Of the wide range of reasons, two stand out as both quite difficult to solve and as needing off-vehicle tools as part of the solution. They are the unanticipated and the disconnected.
When setting up a field for a precision guidance prescription, the farm manager establishes field boundaries, notes obstacles, drainage, and other field characteristics that will aid in both optimizing the operation and avoiding trouble. A well-done field setup allows vehicles pulling different implements or doing different tasks to avoid obstructions or account for variation in soil, terrain and moisture across a field. A fully autonomous and unattended vehicle can execute the prescription without flaw and even account for conditions not specifically in the prescription.
Clearly the road to unattended autonomy is neither short, nor in my case paved. Growers will need a different set of tools if they are to effectively use the precision technology already available in a fully unattended operation
This means that variation like weather, tillage depth variation, crop growth variation, and other factors can be anticipated. What happens, however, when your unattended planter encounters an unanticipated obstacle? It could be something simple. A deer is in the field. With an alert sent to a control room or a farm operator’s tablet, cameras on the vehicle can observe the obstacle and the operator can remotely honk the horn. The deer departs, the operation continues. But what if the deer is injured or dead? Some agronomic operations can tolerate going around an obstacle and dealing with it later, but many operations need to continue as planned or they have a big impact on overall yield. During planting, even short delays or deviations can mean the difference between profitability and failure.
A common problem in tillage is getting a rock or stick caught in part of the equipment such that a mound of dirt starts to build up as you drag the implement further across a field. In an attended operation, the operator simply stops the tractor, loosens the obstruction, and continues. Unless your tractor comes equipped with an all-terrain robot wielding a small sledgehammer, an unattended operation could come to a halt.
Depending on where you are, getting help to the field is either easy or a two-day adventure. Fields in places like western Nebraska, eastern Ukraine, and pretty much anywhere in Australia are large and remote. The chances of someone being nearby are slim and those fields are the ones that could most benefit from unattended autonomy.
Add to this the problem of connectivity. Remote locations are not known for great cellular or other wireless connectivity so current solutions allow the prescription to be loaded in advance so if there is a gap in the communication coverage there are no gaps in the instructions. This allows growers to avoid the high cost and complexity of satellite or self-installed radio networks for their fleet.
Even in a well-connected area weather, terrain, and obstacles like a few trailer-loads of grain can create unexpected gaps in coverage. For a fully autonomous operation, if the vehicle encountered a deer in the field at the exact spot where it was disconnected, then the alert and the instructions would not be sent resulting in a stopped vehicle and no indication as to why it is stopped.
Off-road equipment does not have the advantage of fixed public roadways, frequent traffic, roadside assistance, and emergency services. Simply guiding a technician to a vehicle for a repair can often be troublesome and in some locations the technician may have traveled half a day or more to get there. How then do product developers bring full autonomy to agronomic operations?
The agronomic marketplace is introducing or experimenting with a variety of technical solutions both on and off vehicle. Given my off-vehicle responsibilities, I’ll introduce a few from that category. Even if you are not specifically in the agronomic market, these may apply to some of your connected device use cases.
One of the recent entrants into agronomy is the drone. Field scouting via drone is a popular service and advances in imaging and drone guidance have promising futures. You can potentially solve several problems with a remotely piloted drone that uses a tractor as it’s landing pad and charging station. Your vehicle has an issue and is not connected? Send the drone straight up 100 ft with a cellular radio and WiFi or Bluetooth connected to the tractor. Now pockets of disconnectedness in an otherwise connected location are resolved.
Use your precision maps to guide a drone through a search immediately before an operation to determine if field conditions have changed. Look for unmarked obstacles, standing water, or other conditions that would hinder the task. With that information an operator at a control room could give the go-ahead, modify the prescription, or send the vehicle to another location leaving that field for later.
Crowdsource for help. Need to move a dead animal, tree branch, or stone? Crowdsource the gig economy for locals who can lend a hand for a fraction of what it would cost to send your staff to the location. A solution like this would take some setup, but the success of other forms of crowdsourcing suggest a successful model could be found.
Previous joking aside, there is a place for robotics for dislodging obstructions, reconnecting parts, or doing small things that an operator would do as a quick-fix for an issue or malfunction. Attached or independently operating robots may be a good solution for handling the unanticipated.
Clearly the road to unattended autonomy is neither short, nor in my case paved. Growers will need a different set of tools if they are to effectively use the precision technology already available in a fully unattended operation.