The use of robotics can be of great benefit for many industries, and for various reasons. For example, robotics can be used for tasks when there are concerns over human safety, or when the task is repetitive and can be done more productively by a robot working much longer hours than humans. And then there are times when robots simply offer a required level of precision that humans cannot provide.
The agricultural industry is no different in this regard. In order to remain competitive in what is now a global industry, farmers will have to be more productive, more efficient, and provide consistently good product. On top of this, the industry is also suffering from a reduction in the available skilled and unskilled labour workforce. Robotic solutions leading to autonomous farming can be used to help.
{sidebar id=333}In shifting towards autonomous farming, arguably one of the biggest challenges is in developing and integrating robotic solutions into the farming landscape. The more unstructured and uncertain the environment is, the more machine intelligence is required to achieve any required precision in farming operations. A further challenge will then be to introduce as much structure as possible into the environment prior to such operations.
If we turn our attention to broad-acre agriculture in particular, the farming operations typically involve the use of a tractor for propulsion, and an attached implement to carry out the required task. In carrying out the task, the implement experiences significant interaction with the ground, the condition of which is variable, uncertain, and often undulating. The result of such interaction is the development and application of significant and potentially widely varying forces on the implement and tractor, causing undesirable implement movement and hence presenting a challenge in maintaining precision.
At The University of New South Wales (UNSW), our proposed research aims to design and develop autonomous agricultural machinery, which is capable of carrying out a variety of precise tasks for broad-acre farming when faced with those challenges present in real agricultural conditions. We do this while also keeping track of the “big farming picture”, where the autonomous farming machinery represents one part of a unified systems-of-systems farming approach.
The Farming System
It is proposed that the farming system is in fact viewed and described as a more complex system-of-systems. As depicted in Figure 1, it is driven by a set of inputs to produce a set of outputs, often via a complex and inter-related set of sub-systems.
Various inputs, including information about land geometry, contour maps, available resources, and crop type are considered in order to determine the best or optimal crop layout and thus optimal traffic directions for the machinery. This will improve the crop laying accuracy as well as the efficiency of the machines being operated.
Central to the system structure include two data structures, namely the Precision Farming Data Set (PFDS), and Precision Agriculture Data Set (PADS). The PFDS describes the navigation and spatial accuracy requirements for the crop and provide a basis for other farming machinery sub-systems where spatial accuracy is required. In the case of broad acre farming, the PFDS will take the form of a route map for the tractors.
{sidebar id=289 align=right}On the other hand, the PADS will work in conjunction with the PFDS to ensure the agronomy requirements of the crop are satisfied. This data set is a continually evolving entity, developing as the crop growth continues and when crop sensing and other follow-up operations are taking place.
The autonomous farming machinery is viewed as a sub-system in its own right. It encompasses the operation of all farming machinery, whether partially or completely automated. Such operations include crop seeding, crop sensing, follow-up operations, and harvesting, and must of course be governed and operated in a coordinated fashion.
The operation of seeding is arguably one of the most important, where it must be ensured that the position of each plant is precise. All subsequent machinery-based operations on the crop will be then based on the seeding placement accuracy. As described, it can be difficult to achieve such precision due to several factors, including ground contact forces as well as gravitational effects.
In the crop, various parameters can be measured, such as foliage growth, soil moisture content, and weed prevalence, type, and growth. This falls under the category of crop sensing, which can be done with the aid of the PFDS for ground based vehicles, or alternatively, sensing may take place via aerial means to detect such parameters as foliage growth. The measured parameters are fed into the continually evolving PADS, to ensure the efficient and accurate utilization of the machinery used for follow-up operations.
Follow-up operations include such operations as fertilizing, and application of herbicides and pesticides. These operations are controlled by the PADS which is updated via crop sensing data, as well as the PFDS originally constructed for spatial guidance. Autonomous machinery can be used to undertake these tasks, possibly consisting of a mobile platform such as a tractor, and a means to perform the specific operation.
In the final stage of the crop cycle, harvesting lends itself also to autonomous operation. Harvesting machinery can traverse the crop field once again guided by the PFDS, and may include the use of autonomous grain collecting vehicles operating adjacent to, and coordinated with the harvester. Importantly also, the harvesting stage should accommodate on-the-fly crop yield and quality measurement, input into the PADS.
Automated Farm Machinery
At UNSW, research has led to the automation of a compact agricultural tractor, as well as the design, construction, and automation of “GreenWeeder”, a small ground vehicle robot equipped with the means to carry out non-herbicidal crop weeding. Funding from the Grains Research and Development Corporation (GRDC) in Australia has supported the construction of an active seeding implement to be pulled along by the automated agricultural tractor described.
Precision Seeding Machinery
A major contribution of the research will be in the development of robust control methodologies which will ensure that both the tractor and seeding implement track a pre-specified path not only with sufficient accuracy, but with the ability to counter disturbances from ground contact and the terrain. The tractor used to carry out the research is a John Deere compact agricultural tractor, and provides an excellent facility with which to support precision seeding.
The tractor, pictured in Figure 2, is instrumented such that it can be operated in either manned (manual) or fully autonomous mode. As can be seen, a platform located at the rear of the tractor, is used to mount most necessary equipment, including the on-board computer, motor amplifier, watchdog circuitry for safety, remote start-up circuitry, Inertial Measurement Unit (IMU), modems for all navigation, encoder circuitry, and connector boxes.
{sidebar id=334}Navigation is arguably one of the most important sub-systems on the tractor, and is achieved through the use of dual Differential RTK GPS aided by a tilt sensor and IMU system. The base station GPS and one of the on-board GPS receivers have 2cm accuracy, yielding a high precision measurement of the tractor position. The second on-board GPS receiver has 20cm accuracy, giving a location which when used with that obtained from the first GPS receiver, yields the tractor orientation. The pitch of the tractor is determined via the use of a precision tilt sensor.
An IMU is precisely mounted on the platform also, and provides acceleration information. It is primarily used for short term position tracking in between GPS measurements if required, and as a back-up to the dual differential GPS. Another use of the IMU is in dynamic vehicle modelling, where the rate gyros and accelerometers of the IMU can be used to gather data for either on-line or off-line system identification. It is important to stress the need for tractor orientation information as well as spatial position information. The orientation of the tractor will not only have an influence on the behaviour of the tractor, but also, and more importantly, on any attached implements.
Orientation information becomes a more significant issue in an agricultural setting where there are real conditions to contend with, such as ground undulation and uncertainty, sloping terrain, and tyre slippage. A 3-D angle sensor mounted at the hitch point of the tractor will deliver the global position and orientation of the attached implement. In addition to the above instrumentation, wheel encoders are installed, one on each rear wheel, and enable an back-up measurement of the velocity of the tractor, but more importantly, can be used to provide information about rear wheel slippage when its data is compared to data from the navigation system.
Simple testing of the tractor’s instrumentation and control is undertaken on the University campus. Testing and confirmation of the GPS navigation system was done on the campus oval, and yielded several sets of results, one of which is displayed in Figure 3 with the aid of Google Earth®. Here, the GPS data was obtained in real-time while the tractor was driven manually. A simple client-server application allows the navigation data to be displayed on Google Earth® on-line (of off-line later).
{sidebar id=335}
A vital part of any autonomous and unmanned vehicle is safety. On the John Deere tractor, much of the safety is the responsibility of the watchdog system, which ensures safe operation of the tractor under all circumstances. In the event of any fault condition, the watchdog system is required to halt all mechanical sub-systems of the tractor, in particular, propulsion and steering.
As software forms an integral part of the system, software failures contribute to most of the fault conditions. Software interacts with the watchdog system, by re-triggering a one-shot timer periodically. This re-triggering keeps the watchdog system alive. Failure to re-trigger the timer will result in a halt in tractor operations.
Operation of the tractor is also only enabled if a remote computer is activated and sending a periodic “heart-beat” signal via a wireless local network. Failure of the on-board PC to detect the heart-beat signal will also result in the watchdog timer not being re-triggered.
At present, low level controllers have been designed and implemented to provide automation of the propulsion (for speed) and steering systems of the tractor. This enables robust and remote control of the tractor, thus suitable for unmanned operation. However the primary goal is to eliminate any human operator all together, which requires the deployment of high-level path tracking controllers, designed to generate appropriate propulsion and steering commands for the low level controllers, which ensure the tractor follows a prescribed path or trajectory. This task is ongoing, with several high-level controllers being currently tested in simulation before their deployment onto the tractor.
Non-Herbicidal “GreenWeeder”
In general, weed eradication takes places two to three weeks into most of the broad-acre crop growth. Weed eradication requires the two stages of weed detection and weed destruction. The systems that are currently operating have crude means of detecting weeds. Any plants that appear to absorb more nitrogen are considered a weed. Weed destruction is mostly by spraying a herbicide. The current practices do not allow the herbicide treatments to be optimized to suit the weeds to be eradicated as there are no means of identifying the individual weed types. Hence there is a need to develop methodologies to detect the prevalence and the individual weed types so that the correct treatment and dosage can be applied to individual weed types.
{sidebar id=336 align=right} A more advantageous approach is to find non-herbicidal methods. Methodologies such as electrocution, electroporation, microwaving, heating and cooling etc should be considered as alternatives. This immediately eliminates the need to determine the herbicide formula and dosage and therefore, the need to identify the weed type.
These methods are particularly suitable for crop that is planted according to a PFDS. In general, the weeds that grow on the crop row itself will be defeated by the crop. However, all plants, weeds or otherwise, that grow in the inter-row space will absorb nutrients that were meant for the crop and will cause growth retardation of the crop. The authors have completed preliminary developments of a non-herbicidal weeder that has PFDS/laser/vision guided crop tracking capability with high voltage plasma arcs targeting all plants in the inter-row space.
The small foot print (0.75m x 0.50m x 0.45m) robot shown in Figure 4 has motorized Ackermann steering, electronically geared rear wheel drive and differential, a pair of stereo cameras, a laser range finder, GPS and a long range communication system. A five electrode plasma arc generation system is attached to a well insulated cradle that extends out at the back of the robot.
Research and development of this system is currently progressing with preliminary straight line path tracking control achieved in laboratory tests.
————————–
Dr Ray Eaton is a Lecturer and Director of Academic Studies in the School of Electrical Engineering and Telecommunications, The University of New South Wales (UNSW). He is working in collaboration with A/Prof. Jayantha Katupitiya of UNSW, and A/Prof. Hemanshu Pota of the Australian Defence Force Academy (UNSW-ADFA).
