Delegates at LandWISE 2015 formed small groups to specify a “robot” that would address some issue of relevance.

Development of any such machine should follow a systems engineering approach. Four critical features of systems engineering described by Marcel Bergerman are:

1. Take a top down approach
   • Focus on the “what” not the “how”
2. Formally define requirements
   • A thorough test plan can be developed before the product/system is
3. Have a life cycle orientation
   • Consider all aspects from the cradle to the grave
4. Inter-disciplinary in nature
   • Systems are too complex to do it all on your own

For this exercise, delegates were asked to concentrate on the first two; define what their dream machine had to achieve and describe the constraints under which it must operate.

It is important to take this approach to avoid limiting the possible solutions too early. Deciding your scout robot would be electrically powered three wheel drive eliminates aerial, two wheel or four wheel and wind or fuel powered machines. Sure, the answer may seem obvious, but the final form of many creative solutions that are optimal are not envisaged at first.

Some examples from the exercise include:

A root vegetable harvester that must have minimise soil compaction, remove soil from the produce, minimise mechanical crop damage, and work in a range of weather conditions

A bird scarer for seed crops that must keep birds off the crop, not damage the crop, and must stay within the boundaries of the crop or farm

A pea crop sampler that picks plants randomly through the crop, analyses samples on the go in the paddock and sends the resulting information back to the office. It must cope with the height of the crop and not damage it, and work in a wide range of weather conditions and cope with pugging and mud

An automated machine to remove weeds from vegetable crops grown on beds. It may also monitor plant population and variability, health and vigour, and pests and forecast yields. Working in a range of weather and soil conditions it must have high productivity (ha/hr)

Determine insect populations across field prior to planting of spring crops, including density, location and spread of identified species – include a lure for underground species

Harvest ripe mandarins after determining brix and colour levels and sort according to quality and size with rejects dumped and in-grade fruit transferred to bins for transport

An automatic pine tree pruner that removes every branch up to a height of 6m. Cuts must be clean with no damage to branch or trunk, and it must operate in difficult terrain

An apple harvester that harvests only export grade fruit such that no post-harvest grading is required. Must work 24/7 during harvest within current tree architectures

A fresh broccoli with automated cutting, picking and packing into bins. Will select/sort according to size and colour with no flowering heads.

LandWISE 2015 Presenter Tom Botterill

Tom described advances in developing a viable grape vine pruning robot.  This work is through the MSI-funded project on “Vision Based Automated Pruning”.  Tom’s research interests include 3D reconstruction and modelling using computer vision.

Grape vines must be pruned every year in order to increase yield, prevent disease, and control excess growth. While some simple mechanical pruning systems are available, most of New Zealand’s winegrowers prefer to prune vines manually, even though hand-pruning is often the most expensive task in the vineyard.

To hand-prune a vine, the healthiest canes are selected and the rest are removed. For a robot to prune vines in this way, it must understand the 3D structure of the plant, it must be able to decide which canes to keep or remove, and it must be able to make the required cuts without damaging other canes.

A team at the University of Canterbury are developing just such a robot system. The system is mounted on a platform which straddles the vines and moves along the row, pruning as it goes. The robot uses three high resolution digital cameras to image the vines, then uses computer vision algorithms to make a 3D model of the vine from these images.

Given these vine models, an artificial intelligence system decides which canes to keep and which to cut. The artificial intelligence was “trained” to make good pruning decisions by providing it with examples of how pruning experts prune vines. To prune the vines, a six-jointed robot arm reaches amongst the canes and makes the required cuts with a spinning cutting tool.

The major technical challenges in the current system were building a 3D model of the complex vine structure, then reaching to cut the vines while the platform moves. After four years of development, the 3D models are finally complete and correct enough to make decisions about where to prune, however the dynamic robot arm control is still under development, so the robot must stop at each plant to make the cuts.

Our system has been made possible by rapid advances in computer vision and robot technology over the last decade, and it is close-to-working despite being far from the simplest way to automate pruning. The algorithms and techonology are now available to automate many harvesting, pruning and precision-agriculture tasks, and there are now no technical barriers to seeing all of these tasks automated out in the field.

This project is funded by MBIE and is a partnership between the University of Canterbury, NZ Winegrowers, Pernot Ricard NZ, Scott Technology, Lincoln University and the University of Otago.

Tom’s publications and more information: http://hilandtom.com/tombotterill/