Weed-identifying robots at your command

Fast fact

  • Robots with superior visual programming can provide customised weed scouting and control tailored to a grower's particular farming system.
Photo of AgBot II

Visual weed sensors are being tested on the AgBot II robot to see how automation systems can be tailored to individual farms.

PHOTO: Queensland University of Technology

Young Queensland engineer David Hall hopes to deliver an individualised solution to the future of automated weed-management systems

Imagine a weed-control robot that could go into the paddock to find weeds, take pictures and come back to you for instructions, before returning to implement your weed-management strategy. It might sound like science fiction, but is actually a very realistic scenario for the future of broadacre weed control.

Agricultural robots being developed for weed control make strong assumptions about what weed species are expected in the field. Although this might be appropriate in some situations, it could be limiting in others as different farms can have different weed species and weeding requirements.

So a more adaptable robot is required – one that can provide growers with detailed information about weed species distribution and allow the grower to call the shots, selecting the preferred automated species-specific solutions. These could include a particular spray treatment or tactical tillage.

David Hall, a PhD student based at the Queensland University of Technology (QUT), is developing solutions to allow robots to visually detect and classify plants in a paddock and use this information to implement a weed-control plan.

Mr Hall’s vision is for a robot that can scout a paddock without any assumed knowledge and then present the grower with sample weed images and mapping. Once the grower identifies the weed species and the preferred control method, the robot would continue to scout and apply the control mechanism wherever it finds the weeds. If a weed is found that doesn’t fit the previous experience, the robot would again seek instructions from the grower.

“I think the reason why agricultural robotics haven’t been seen as being able to provide integrated weed management is, at least in part, due to the assumption that we already know what weeds to expect in any given field,” he says.

Mr Hall’s QUT supervisors, Dr Feras Dayoub, Dr Chris McCool and Professor Tristan Perez, are specialists in the field of robotic vision. Dr Dayoub says the robot of the future will quickly collect data and do clustering and analysis to provide the grower with a weed-distribution map.

Growers are now looking to target individual patches of weeds before they spread. When done well this approach can provide savings in weeding resources and increase weeding efficacy, but it can be hard to achieve using large machinery and is time-consuming.

As the accuracy of robotic systems improves, and the sensors and hardware become cheaper and more widely available, the use of technology such as that being developed in the QUT project could become the norm.

More information:

Professor Tristan Perez,
07 3138 9076,
tristan.perez@qut.edu.au

Dr Feras Dayoub,
feras.dayoub@qut.edu.au

Dr Chris McCool,
c.mccool@qut.edu.au

David Hall,
d20.hall@qut.edu.au

QUT Digital farming

End of GroundCoverTMSupplement 'Weeds'
Read the accompanying:

GroundCoverTM issue 128: May–June 2017

GRDC Project Code GRS10926

Region North