Can AI and robotics help harvest bananas, the world’s most popular fruit?

ARM Hub participated in a project investigating the feasibility of automating banana de-handing, in collaboration with Hort Innovation and the Australian Banana Growers Council, to provide the banana industry with information on what is required to progress towards automation.

Banana de-handing is the action of separating the banana hand (fruit) from the bunch stalk (stalk).

A key advancement in recent years is computer vision, an AI tool to help robots adapt to natural variation.

In the case of banana de-handing, computer vision built into the robotic system can identify appropriate cutting locations, assisted by the contrasting green and black abscission line present where the hand connects to the stalk.

Challenges include sap, delicate banana skins and, of course, the use of knives and other cutting tools in close proximity to other staff.

Banana drama

De-handing is a crucial process on any commercial banana farm. Those responsible for the task set the pace of the packing shed and, as such, the quality of fruit heading to market.

De-handing is a skilled task but also happens to be manual, repetitive, and labour intensive, relying heavily on being able to hire and train experienced staff.

ARM Hub CEO, Associate Professor Cori Stewart, said Australia is known for its fresh fruit, however challenges, including natural disasters and working visa reductions have hit hard.

“Together, we need to take a proactive approach to address these challenges, to reduce crop loss, damage to land, as well as banana price hikes in supermarkets,” she said.

Dr Rosie Godwin, Research and Development Manager from the Australian Banana Growers’ Council, said automation could provide some practical, positive change for industry.

“This is an option we’ve been wanting to investigate for some time,” she said.

“It’s been wonderful to have ARM Hub  and QUT come on board and really hit the ground running.”

The challenge

The idea of automating this task is not new – in fact, ARM Hub found the first recorded attempt in 1985, consisting of a set of spinning blades suspended above a tank of water.

The spinning blades, and other earlier attempts, could not account for natural variations and resulted in damage to fruit.

Fast forward to 2022 and new technology has opened up a world of possibilities.

Dr Godwin and Wayne Austin from Australian Hydraulic Services (Innisfail) have been closely involved with the project, including facilitating a trip to Far North Queensland to study packing sheds and speak directly with farmers.

QUT expert and Chief Investigator, Dr Chris Lehnert, along with ARM Hub’s mechatronics engineer Dr Troy Cordie and Industrial Designer Mr Anthony Franze were able to get a detailed view of the production process and identify potential opportunities.

“Engaging with growers from various sized operations supported the ARM Hub team to develop deep insights into the banana de-handing process and understand the challenges associated with the automation task,” Mr Franze said.

What we did

• Time and motion study across five farms
• Data recording and interviews with operational staff
• Literature review of existing R&D
• Online forum with growers
• ARM Hub team developed and analysed options for automation

What the project found

The project identified three possible automation solutions.

• Option 1: Using two robotic arms and a perception package, the robot cuts the banana hands from the bunch and places them on the wheel/trough.
• Option 2: The bananas are suspended in a tank of water and slowly lifted. As the bunch is lifted, the individual hands are removed.
• Option 3: The banana bunch is angled to a tub of water and the hands are removed just above the water. Once in the water the hands are moved away with water jets.

A closer look at Option 2

• The submersion de-hander uses a robot arm, a lifting and positioning mechanism, a perception package and a cutting method (eg knife, water jet or laser).
• Standalone solution or could operate alongside one or more de-handers.
• As the banana bunch arrives on the overhead rail, the bunch is attached to the lifting and positioning mechanism and lowered into the tank. Once fully submerged, the lifting and positioning mechanism gradually lifts the bunch out of the tank. As the bunch rises out of the tank, the perception system identifies each hand and location for the cut. The robot arm cuts the hand from the bunch. Once removed from the bunch, the hand is moved away with a water jet.
• No additional requirement to handle the banana. The process is repeated for each hand on the bunch.
• A conveyor lifts the hands from the tank onto the trough/wheel.

Recommendations from the research included:

• Previously proposed automated de-handing solutions and inventions are unlikely to be viable for adoption in Australia and do not warrant further investigation or development.  These and other mechanical banana de-handing technologies are likely to have unacceptable impacts on fruit quality, including bruising.
• Advances in computer vision, machine learning and robotic handling methods provide promising opportunities for the development of a viable de-handing solution.
• For a commercial solution to be developed, the detailed design of two robotic options should be progressed to the next stage of development. Development of an automated robotic de-handing solution is being researched further.
• A cost benefit analysis would best be conducted once more information is available.
• Greater farm-level information is needed to determine benchmark/s that would make an automated system viable for use in Australia.
• Further research and development efforts should be delivered in partnership with producers and technology providers.

Read the final report via Hort Innovation

More information: ARM Hub – info@armhub.com.au