University researchers may have found the answer to Australia's chronic shortage of seasonal fruit pickers by developing a robotic apple harvester.
A research team led by Dr Chao Chen from Monash University's Department of Mechanical and Aerospace Engineering has developed an autonomous robot capable of identifying, picking and depositing apples in as little as seven seconds at full speed.
The robot clasps the apple in a set of suction-aided jaws, twists the fruit to break the stem and then drops it into a basket.
During trials in February and March at Fankhauser Apples in Drouin, Victoria, the robot harvested more than 85 per cent of all reachable apples in the canopy as identified by its vision system.
Dr Chen said less than six per cent of the fruit were damaged due to stem removal.
Apples without stems can still be sold but don't necessarily fit the cosmetic guidelines of some retailers, he said.
With the robot limited to half its top speed, the median harvest rate was 12.6 seconds per apple but in streamlined pick-and-drop scenarios the time was cut to about nine seconds.
By using the robot's top speed, individual apple harvesting time can drop to seven seconds.
"Our developed vision system can not only positively identify apples in a tree within its range in an outdoors orchard environment by means of deep learning but also identify and categorise obstacles, such as leaves and branches, to calculate the optimum trajectory for apple extraction," Dr Chen said.
Automatic harvesting robots, while a promising technology for agriculture, posed challenges for fruit and vegetable growers.
Robotic harvesting of fruit and vegetables required the vision system to detect and localise the produce.
To increase the success rate and reduce damage to produce during harvesting, information on the shape and stem-branch joint location and orientation were also required.
More Machinery Reading
To counter this problem researchers created a state-of-the-art motion-planning algorithm featuring fast-generation of collision-free trajectories to minimise processing and travel times between apples, reducing harvesting time and maximising the number of apples that can be harvested at a single location.
The robot's vision system can identify more than 90 per cent of all visible apples seen within the camera's view from a distance of about 1.2 metres.
The system can work in all types of lighting and weather conditions including intense sunlight and rain and takes less than 200 milliseconds to process the image of an apple.
"We also implemented a 'path-planning' algorithm that was able to generate collision-free trajectories for more than 95pc of all reachable apples in the canopy," Dr Chen said.
"It takes just eight seconds to plan the entire trajectory for the robot to grasp and deposit an apple.
"The robot grasps apples with a specially designed, pneumatically powered soft gripper with four independently actuated fingers and suction system that grasps and extracts apples efficiently while minimising damage to the fruit and the tree itself.
"In addition, the suction system draws the apple from the canopy into the gripper, reducing the need for the gripper to reach into the canopy and potentially damaging its surroundings. The gripper can extract more than 85pc of all apples from the canopy that were planned for harvesting."