Optimising Fruit Harvesting Paths: A Mapless, Occlusion-Aware Picking Framework.

Area of Science:

• Robotics
• Computer Vision
• Agricultural Technology

Background:

• Fruit harvesting is labor-intensive and faces agricultural labor shortages.
• Occlusion presents a significant challenge in automated fruit picking viewpoint planning.

Purpose of the Study:

• To develop a mapless picking guidance framework for autonomous fruit harvesting.
• To address viewpoint planning challenges posed by fruit occlusion in unstructured environments.

Main Methods:

• A novel framework directly predicts the next viewing direction and estimates fruit occlusion.
• The method avoids pre-built maps and candidate-viewpoint sampling, reducing computational cost.
• An adaptive strategy balances viewpoint exploration and target approach during planning.

Main Results:

• Achieved 80.46% success rate on in-distribution data and 77.58% on unseen fruit.
• Reduced occlusion by 80.56% and 79.16% respectively for the tested datasets.
• Demonstrated effectiveness in unstructured orchard environments.

Conclusions:

• The proposed framework offers an effective solution for occlusion-aware fruit viewpoint planning.
• This approach enhances the feasibility of automated harvesting in complex agricultural settings.
• The mapless strategy alleviates local optimum bias and reduces computational load.