Extracting Objects for Aerial Manipulation on UAVs Using Low Cost Stereo Sensors

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Unmanned aerial vehicles (UAVs) offer expanded application possibilities with object manipulation capabilities.
• Rotary-wing UAVs, with hovering ability, are particularly suited for in-flight manipulation tasks.
• UAV manipulation must function effectively in natural, partially known environments.

Purpose of the Study:

• To develop an onboard object extraction method for autonomous grasping by UAVs.
• To enable grasping without requiring a predefined object shape model.
• To provide essential information for a UAV's grasping module.

Main Methods:

• Generating a local map using depth information to identify object candidates.
• Utilizing Support Vector Machine (SVM) classification to recognize or reject object candidates.
• Creating sparse cloud representations to determine object centroids and dominant axes.

Main Results:

• The developed method successfully extracts object information for autonomous grasping.
• The system demonstrates robustness in processing noisy point clouds from low-cost cameras.
• Accurate results were achieved for object extraction and grasping parameter calculation.

Conclusions:

• The proposed method facilitates autonomous object grasping for UAVs in natural environments.
• The approach is effective even with limited prior knowledge of object shapes.
• This research enhances the manipulation capabilities of UAVs for diverse applications.