Towards Real-World Aerial Vision Guidance With Categorical 6D Pose Tracker.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Accurate six-degrees-of-freedom (6-DoF) object pose tracking is essential for robotic manipulation and real-world applications.
• Aerial robotics manipulation faces unique challenges, including rapid viewpoint fluctuations and significant inter-frame variations.

Purpose of the Study:

• To develop a robust category-level 6-DoF pose tracking system for aerial vision guidance.
• To address the specific challenges posed by aerial conditions in robotic manipulation tasks.

Main Methods:

• Introduction of Robust6DoF, a category-level 6-DoF pose tracker utilizing shape and temporal priors for optimal keypoint pair exploration.
• Implementation of a Spatial-Temporal Augmentation module within Robust6DoF to mitigate inter-frame differences and intra-class shape variations.
• Development of a Pose-Aware Discrete Servo strategy (PAD-Servo) with two servo action policies for aerial robotics manipulation.

Main Results:

• Robust6DoF demonstrated superior performance across four public benchmarks.
• Real-world tests confirmed the effectiveness of Robust6DoF and PAD-Servo for aerial robotic applications.
• The proposed methods successfully handle rapid viewpoint changes and inter-frame differences inherent in aerial conditions.

Conclusions:

• Robust6DoF provides a robust solution for category-level 6-DoF pose tracking in challenging aerial environments.
• The combined Robust6DoF and PAD-Servo system is suitable for practical aerial robotic manipulation tasks.
• The research advances the capabilities of aerial robots in performing complex manipulation tasks through enhanced vision guidance.