Vision for Robust Robot Manipulation

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Domestic robots require robust object manipulation capabilities.
• Existing proprioceptive sensors for grip detection can be costly and complex.
• Vision-based sensing offers a rich alternative for robot manipulation tasks.

Purpose of the Study:

• To develop a vision-based system for robustly evaluating object manipulation success in domestic robots.
• To enable continuous monitoring of object presence and facilitate automatic recovery from manipulation failures.

Main Methods:

• Utilizing depth cameras for enhanced environmental perception.
• Employing Lab-colour segmentation to identify robot manipulators within images.
• Leveraging depth information to detect contact edges between manipulators and objects.

Main Results:

• The system accurately detects the presence or absence of contact points.
• It enables continuous reporting of object loss during manipulation.
• Experimental validation in realistic indoor environments confirms the approach's effectiveness.

Conclusions:

• Depth-vision-based sensing provides a robust solution for evaluating robotic manipulation success.
• This method enhances the reliability of domestic assistant robots by enabling object-loss recovery.
• The approach offers a flexible and cost-effective alternative to traditional proprioceptive sensors.