Human-manipulator interface using particle filter.

Area of Science:

• Robotics
• Human-Computer Interaction
• Computer Vision

Background:

• Human-robot interaction requires accurate hand pose tracking for effective control.
• Existing methods like particle filters (PF) face translation errors due to sensor limitations.
• Human operators have perceptive and motor limitations affecting high-precision tasks.

Purpose of the Study:

• To develop a human-robot interface system for controlling robot manipulators.
• To improve the accuracy and reliability of human hand pose estimation.
• To address translation errors in particle filter-based tracking and assist operators in high-precision operations.

Main Methods:

• Utilized a 3D camera (Kinect) for capturing human hand orientation and translation.
• Employed the Camshift algorithm for initial hand tracking.
• Integrated particle filter (PF) for translation estimation and adaptive multispace transformation (AMT) for pose correction.
• Developed a methodology to correct translation errors when sensors fail.

Main Results:

• The human-robot interface system successfully tracked human hand pose.
• The adaptive multispace transformation (AMT) method improved accuracy and reliability in determining robot pose.
• Experimental tests in a lab environment validated the system's effectiveness in controlling a robot manipulator.

Conclusions:

• The proposed human-robot interface system, incorporating PF and AMT, effectively controls robot manipulators.
• The AMT method enhances precision and reliability, overcoming limitations of PF-based tracking.
• This system offers a viable solution for improving human operator performance in robot control tasks.