Robust Kalman filtering cooperated Elman neural network learning for vision-sensing-based robotic manipulation with...

Area of Science:

• Robotics
• Computer Vision
• Machine Learning

Background:

• Uncalibrated robotic manipulation requires accurate models and camera parameters, which are often difficult to obtain.
• Existing visual servoing methods can be sensitive to calibration and modeling errors, limiting performance.

Purpose of the Study:

• To propose a global-state-space visual servoing scheme for uncalibrated, model-independent robotic manipulation.
• To develop a method that avoids performance degradation due to calibration and modeling errors.

Main Methods:

• Utilized robust Kalman filtering (KF) and Elman neural network (ENN) learning techniques.
• Learned the global map between vision and robotic workspace using an ENN, approximating the global space Jacobian.
• Employed KF to refine ENN learning and update ENN weights for precise pose convergence and global stability.

Main Results:

• The proposed scheme successfully achieved precise convergence of the desired robotic pose.
• Demonstrated robust performance in simulation and experimental results using a six-degree-of-freedom robotic manipulator.
• Validated the model-independent and uncalibrated nature of the visual servoing scheme.

Conclusions:

• The global-state-space visual servoing scheme offers a robust solution for uncalibrated robotic manipulation.
• The integration of ENN and KF effectively handles the learning and refinement of the robot's workspace mapping.
• This approach enhances robotic manipulation accuracy and stability without relying on camera or model parameters.