Mehdi Sadeghzadeh, David Calvert, and Hussein A. Abdullah
Visual servoing, Reinforcement learning, Q-learning, Fuzzy neural networks
This paper proposes a self-learning approach for an autonomous robot manipulator visual servoing system. The self-learning system uses Q-learning to find the optimal policy based on reinforcement methodology. This policy is used by the robot to reach a predetermined object that has been randomly placed in the environment. The Q-learning algorithm is implemented using fuzzy neural networks to estimate the Q-evaluation function for each robot action. Each fuzzy neural network is trained using the input state and the Q-value for the basic action in on-line training episodes. The input state consists of the extracted image features. A camera mounted on the robot end-effector captures the target image in each iteration and sends it to a feature extraction unit. This self-learning system learns the optimal policy in order to select the best basic action that maximizes the cumulative reward received at each time step. This learning approach does not use robot or camera models, or require calibration. The results demonstrate the effectiveness of the system to learn the highly non-linear mapping between the continuous work-space and the optimal action policy.
Important Links:
Go Back
