원문정보
초록
영어
This article examines navigation of a flying robot inside a building environment in three dimensional spaces in which the size and location of some obstacles are not determined and other obstacles and target can be moving. This article suggests a new method by combining Q-learning algorithm and Monte Carlo algorithm on optimal navigation by the flying robot. The rewards are intended to be maximized when the robot flies in the right route; moreover, the maximum performance power would be measured according to the future predictions and the well-doing of that action would be also measured. Here, this method has been implemented with Webots simulator, and simulated data are analyzed by MATLAB. The simulation results show that control of the policy obtained from Q-learning and Monte Carlo methods is more efficient compared to traditional methods in controlling flying robot navigation.
목차
1. Introduction
2. Literature Review
3. Kinematics and Robot Model
4. Hybrid Algorithm of Q-learning and Monte Carlo
4.1. Q-learning
4.2. Mont Carlo Method
5. Hybrid Algorithm
6. Results of Combined Learning Simulation in Webots
6.1. Implementation of Webots Simulation
6.2. Analysis of Hybrid Learning Simulation Results in MATLAB
7. Conclusion
References