Adaptive behaviours for robotics

• Main Author: Huang, Jiqing.
• Other Authors: Lin Zhiping
• Format: Final Year Project
• Language: English
• Published: 2011
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics
• Online Access: http://hdl.handle.net/10356/46004
• Institution: Nanyang Technological University

Unmanned Ground Vehicles (UGVs) have been playing more and more important roles in both civilian and military world. Most of the current UGV uses an algorithmic approach which is predictable and works only in certain fixed situation. Our main focus is to develop an adaptive type of UGV with smart reasoning. This is done by implementing the different clustering techniques, reinforcement learning and exploration strategies into the AI Engine which functions like the learning tool of the UGV. Comparisons between the different kinds of techniques and strategies were made using the first simulation. From the simulations, we can conclude that using Q-learning as our reinforcement theory and Boltzmann Exploration as our exploration strategy would obtain better results. It is also noted that using two vectors instead of three to represents the state, action and reward of a UGV would reduce the number of rules generated and hence reduce the time required for selection and learning at each time step. Lastly, updating of the Q-value at the end of the episode has also been proven to have a faster convergence of the Q-value. Both the second and third simulations were designed to test the AI Engine‟s practicability and capability to execute in a real time environment. We have obtained a near optimum result for the second simulation where the AI Engine learned a series of actions in order to obtain a reward. In the third simulation, the AI Engine managed to learn the desired action without going through a number of episodes. All of these simulations have been proven successful. Other kinds of artificial intelligence and data mining techniques could be implemented and compared in the future.