Energy-efficient Q-learning for Collision Avoidance of Autonomous Robots
MetadataShow full item record
Recently, many companies have been studying intelligent cars, and improvements in sensor technology and computing are required. The intelligent cars use GPS to know where they are. The cars use sensors to detect objects in all directions, including people, vehicles and animals. Based on what the sensors detect, the software processes the information to help the cars predict what all the objects around the cars might do. Based on the prediction, the cars choose a safe speed and trajectory for themselves. Therefore, the current research on intelligent cars is mainly focused on collision avoidance because of safety. Every year, there are over two million traffic-related deaths worldwide. This number could be dramatically reduced, especially since 94% of accidents in the U.S. involve human error. In this work, we show the design, development, and testing of an autonomous ground vehicle for testing energy-efficient of Q-learning in robotics. The vehicle platform is based on the Boe-Bot chassis, which is an education shield for Arduino. The Boe-Bot is a common research testbed that encourages the use of low-cost hardware and open-source software. The research platform uses Raspberry pi 2 Model B as its on-board computer for handling Q-Table collision avoidance like processing sensor data and computing deciding actions. The robot is then tested by using a modified Q-Learning algorithm to avoid collisions. We design hardware-friendly software, and the robot can use the same Q-Table as the software. The robot runs in several maps to find relationships between reward function and time and energy consumption. One target of this work is to find a relationship between reward function and time and energy consumption. This study will help us to understand feasibility of saving time and energy for intelligent cars. The aim of this master’s thesis is to investigate energy-efficient Q-learning collision avoidance in robotics. Unlike, the existing research, we consider saving energy for collision avoidance. Furthermore, we could use same the Q-Table on the actual robot because we designed the software as hardware-friendly.
Ahn, Seungjai (2017). Energy-efficient Q-learning for Collision Avoidance of Autonomous Robots. Master's thesis, Texas A & M University. Available electronically from