Investigation and Design of Swarm Intelligence Methodologies applying Machine Learning for Terrain Mapping
Abstract
Our goal in this project is to build a system of robots that can autonomously generate a map of an unknown location. We use three ground robots with movement and obstacle avoiding capabilities.
We have used and tested various swarm dispersion and exploration algorithms, both in simulations and in physical robots. These algorithms allow the robots to use simple behaviours to efficiently cover a given environment.
Two types of terrain mapping, namely radar and computer vision systems were evaluated. We concluded that a computer vision system that performs visual SLAM (Simultaneous Localization And Mapping) is preferable to a radar-based system.
We have also implemented a machine learning system that is capable of traversing uneven terrain in simulation by solving the mountain car problem from OpenAI gym.
Implementation
Software
Machine Learning
- OpenAI
- Mountain Car Problem
- Kernel Functions
- Using linear approximators to do nonlinear classification