Scouting Robot with Search-Space Reducing Hybrid Networks for Unknown Environment Path Planning

Abstract

Autonomous vehicle navigation is becoming an important problem as fully self-driving cars are becoming a possibility in the next few decades and space exploration is gaining more momentum. One of the most important aspects of autonomous vehicle navigation is a solid path planning algorithm. Most path planning algorithms can be categorized into 2 groups, classical and machine learning algorithms. Classical algorithms can find the shortest path and usually have a 100% accuracy, but it can't function when its environment isn’t 100% mapped out. On the other hand, ML algorithms can operate on partial maps or no maps at all, but they have poor success rate and produce long paths. However, a hybrid approach to a path planning algorithm could eliminate the downside of both categories of algorithms. A robot with a hybrid algorithm could have 100% accuracy while maintaining a near optimum route without a map of the environment. This paper proposes a search space reduction hybrid network (SRHN) path planning algorithm that not only combines the advantages of classical methods and machine learning methods, but also reduces the search space and memory usage. SRHN works by dividing up the distance between a start point and an endpoint with landmarks and paths from its current landmark to the next. While calculating an approximate optimal path, it significantly reduces search space. To test the result of SRHN, experimentation was conducted in the real world. Excellent results have been achieved in the real world 2D tests that were conducted.