Author(s)

Henry Zhao

Corresponding Author

Henry Zhao

ABSTRACT

The problem of path planning through rough and uneven terrain has been studied extensively in recent decades. However, many such methods are designed to optimize path length, and are not fit for potentially lower-cost legged robots that may drift or suffer from external influence. This paper proposes a method for inferring the presence and heights of objects unseen by camera angles as well as a method to navigate through rough terrain while minimizing risk taken by a robot. The proposed method utilizes and combines multiple camera viewpoints to fill in areas blocked by an object, while allowing for additional viewpoints to be appended. The method also marks out areas that have not yet been seen, allowing a robot to explore said areas, should no other paths be apparent. The path planning section of this method takes into account a variety of costs, notably the total elevation gained and lost, as well as the distance from the path to the closest obstacle. Thus, it prioritizes flat terrain over rough as well as attempting to stay in the middle of two objects, greatly reducing the risk of collision or falling. The proposed methods have been tested both in simulated environments as well as integrated into and tested on a small quadruped robot. The algorithm results in a lowered average distance to obstacles, as well as a lowered total elevation gained or lost, in return for a longer path length. Even with a 5% lateral drift on the robot, this algorithm successfully navigates around obstacles without contact.

KEYWORDS

Navigation, Path planning, A-star, Rough terrain, Quadruped robot