Using GPS Data to Control an Agent in a Realistic 3D Environment

In this paper we focus on controlling an agent in a virtual environment using only GPS information collected with off-the-shelf devices. GPS data in itself is too noisy to be used as waypoint data, as even minor errors in waypoints may cause agents to wander into illegal locations or get stuck between obstacles. To improve the inaccurate waypoint coordinates, we created a method based on map matching and A* path finding utilizing a navigation mesh. To demonstrate the competence of our method we compare it with two other methods, which are map matching without path finding and a classifier based GPS averaging without a navigation mesh. We found that using a navigation mesh with map matching and A* based path finding, safe routes can always be guaranteed for the agent even though some unnaturalness is sometimes introduced in the paths.