RAM based neural-network controlled vehicle: path-tracking & collision avoidance

Various AI algorithms exist that can serve vehicle autonomy domain within limits and present different critical levels of time-space complexity in real time scenarios. This paper evaluates the performance of WISARD-net for autonomous vehicle vision based maneuver control system for road-following and obstacle avoidance tasks. A novel Grid-WISARD algorithm is proposed and examined for obstacle avoidance maneuver. The vehicle is trained for a small stretch of artificial road with a 3 to 4 sets of road images and achieved successful road-tracking, obstacle avoidance and driving on the left-side of the road. Whereas ANN models normally require long training, the proposed model was able to demonstrate successful maneuvers with comparatively considerably few training samples and training duration.