Coding Random forest algorithm for Autonomous vehicle hazard detection (Msc Project)

Modified random forest Algorithm for extreme road incident detection in Autonomous vehicles

Statement of Problem
Driverless cars use a number of analytical Algorithms and machine learning algorithms for object detection and tracking, sensor data-based mapping and planning/decision making. A key technology in ensuring the autonomous vehicle driving safety is the Driver Decision Making Mechanism (DDM). The is the algorithm of interest for this project.

Current fully autonomous systems utilize perception localization, mission planning, motion planning and trajectory tracking to navigate road networks. However in a dynamic urban setting, the response of an AV to any extremely unpredictable and rare event has not been specifically addressed. Unpredictable urban driving scenarios such as breakout of/approaching bush wild fires, approaching flash floods, road accident inferno, sudden road obstruction e.t.c requiring the vehicle to reverse from the potentially dangerous event and re-route are rare but probable in occurrence. When these events occur, existing localization algorithms and perception systems may leave a fully autonomous vehicle stuck without the ability to differentiate between a temporary traffic obstruction from a permanent route closing obstruction. Without a dedicated extreme event (EE) detection subroutine in the overall planning algorithm, the AV may wait endlessly at a spot after initial scenario identification of a sudden obstruction to forward motion.

This project is to design the subroutine which enables the AV identify the scenario and differentiate it from a traffic gridlock with a high degree of accuracy within a safe time frame.