LEARNING OBSTACLE BEHAVIOR FOR IMPROVED THREAT MAPPING DURING NAVIGATION
Johns Hopkins University
A novel obstacle evaluation system for the visually impaired is proposed and evaluated in this thesis. This system assesses the threat in the environment posed by various kinds of stationary and moving obstacles. It evaluates the behaviors of different classes of obstacles and learns features based on their collision avoidance strategies. For the purpose of crowd simulation, Reciprocal Collision Avoidance strategy has been used here. This idea can further be generalized to any multi-agent environment and can serve as the building block for new obstacle avoidance algorithms.
visually impaired, obstacle avoidance, collision avoidance, multi agent, threat map