Abstract
Finding a path free from obstacles that poses minimal risk is critical forsafe navigation. People who are sighted and people who are visually impairedrequire navigation safety while walking on a sidewalk. In this research wedeveloped an assistive navigation on a sidewalk by integrating sensory inputsusing reinforcement learning. We trained a Sidewalk Obstacle Avoidance Agent(SOAA) through reinforcement learning in a simulated robotic environment. ASidewalk Obstacle Conversational Agent (SOCA) is built by training a naturallanguage conversation agent with real conversation data. The SOAA along withSOCA was integrated in a prototype device called augmented guide (AG).Empirical analysis showed that this prototype improved the obstacle avoidanceexperience about 5% from a base case of 81.29%