A Robot Obstacle Avoidance Approach with LiDAR and RGB Camera Data Combined

A robot obstacle avoidance approach is proposed in this paper, which combines 2D LiDAR data with images from the RGB camera attached to the mobile robot. With accurate information of distance to obstacles provided by the LiDAR, conventional obstacle avoidance methods have good performance navigating in simple environments, which means that the ground is level and no low reflectivity object is blocking the robot's way. However, lacking global awareness of its surrounding environment, the robot may not perform well in complex environments. In our approach to obstacle avoidance, the robot uses the camera as a second data source to get more information about its surroundings. Based on the end-to-end deep learning algorithm proposed in this paper, images from the camera are classified into two categories, telling the possibility of whether there are obstacles ahead. A method integrating the possibility to the navigation process is also proposed, making the robot more reliable and less likely to run into an obstacle while navigating in a complex environment. Finally, The Timed Elastic Band (TEB) local path planning method is used to produce the local path and generate the velocity command for the robot. An experiment is conducted to verify the overall performance by deploying the approach to an embedded system on a mobile robot. By evaluating the experimental result, we found the method effective and reliable for obstacle avoidance.