LiDAR Scan Images for Mobile Robot Motion Planners

During autonomous navigation, mobile robots are required to avoid obstacles ahead in some cases. We assume a mobile robot equipped with a
$2D$
LiDAR. For the robot, we have proposed motion planners based on deep neural networks, such as multilayer perceptrons, MLPs. The networks were trained through imitation learning. However, the robot based on these motion planners using range data inputs sometimes failed to avoid obstacles in unknown environments. This is because that the robot planned avoidance motions without considering the spatial configuration. In this paper, therefore, we focus on the obstacle avoidance capability of the robot, and propose to use LiDAR scan images as inputs. The LiDAR scan images are generated from the range data, and used as inputs to four types of motion planners based on MLP and convolutional neural network, CNN. In the navigation experiments, the effectiveness of the motion planners for improving the obstacle avoidance capability is discussed. Moreover, we examine the applicability of the motion planners to the actual mobile robot for autonomous navigation in a real environment.