End-to-end navigation for robots using learning from demonstration with 2D-laser range data

One of the continuing research topics is the application of learning from demonstration to motion planning. The purpose of this research is to offer a model that is capable of learning the complex mapping from raw 2D-laser range discoveries and a target location to the necessary steering signals for the robot. This work introduces the first approach that, to the best of our knowledge, trains a target-oriented end to-end navigation model for a robotic platform. The expert demonstrations that are generated in simulation with an existing motion planner serve as the foundation for the supervised model training sessions. We provide evidence that the navigation model that has been taught may be directly transferred to environments that have not been encountered before, both in the virtual world and, in the real world. It can safely lead the robot through areas that are packed with obstacles to reach the targets that have been set. A comprehensive quantitative and qualitative evaluation of the motion planner is presented, and a comparison is made between it and a grid-based global strategy, which was tested in simulation as well as in trials conducted in the real world.