The challenge of robot navigation in uneven outdoor environments is demanding but important application for outdoor mobile robots. Classical geometric approaches have been employed to solve this problem but with limited success. When the terrain becomes exceedingly irregular or if there are tall grasses present, classical methods do not perform sufficiently well. Therefore, in this project we will use data from expert human demonstrations with inverse reinforcement learning methods to train a robot to navigate in uneven terrain with obstacles such as tall grass. The objective is to train a robot agent that can overcome the limitations of classical navigation methods by utilising learning methods. The robot agent will be trained in simulation using the GAIL algorithm and the resulting agent will be tested in simulation and if time allows also in the real world.
