Imagination-augmented Navigation Based on 2D Laser Sensor Observations

Autonomous navigation of mobile robots is an essential task for various industries. Sensor data is crucial to ensure safe and reliable navigation. However, sensor observations are often limited by different factors. Imagination can assist to enhance the view and aid navigation in dangerous or unknown situations where only limited sensor observation is available. In this paper, we propose an imagination-enhanced navigation based on 2D semantic laser scan data. The system contains an imagination module, which can predict the entire occupied area of the object. The imagination module is trained in a supervised manner using a collected training dataset from a 2D simulator. Four different imagination models are trained, and the imagination results are evaluated. Subsequently, the imagination results are integrated into the local and global cost map to benefit the navigation procedure. The approach is validated on three different test maps, with seven different paths for each map. The quality and numeric results showed that the agent with the imagination module could generate more reliable paths without passing beneath the object, with the cost of a longer path and slower velocity.