Over the last decade deep learning (DL) has shown to be a powerful paradigm for the development of robots capable of autonomously navigating, interacting and making decisions in complex environments, thanks to the concurrent advances in high-performance computing. 

However, DL-based autonomous navigation tends to be difficult to interpret, especially in so called end-to-end methods, where the visual input is used to directly infer a control action. This often makes experimental validation in real-world applications arduous. Similarly, it is hard to ensure robot safety, especially during testing stages. These issues have limited the application of DL to outdoor unstructured environments, where mobile robots will have a strong impact, as in, e.g., urban and non-urban search and rescue, planetary exploration, precision farming, to name a few. 

The goal of the workshop is to gather the current results lying at the intersection of deep learning and field robotics in outdoor unstructured environments, embracing the broad realm of unmanned aerial, ground, and marine vehicles. The ultimate goal is to foster interaction between the two related communities, by sharing successful stories and promising directions to solve the current open challenges in deploying DL-based solutions in the wild. Special attention will be given to simulators and benchmarking tools as crucial elements in the development and testing of novel methods in hazardous environments where the robot may be irreparably damaged.