Long Range Neural Navigation Policies for the Real World

Marek Fiser
Tsang-Wei Lee
Alexander Toshev
Google Scholar


Learned Neural Network based policies have shown promising results for robot navigation. However, most of these approaches fall short of being used on a real robot -- they require extensive training in environments, most of which do not simulate the visuals and the dynamics of the real world well enough that the resulting policies can be easily deployed. We present a novel Neural Net based policy which allows for easy deployment on a real robot. It consists of two sub policies -- a high level policy which can understand real images and perform long range planning expressed in high level commands; a low level policy that can translate the long range plan into low level commands on a specific platform in a safe and robust manner. For every new deployment, these policies can be successfully trained on two different types of data -- an easily obtainable scan of the deployment world modeling its visuals and layout; a generic synthetic environment modeling the robot physics. We detail the design of such environments and how one can use them for training a final navigation policy. We demonstrate a deployment of the model in a large office building and test it extensively, achieving $0.80$ success rate over long navigation runs and outperforming SLAM-based models in the same settings.

Research Areas