Domain Adaptation Using Adversarial Learning for Autonomous Navigation
Autonomous navigation has become an increasingly popular machine learning application. Recent advances in deep learning have also brought huge improvements to autonomous navigation. However, prior outdoor autonomous navigation methods depended on various expensive sensors or expensive and sometimes erroneously labeled real data. In this paper, we propose an autonomous navigation method that does not require expensive labeled real images and uses only a relatively inexpensive monocular camera. Our proposed method is based on (1) domain adaptation with an adversarial learning framework and (2) exploiting synthetic data from a simulator. To the best of the authors' knowledge, this is the first work to apply domain adaptation with adversarial networks to autonomous navigation. We present empirical results on navigation in outdoor courses using an unmanned aerial vehicle. The performance of our method is comparable to that of a supervised model with labeled real data, although our method does not require any label information for the real data. Our proposal includes a theoretical analysis that supports the applicability of our approach.
READ FULL TEXT