• <xmp id="om0om">
  • <table id="om0om"><noscript id="om0om"></noscript></table>
  • Han-Pang Chiu, SRI International
    gtc-dc 2019
    We’ll discuss vision-based navigation, which is an important and challenging technology. Traditional approaches rely on an accurate visual feature map that’s built beforehand or constructed using simultaneous localization and mapping algorithms. However, the changing real world requires that the navigation system adapt to changes in appearance caused by weather, time, and illumination variations. Maps need to be continuously updated, which is difficult and time consuming. We’ll describe a new hybrid approach to this problem, in which we utilize both semantic and metric inference. Semantic information is more adaptive to scene changes in comparison to low-level features, and can be matched across time. We’ll explain our development of various AI-enabled methods that incorporate semantic information to improve navigation performance. These are all built into our navigation system, which runs on an NVIDIA Jetson TX2.