ABSTRACT
Navigation is one of the fundamental requirements of a driverless car. This chapter gives a brief introduction to localization and mapping, discusses various use cases and explains different VSLAM approaches. Additionally the HD maps discussed earlier can provide absolute localization for pre-mapped areas because they are created by using precise GPS initialization. Active localization is a key domain where, in addition to the conventional mapping step, the algorithm decides the actions of the ego vehicle or robots, thus planning to reduce the uncertainty. The rest of the chapter demonstrates camera-based VSLAM and range sensor-based SLAM. Using deep learning modules for perception and deep RL for path planning algorithms to achieve active localization in an OpenAI-simulated environment were discussed. The authors evaluated the performance of agents in a simulated 2D environment.
