Berkeley researchers to have developed new robot navigation for self-driving systems

Researchers at the University of California at Berkeley have recently developed a wheeled robot that can travel kilometers over suburban terrain.

Indeed, the robot relies on heuristics picked up from thirty hours of video of previous runs and maps of the terrain in order to have an improved schematic of the way instead of only mapping its environment. The research, called ‘ViKiNG: Vision-Based Kilometer-Scale Navigation with Geographic Hints’, focuses on reinforcement learning, which is a form of deep learning AI where neural networks are trained to advance in stages toward a goal.

The software then uses overhead satellite images of the new terrain or overhead maps, enabling it to go further and faster. This is done through increased training data and a neural network. In order to get results, the program uses the combination of low-level learned control approaches, for moment-by-moment navigation, and higher-level planning.

Hence, it is believed that this is the key toward more complex navigation such as autonomously driven vehicles.