Navigating cities has always been a big challenge for drivers, especially when there is no map on hand. Somewhere down the road, you have to use a GPS or map to figure out how to reach a certain place of interest, even if you have been a longtime inhabitant of the city. Today, Artificial intelligence is making inroads into a variety of areas, sometimes even exceeding human capabilities in carrying out tasks and solving problems. Chief among them is the ability to navigate cities.
Google DeepMind AI, which is famously known for its AI programs such as AlphaGo, has been playing a crucial role in teaching AI how to steer through cities without using any maps. All the AI needs to work out its location is a collection of images captured via Google Street View. As AI is currently operating only as a virtual program, it is carrying out this learning activity by navigating virtual worlds gathered using Google Maps data. This is slowly paying off as they have started gaining and improving their navigation skills on an ongoing basis.
An info video released by Google DeepMind demonstrates how the AI learns and finds out its exact location based on visual inputs from Street View images without using maps or having any previous knowledge about the neighborhood. In the near future, AI programs will be able to gain the knowledge required to travel from one place to another using only this data. Above all, the AI will be able to create its own map of different locations without even going through an actual map.
Google’s DeepMind Technologies Inc says that it offers an end-to-end deep reinforcement learning approach to help in the whole navigation process. The AI has already been tested in popular localities across New York, like Lower Manhattan, Harlem, Midtown and Greenwich Village, in addition to places like Paris Rive Gauche and London Central. DeepMind’s researchers also claim that their AI could help in gaining a precise and deep understanding of how human beings learn about their surroundings.
The AI has to go through training whenever it goes through a new city and this is why it is taking time for real-world application. Once the AI is tested and deployed for commercial use, autonomous cars can learn how to navigate in cities even without using any GPS assistance. One of the main concerns of self-driving cars today is to continue navigation even when there is no cell service. This is where the AI’s ability to get familiar with city lanes can come in handy.
Some Final Words
Who knows, AI programs may develop robot bodies in the future. We may soon see autonomous cars and drones weaving in and out of city traffic better than professional drivers can do. Also, it remains to be seen whether there are going to be risks associated with deploying such a powerful technology on self-driving, autonomous vehicles. For sure, the success of such an AI system will depend a lot on how well it is tested before being implemented in our busy cities.