Robots Can Now Navigate Without Needing a Map Using AI
Sat, April 10, 2021

Robots Can Now Navigate Without Needing a Map Using AI

Facebook AI's team developed an algorithm that lets a robot find its way in an unfamiliar environment without using a map / Credits: Zapp2Photo via Shutterstock

 

Robots can be seen anywhere, helping industries automate tasks and increase their sales and revenue. They can do anything that they are programmed to do with the help of artificial intelligence and other technologies. Through sensors, they can navigate anywhere they go as well as keep their balance and avoid obstacles. However, when they are placed in an unfamiliar street and left to find their way, they would definitely fail because they are not programmed to have their own sense of direction. 

To address this problem, a team at Facebook AI developed a reinforcement learning algorithm that lets a robot find its way in an unfamiliar environment without using a map. According to MIT Technology Review, an innovative, digitally oriented global media company that aims to bring about better-informed and more conscious decisions about technology through authoritative, influential, and trustworthy journalism, the algorithm helps a robot find the shortest route in unfamiliar environments.

With the help of a depth-sensing camera, GPS, and compass data, robots can navigate with no wrong turns, no backtracking, and no exploration. Compared to previous systems that reached a 92% success rate on these tasks, this new system can reach their goal 99.9% of the time. The researchers trained the algorithm, which is called decentralized distributed proximal policy optimization or DD-PPO, to achieve nearly 100% success in a variety of virtual environments, such as houses and office buildings.

According to the Facebook AI’s page, the developers are hoping to create systems that accomplish point-goal navigation with only camera input. This will help them build agents that work in common settings, such as inside office buildings or laboratories, where these additional data points aren’t available. Also, they trained and evaluated DD-PPO using a modular framework with a highly performant and stable simulator called AI Habitat platform. This made it an ideal framework for simulating billions of steps of experience.