Facebook Creates Method May Allow AI Robots To Navigate Without Map

Facebook has recently created an algorithm that enhances an AI agent’s ability to navigate an environment, letting the agent determine the shortest route through new environments without access to a map. While mobile robots typically have a map programmed into them, the new algorithm that Facebook designed could enable the creation of robots that can navigate environments without the need for maps.

According to a post created by Facebook researchers, a major challenge for robot navigation is endowing AI systems with the ability to navigate through novel environments and reaching programmed destinations without a map. In order to tackle this challenge, Facebook created a reinforcement learning algorithm distributed across multiple learners. The algorithm was called decentralized distributed proximal policy optimization (DD-PPO). DD-PPO was given only compass data, GPS data, and access to an RGB-D camera, but was reportedly able to navigate a virtual environment and get to a goal without any map data.

According to the researchers, the agents were trained in virtual environments like office buildings and houses. The resulting algorithm was capable of navigating a simulated indoor environment, choosing the correct fork in a path, and quickly recovering from errors if it chose the wrong path. The virtual environment results were promising, and it’s important that the agents are able to reliably navigate these common environments, as in the real world an agent could damage itself or its surroundings if it fails.

The Facebook research team explained that the focus of their project was assistive robots, as proper, reliable navigation for assistive robots and AI agents is essential. The research team explained that navigation is essential for a wide variety of assistive AI systems, from robots that perform tasks around the house to AI-driven devices that help people with visual impairments. The research team also argued that AI creators should move away from map usage in general, as maps are often outdated as soon as they are drawn, and in the real world environments, they are constantly changing and evolving.

As TechExplore reported, the Facebook research team made use of the open-source AI Habitat platform, which enabled them to train embodied agents in photorealistic 3-D environments in a timely fashion. Haven provided access to a set of simulated environments, and these environments are realistic enough that the data generated by the AI model can be applied ot real-world cases. Douglas Heaven in MIT Technology Review explained the intensity of the model’s training:

“Facebook trained bots for three days inside AI Habitat, a photorealistic virtual mock-up of the interior of a building, with rooms and corridors and furniture. In that time they took 2.5 billion steps—the equivalent of 80 years of human experience.”

Due to the sheer complexity of the training task, the researchers reportedly culled the weak learners as the training continued in order to speed up training time. The research team hopes to take their current model further and go on to create algorithms that can navigate complex environments using only camera data, dropping the GPS data and compass. The reason for this is that GPS data and compass data can often be thrown off indoors, be too noisy, or just be unavailable.

While the technology has yet to be tested outdoors and has trouble navigating over long-distances, the development of the algorithm is an important step in the development of the next generation of robots, especially delivery drones, and robots that operate in offices or homes.