Egocentric perception is how people see the world around them, a first-person perspective, which is difficult for AI to comprehend since it currently must rely on learning from a third-person view. Facebook AI is aiming to solve this problem to take artificial intelligence to the next level and “unlock a new era of immersive experiences.” The company gathered a group of researchers from 13 universities and laboratories across nine countries to collect large amounts of data for the project. The global team managed to collect 2,200 hours of first-person video from over 700 participants living their daily lives, all of which will be used to teach AI. A scientist on the project says that the AI field needs to have a new standard so artificial intelligence can learn from a first-person perspective and understand real-time motion. The company brings up the example of a roller coaster. From a third-person perspective, AI can understand what it’s looking at, but strapped into the seat of a roller coaster, it doesn’t know what’s going on. The team has five benchmarks it wants to achieve, including having the AI understand social interactions, manipulate objects, and plan for the future like a person can. The data Facebook has collected will be publicly available for researchers in November. Then, in early 2022, there will be a research challenge for AI experts worldwide to teach other machines artificial intelligence egocentric perception.
