Soon the virtual assistants will be able to distinguish between the different rooms in your house and even help you find missing items such as keys and books. This and more will all be possible thanks to Facebook’s aim to create more realistic AR and VR. The open-source simulation and dataset developed by the social media giant will eventually allow virtual assistants to learn about your physical surroundings.
Researchers at Facebook created a new open platform for embodied AI research titled Al Habitat. Facebook Reality Labs (previously known as Oculus Research) also released a dataset of photorealistic sample spaces called Replica. Both the methods are now available for the researchers on GitHub and will allow them to train AI bots to replicate in a physical environment.
On the Replica data set, the researchers can see 18 different sample spaces including a living room, a two-story house, and a conference room. Through these sample 3D simulated areas, the researchers can train the AI bots to respond to common commands like ‘bring my keys from the desk in the living room’ and ultimately adopt the same in real-life situations.
The simulation area of Replica dataset mimics actual houses and captures all the details one might find in a real house including decorative pieces. The 3D simulations are realistic looking and manage to replicate even the flooring and textures in sharp detail – an essential protocol to train bots and enable them to work in our places.
Several researchers are already testing the features of Replica and Al Habitat while Facebook AI recently hosted an autonomous navigation challenge on the platform. Facebook will be announcing the winning research team of the challenge at the Conference on Computer Vision and Pattern Recognition (CVPR) scheduled on Sunday.
Read next: Facebook Successfully Gained Support of High-Profile Tech Giants like Mastercard, PayPal and Uber for its Blockchain Project
Featured photo: Bloomberg via Getty Images
Researchers at Facebook created a new open platform for embodied AI research titled Al Habitat. Facebook Reality Labs (previously known as Oculus Research) also released a dataset of photorealistic sample spaces called Replica. Both the methods are now available for the researchers on GitHub and will allow them to train AI bots to replicate in a physical environment.
On the Replica data set, the researchers can see 18 different sample spaces including a living room, a two-story house, and a conference room. Through these sample 3D simulated areas, the researchers can train the AI bots to respond to common commands like ‘bring my keys from the desk in the living room’ and ultimately adopt the same in real-life situations.
The simulation area of Replica dataset mimics actual houses and captures all the details one might find in a real house including decorative pieces. The 3D simulations are realistic looking and manage to replicate even the flooring and textures in sharp detail – an essential protocol to train bots and enable them to work in our places.
Several researchers are already testing the features of Replica and Al Habitat while Facebook AI recently hosted an autonomous navigation challenge on the platform. Facebook will be announcing the winning research team of the challenge at the Conference on Computer Vision and Pattern Recognition (CVPR) scheduled on Sunday.
Read next: Facebook Successfully Gained Support of High-Profile Tech Giants like Mastercard, PayPal and Uber for its Blockchain Project
Featured photo: Bloomberg via Getty Images
