Researchers from Allen Institute for Artificial Intelligence, Seattle have developed a new open source virtual reality training ground called AI2-THOR to boost efficiency of the robots.
Advanced robots are getting better with performing daily tasks. However, they are not capable of performing complex tasks, which are not programmed as a part of their system. Open source developed by Allen Institute for Artificial Intelligence aims to provide virtual reality training ground called AI2-THOR. Most of the robots are programmed with commercial machine learning algorithms to function according to practical requirement. Data sets made up of video and still images help to enhance the performance of machine learning.
‘The House Of inteRactions’ (THOR) is AI training ground, which is created using graphical engine unity. The project is in development since 2016. The software was provided in 120 different scenarios based around kitchens, living room, bedroom, and bathroom settings. Location specific features such as open microwave in kitchen are taken into consideration. The details such as empty and full bathtubs and sliceable apples are also noted.
AI2-THOR team working on expanding the applications in near future by adding objects with non-rigid physics, letting robots get valuable practice at making beds or moving items of clothing. The open-source software is available for user with simple download. It customizes the various scenes according to the specifications. Such initiatives are expected to improve performance of robots to get even smarter.