A team of roboticists at Carnegie Mellon University has developed a training system that allows a robot to start with limited capabilities, such as performing a specific task such as opening doors or drawers, and improve as it teaches itself how to modify its techniques, while facing never-before-seen challenges.
According to what was reported by the “techxplore” website, the researchers proposed that the only way to train robots is under real-world conditions, and to this end, they developed an adaptive learning approach that allows the robot to learn by starting with a limited knowledge base and adding to it through practical experience.
The research team tested their ideas by creating their own four-wheeled robot with one arm and a hand unit from off-the-shelf components.
The robot’s sole purpose was to approach a door or staircase and then use the clamp to grab the door, turn, push, or whatever else was needed to open the door.
The researchers showed how to manipulate a number of door handles to enter a room or building, and at an outdoor test site, the researchers allowed the robot to attempt to open a door or drawer.
They noted that if the type of handle has already been learned, the robot uses its knowledge to open the door immediately, but if it is unfamiliar, the robot will use what it knows about other door handles to try to gain access.
The researchers found that given enough time the robot usually figures out how to open the door or drawer, showing a 95% success rate.
#robots #open #kinds #doors #Find #details
#Technology