Monday 26 October 2015

How Robots Can Learn New Tasks by Observing


Robot Training Academies – Aid Industrial Robots to Perform Difficult Chores

Robot `training academies’ could aid industrial robots to learn performing difficult chores by first, observing how humans tend to do it. It could probably take weeks in reprogramming an industrial robot to accomplish a complicated new job which would make retooling modern manufacturing route expensive and slow.

The development could speed up if robots were capable of learning how to perform a new task by observing others first. This is the idea behind the project which is in process at the University of Maryland, wherein researchers are training robots to be attentive students.

Yezhou Yang, a graduate student in the Autonomy, Robotics and Cognition Lab at the University of Maryland states that it is a robot training academy. They have asked an expert to show the robot a task and let the robot figure out most parts of sequences of things which need to be done and then fine tune things to make it work. The researchers at a recent conference in St. Louis had demonstrated a cocktail making robot which tends to use the approaches they work on.

Two Armed Industrial Machine Robot – Rethink Robotics

A two armed industrial machine robot, made by a Boston based company known as Rethink robotics, had watched a person mix drink by pouring liquid from many bottles into a jug and then copied these actions, clutching bottles in the proper order prior to pouring the precise quantities into a jug.

Yang performed the task with Yiannis Aloimonos and Cornelia Fermuller, who were the two professors of computer science at the University of Maryland. The approach is said to comprise of training a computer system to link specific actions of the robot with video footage portraying the performance of people with various tasks.

 For instance, a recent paper from the group portrayed that a robot can learn how to pick various objects by utilising two differentsystems on observing thousands of instructional YouTube videos wherein one system studies to recognize various object while the other identifies various types of grip.

System of Learning Adopted – Advanced Artificial Neural Network

Observing many YouTube video could be time consuming, however the learning technique seems to be much more efficient than programing a robot in managing various different matters and enabling the robot to handle new object.

The system of learning adopted in gripping work tends to include advanced artificial neural network which has envisaged great progress recently and are now being utilised in several areas of robotics. Researchers are coordinating with many manufacturing companies comprising of electronics business as well as car makers in adapting the technology to be used in factories.

 The companies are looking out for ways to quicken the process wherein engineers reprogram the machines. Yang is of the opinion that at several companies it tends to make a month and a half or more to reprogram a robot. So what would be the current AI potentials that can be used in shortening this span to half?

The project tends to reflect two styles in robotics, while one is to find new approaches in learning and the other is robots working in closeness with people. The Maryland researchers like the other group wishes to connect actions to language in improving the ability of robots to analyse spoken or written instructions.

No comments:

Post a Comment

Note: only a member of this blog may post a comment.