Training robots like we train children

Training robots like we train children

What is the first thing that comes to your mind when you hear the word ‘Robot’? Some Arnold fans might say, Terminators, while a die-hard Star Wars fan would quickly respond with either BB-8 or C-3PO.

via GIPHY

As shown in these movies, robots are gaining a lot of popularity among humans. According to a report by Statista, worldwide shipments of industrial robots came to around 294,000 units in 2016, up from around 159,000 in 2012.

The tremendous growth in sales demonstrates our increasing dependence on the robots.  One of the major concerns with the using robots is that it is difficult to train them. There are still certain tasks easily performed by humans but robots are unable to perform such tasks without extensive programming.

However, University of Connecticut has come up with a unique solution to overcome this problem.

PATENTED TECHNOLOGY

According to a patent application filed by the University of Connecticut, a human trainer demonstrates a task to a robot. The human trainer can demonstrate any particular task such as loading/unloading a dishwasher or placing food inside a microwave. The human trainer demonstrates task in front of a three-dimensional (3D) video camera such as Microsoft Kinect for Windows.

The invention obtains the footage captured by the 3D camera and processes the footage to obtain information related to various parameters such as position, velocity, and acceleration of the trainer’s movements while performing the task. The processed information is then fed to a training unit where the weights for Neural Networks or Gaussian Mixture Model are obtained. These weights are used to model a Dynamic System which will help the robot to mimic trainer’s motion. Using the Neural Networks and Gaussian Mixture Model, a trajectory corresponding to a task is determined.  Further, a motion planning unit is used to transfer these trajectories to the robot. Finally, the robot mimics actions performed by the trainer to perform a task.

ADVANTAGES

This technology can help in bringing the robots into the mainstream. Any user can easily train robots to perform specific tasks. Especially, it can be used by non-technical people to interact more easily with the robots.

WHAT IS YOUR TAKE?

So what do you think of this latest invention by University of Connecticut? Let us know in the comments below.

Patent Information
Publication Number: US20180029226
Patent Title: Skill transfer from a person to a robot
Publication date: 1 Feb 2018
Filing date: 26 Jul 2017
Inventors: Ashwin Dani; Harish Ravichandar
Original Assignee: University of Connecticut

US20180029226