An addition to the learning process of a robot, researchers have developed a new algorithm to teach pre-grasping manipulation techniques. Researchers at Karlsruhe Institute of Technology provided successful results after several attempts. Further, the study is published in arXiv states that there were several attempts to impart human functionality in robots. However, most failed due to the lack focus on strategies to manipulate events prior to conducting the desired functions.
According to Lars Berscheid, one of the researchers, grasping is a well-understood concept in modern robotics. However, the challenge lies in making the robot do required tasks before it grasps an object. As a result, makes it very difficult for machines get hold of the objects from a clutter or extremely tight spaces.
Berscheid further stated that their study focuses on more than evolving grasping techniques for the robots. Also, it focuses on implementation of advanced robotics learning algorithms to manipulate pre-grasping operations. This, as a result, will provide more human-like functionality to the robots.
How Researchers Trained the Robot the Trick of Pre-Grasping?
During its learning stage, the robot completed its tasks based on the data fed to it. By processing the acquired data, the robot can decide on the action to perform to complete the task. Researchers trained the robot to pick objects from random bins. Further, they input the data to complete the action through grasper and various imaging techniques. Along with this, researchers also trained the robot to either shift or move the object that does not match with the input data.
Bercheid continues to explain that at first it seems easy to train the robot. However, the process to inculcate this algorithm into an existing machine learning algorithm is tricky. Also, he states that the challenge of training the robot in the first place is quite tricky. This is because training must be automated in order to provide desired result.
As the research is progresses, researchers believe that implementing more pre-operation functions to a robotic system to give it more of a human touch is possible.