After more than three decades of research, UC Berkeley engineers Ken Goldberg and Jeff Mahler introduced the design of a novel “ambidextrous” robot in a study Jan. 16.
With the capability of revolutionizing factory robots, this new invention is equipped with a number of different grippers that can handle a variety of different objects. The study was published in the journal Science Robotics.
Mahler, a campus professor of industrial engineering and operations research as well as electrical engineering and computer sciences, said in an email that he hopes the robot will have a “tangible impact in the world,” adding that it will likely be applied to logistical work, potentially doing tasks such as fulfilling orders in e-commerce.
The idea to create this robot was sparked in 2017 after Goldberg and Mahler, who is a UC Berkeley postdoctoral researcher, published two research papers focused on the gripping mechanisms of two different types of robots: a parallel-jaw gripper and a vacuum-based suction cup. Through this project, the engineers realized that the strengths and weaknesses of each type of robot gripper complemented one another.
“The natural idea was to combine the two modalities to see if we could enable the robot to grasp a larger diversity of objects,” Mahler said in the email. “The challenge we tackled in the paper was to figure out how to get the robot to decide which gripper to use for a given object.”
According to Mahler, the initial step in developing the robot was creating an experimental benchmark system that could help the researchers measure progress.
The researchers also had to re-examine data with their previous methods involving two individual grippers: Dex-Net 2.0 and 3.0. These grippers were designed to plan grasps from synthetic data that often came from 3D object models in conjunction with analytic models of geometry and contact physics.
Though Mahler said that although he feels proud of the progress made, he still believes there is more work to do.
Mahler added that there are a lot of problems the researchers have to consider, especially with regard to the robot’s ability to grasp an object. More specifically, the robot needs to be able to identify objects that must be placed in a particular orientation — for instance, getting the robot to know to place a cup on its base, according to Mahler.
The researchers have also encountered limitations, including certain transparent objects that may not be able to be handled reliably. The robot is also very slow in learning from its past experiences, Mahler added.
“It is nice to have all of the experiments done! But the project is not done, by any means,” Mahler said in the email. “I hope that it will make a tangible impact in the world. … I also hope it helps inspire future generations of roboticists.”