The Robot Report

“This research may open up new possibilities in manipulation with a simple gripper,” said study Wenxuan Zhou at Carnegie Mellon University. “Potential applications include warehouse robots or housekeeping robots that help people to organize their home.”

Inside the system

The team used reinforcement learning to train a neural network and tested the system in the MuJoCo physics simulator. The simulated robot attempted to grasp objects and was rewarded when successful.

The researchers then zero-shot transferred the technology to a physical robotics system, which consisted of a Frank Emika robotic arm and a parallel gripper. The robot grasped items confined within an open bin that were initially oriented in ways the robot couldn’t handle. It then needed to figure out how to use its environment to create success picks, for example, by pushing an item against the wall of the bin to reorient said object.

Here’s an example of how the gripper could use its environment to help grasp an object. The gripper first moves close to the object and makes contact on the side of the object with the top finger. It then pushes the object against the wall to rotate it. During this stage, the gripper

With the exploitation of external resources such as contact surfaces or gravity, even simple grippers can perform skillful maneuvers that are typically studied with a multifingered dexterous hand.

| Carnegie Mellon University

usually maintains a fixed or rolling contact with the object, but sliding also occurs. The object might have sliding or sticking contacts with the wall and the ground. After the gripper has rotated a bit further and the bottom fingertip is below the object, the gripper will let the object drop on the bottom finger. After that, the gripper will try to match the desired pose more precisely. At this point, the policy has executed the grasp successfully and it is ready to close the gripper.

According to the researchers, one of the key decisions in this strategy is to use the top finger to rotate the object instead of the bottom finger. One might suppose an alternative approach which is to use the bottom finger to scoop the object against the wall and then directly roll the finger underneath the object to reach the grasp. However, this strategy is not physically feasible on the parallel gripper due to the limited degree of freedom of the finger.

Another successful strategy is to flip the object to stand on its side and then move to the grasp. This strategy relies on the fact that the object remains stable after the rotation.

The researchers tested the system on items that varied in weight, shape and slickness, including cardboard boxes, plastic bottles and a container

Advanced Products for Robotics and Automation

CGI Motion standard products are designed with customization in mind. Our team of experts will work with you on selecting the optimal base product and craft a unique solution to help di erentiate your product or application. So when you think customization, think standard CGI assemblies.

Connect with us today to explore what CGI Motion can do for you.