Helping Robots Get a Grasp

Jan-29-19 A new probabilistic grasp planner able to plan grasps in real time could allow robots to complete a wide array of tasks with different types of objects.

Grasping different objects safely and effectively requires an understanding of how to approach the object—such as a full-handed approach or a more delicate, multi-finger approach. To allow robots more dexterity across a wider range of objects, the team from the University of Utah developed a grasp planning system that enables the robot to learn the best grasping method as it practices different grasps on various objects. That collected data is then used to predict which grasp is most effective, and can also be used to enable the robot to choose a grasp for an unfamiliar object or one that is only partially visible.

More Info about this Invention:

[TECHXPLORE.COM]
[ROBOT-LEARNING.CS.UTAH.EDU]

---