Various aspects of techniques, systems, and use cases for selecting grasping configurations for a robot are disclosed. Geometric primitives are generated to model the robot for grasping and manipulation by the robot. The geometric primitives are combined using various functions to determine which configuration to use. The instantaneous configuration is determined, as well as the forward kinematics and links to determine active geometric primitives of the gripper. The active geometric primitives are used to approximate an x, y, and z coordinate of each point of the primitives, a distance between the point and a grasping target, and an associated surface link. The configurations are ranked based on grasping metrics and one of the configurations selected to use accordingly.

Deep machine learning methods and apparatus related to manipulation of an object by an end effector of a robot. Some implementations relate to training a deep neural network to predict a measure that candidate motion data for an end effector of a robot will result in a successful grasp of one or more objects by the end effector. Some implementations are directed to utilization of the trained deep neural network to servo a grasping end effector of a robot to achieve a successful grasp of an object by the grasping end effector. For example, the trained deep neural network may be utilized in the iterative updating of motion control commands for one or more actuators of a robot that control the pose of a grasping end effector of the robot, and to determine when to generate grasping control commands to effectuate an attempted grasp by the grasping end effector.

Technologies are generally described for grasp point selection for robotic grippers through machine vision and ultrasound beam based deformation. The grasp point selection may aim to increase a probability that the grasp points on an object behave in a substantially similar way when a robotic gripper executes a corresponding grasp on the object. According to some examples, an outline of an object may be extracted from a three-dimensional (3D) image of the object and a set of points may be selected as candidate grasp points from the outline based on the candidate grasp points' potential to achieve force closure. One or more ultrasound transducers may be used to exert a local force on the candidate grasp points through an ultrasound beam and resulting local deformations may be recorded. Final grasp points may be selected based on having similar response to the force applied by the ultrasound transducers.