Determining grasping parameters for grasping of an object by a robot grasping end effector

A method, comprising:

• identifying image data for an electronic image of an object, the image data comprising one or more channels;
• identifying a plurality of labeled valid grasps for the object in the electronic image, each of the labeled valid grasps defining grasping parameters for a respective one of the grasps for the object;
• selecting, for a cell of the image data, only a single grasp of the labeled valid grasps for the object, the single grasp being one of multiple of the labeled valid grasps contained in the cell;
• generating training example output that comprises the grasping parameters for the selected single grasp and that omits the grasping parameters for all other of the multiple of the labeled valid grasps contained in the cell;
• wherein the grasping parameters for the selected single grasp define at least a two-dimensional grasping coordinate defining a position of a robot grasping end effector for the grasp of the object and an orientation parameter defining an orientation of the robot grasping end effector for the grasp of the object;
• providing, as a training example to a convolutional neural network, the image data as input of the training example and the training example output as output of the training example;
• and
• training the convolutional neural network based on the training example;
• after training, generating additional grasping parameters for an additional object in a single regression through the convolutional neural network, wherein generating the additional grasping parameters in the single regression comprises:
• applying additional image data, for the additional object, as input to the convolutional neural network, and
• generating the additional grasping parameters as output of the single regression;
• and
• positioning an end effector of a robot relative to the object based on the additional grasping parameters generated in the single regression through the convolutional neural network.