Task execution system, task execution method, training apparatus, and training method

#This application claims priority to Japanese Patent Application No. 2017-093222 filed May 9, 2017, the entire contents of which are incorporated herein by reference.

#The disclosure relates to a system that realizes execution of a task, a method for realizing execution of a task, and a training apparatus and a training method for the same.

#Conventionally, it is known that machine learning technology including a neural network is used to control a system to cause the system to execute a predetermined task. For example, JP 2017-20135A discloses that machine learning is applied to the picking of grip-target objects that are piled up in bulk, so that target objects with a high gripping success rate are learnt, and picking is performed. Also, for example, JP 3978098B2 discloses that machine learning (with a rule-based classifier) is applied to classification processing that is performed to determine whether or not samples have a defect, using captured images, and a user sets the configuration of the classifier, in advance of learning.

#FIG. 1 is a diagram illustrating a concept of an overall system that includes a task execution system according to one or more embodiments.

#FIG. 2 is a block diagram illustrating an example of a functional configuration a task execution system according to one or more embodiments.

#FIG. 3 is a diagram illustrating an example of a hardware configuration of a task execution system according to one or more embodiments.

#Hereinafter, one or more embodiments will be described in detail with reference to the drawings. Note that identical elements are denoted by identical reference signs, and redundant description thereof is omitted. Also, the following one or more embodiments are examples that illustrate the present invention, and is not intended to limit the present invention to only the one or more embodiments. Furthermore, the present invention may be variously modified without departing from the spirit thereof.

#FIG. 1 is a diagram showing a concept of an overall system 1 that includes a system 10 (hereinafter also referred to as “task execution system 10”) that realizes execution of a predetermined task according to one or more embodiments. The task execution system 10 is a system that realizes execution of a predetermined task, using a learning module 16. Examples of tasks that are executed may include, but are not limited to, the task of determining the quality of a product and outputs the result of inspection on a display device, the task of outputting an operational instruction to a robot to instruct the robot to grip an object, and so on. Note that, as described below, the learning module 16 includes one unit of a dedicated or general-purpose piece of hardware or software that has the ability to learn through machine learning, or one unit composed of a given combination of such units. That is, the learning module 16 includes a software program that includes at least one of: a learning model that has the ability to learn through machine learning; and a trained model that has acquired a predetermined ability through machine learning.

#The learning module 16 includes a computation apparatus that includes: a storage apparatus in which the software program is stored; and a hardware processor that reads out and executes the software program. “Realizing execution of a predetermined task” refers to a case where an external system 50 is caused to execute a predetermined task, and may also refer to a case where the task execution system 10 itself executes a predetermined task. The task execution system 10 is applicable not only to a case where an actual system is caused to execute a task, but also to a case where a simulator, which is a virtual system, is caused to execute a task. In such a case, targets that are to be controlled are a virtual sensor, a virtual robot, a virtual system, and so on.