Arithmetic device, control program, machine learner, grasping apparatus, and control method

A non-transitory computer readable medium storing a control program for controlling a motion of a grasping system configured to perform work involving a motion in which a grasping part or a grasped object grasped by the grasping part is brought into contact with a target object to slide the grasped object, the control program causing a computer to:

• acquire, as a state variable indicating a state of the grasping system when the grasping system is performing the work, any one of:
• a position and a speed of the target object;
• a type, a shape, and a state of the target object;
• a relative position and a relative speed of the grasping part or the grasped object with respect to the target object;
• a presence or absence of contact between the grasping part or the grasped object and the target object;
• an operating force of the grasping part or an operating force of each joint part related to a grasping of the grasped object;
• a vibration of a contact part between the grasping part or the grasped object and the target object;
• a sliding sound emitted from the contact part between the grasping part or the grasped object and the target object;
• and a difference between a target posture and an actual posture or an adaptive amount of a control system;
• calculate a target value of each of various actuators related to the work in the grasping system by inputting the state variable to the learned machine learner read from a storage unit storing a learned machine learner that has been learned by receiving a plurality of training data sets composed of a combination of the state variable acquired in advance and correct answer data corresponding to the state variable;
• and
• output the target value of each of the various actuators to the grasping system.

A machine learner configured to determine a target value of each of various actuators of a grasping system configured to perform work involving a motion in which a grasping part or a grasped object grasped by the grasping part is brought into contact with a target object to slide the grasped object, wherein

• the machine learner:
• uses, as a state variable that is acquired in advance and indicates a state of the grasping system when the grasping system is performing the work, any one of:
• a position and a speed of the target object;
• a type, a shape, and a state of the target object;
• a relative position and a relative speed of the grasping part or the grasped object with respect to the target object;
• a presence or absence of contact between the grasping part or the grasped object and the target object;
• an operating force of the grasping part or an operating force of each joint part related to a grasping of the grasped object;
• a vibration of a contact part between the grasping part or the grasped object and the target object;
• a sliding sound emitted from the contact part between the grasping part or the grasped object and the target object;
• and a difference between a target posture and an actual posture or an adaptive amount of a control system;
• and performs learning by receiving a plurality of training data sets composed of a combination of the state variable and a target value of each of the various actuators of the grasping system in the motion of sliding the grasped object which is correct answer data corresponding to the state variable.

A grasping system comprising:

• a grasping part;
• various actuators related to work involving a motion in which the grasping part or a grasped object grasped by the grasping part is brought into contact with a target object to slide the grasped object;
• an acquisition unit configured to acquire, as a state variable indicating a state of the grasping system when the grasping system is performing the work, any one of:
• a position and a speed of the target object;
• a type, a shape, and a state of the target object;
• a relative position and a relative speed of the grasping part or the grasped object with respect to the target object;
• a presence or absence of contact between the grasping part or the grasped object and the target object;
• an operating force of the grasping part or an operating force of each joint part related to a grasping of the grasped object;
• a vibration of a contact part between the grasping part or the grasped object and the target object;
• a sliding sound emitted from the contact part between the grasping part or the grasped object and the target object;
• and a difference between a target posture and an actual posture or an adaptive amount of a control system;
• a storage unit storing a learned machine learner that has been learned by receiving a plurality of training data sets composed of a combination of the state variable acquired in advance and correct answer data corresponding to the state variable;
• an arithmetic unit configured to calculate a target value of each of the various actuators by inputting the state variable to the learned machine learner read from the storage unit;
• and
• an output unit configured to output the target value of each of the various actuators to the grasping system.

A control method for controlling a motion of a grasping system configured to perform work involving a motion in which a grasping part or a grasped object grasped by the grasping part is brought into contact with a target object to slide the grasped object, the control method comprising:

• acquiring, as a state variable indicating a state of the grasping system when the grasping system is performing the work, any one of:
• a position and a speed of the target object;
• a type, a shape, and a state of the target object;
• a relative position and a relative speed of the grasping part or the grasped object with respect to the target object;
• a presence or absence of contact between the grasping part or the grasped object and the target object;
• an operating force of the grasping part or an operating force of each joint part related to a grasping of the grasped object;
• a vibration of a contact part between the grasping part or the grasped object and the target object;
• a sliding sound emitted from the contact part between the grasping part or the grasped object and the target object;
• and a difference between a target posture and an actual posture or an adaptive amount of a control system;
• and
• calculating a target value of each of various actuators related to the work in the grasping system by inputting the state variable to the learned machine learner read from a storage unit storing a learned machine learner that has been learned by receiving a plurality of training data sets composed of a combination of the state variable acquired in advance and correct answer data corresponding to the state variable.