Machine learning device, robot controller, robot system, and machine learning method for learning action pattern of human
A machine learning device for a robot that allows a human and the robot to work cooperatively, the machine learning device including a state observation unit that observes a state variable representing a state of the robot during a period in that the human and the robot work cooperatively; a determination data obtaining unit that obtains determination data for at least one of a level of burden on the human and a working efficiency; and a learning unit that learns a training data set for setting an action of the robot, based on the state variable and the determination data.
1. A robot system, comprising:
at least one memory; and
at least one processor configured to:
obtain a state variable representing a state of a first robot when a first human and the first robot work cooperatively,
obtain an output from a neural network having the state variable as input information, and
control, based on the output from the neural network, an action of the first robot when the first human and the first robot work cooperatively,
wherein the neural network has been updated, based on reinforcement learning using a reward for at least a working efficiency when at least one human and at least one robot work cooperatively, to set the action of the first robot that leads to at least a higher working efficiency when the first human and the first robot work cooperatively,
wherein the at least one human includes at least one of the first human or a second, different human,
wherein the at least one robot includes at least one of the first robot or a second, different robot, and
wherein the reward is set based on at least a value of the working efficiency when the at least one human and the at least one robot work cooperatively.
2. The robot system according to claim 1 , wherein
the state variable includes information from at least one of a rotation angle detector, a camera, a motion sensor, a pressure sensor, a torque sensor for a motor, or a contact sensor.
3. The robot system according to claim 1 , wherein
the state variable includes information in relation to at least one of a position, an orientation, a velocity, or an acceleration of the first robot.
4. The robot system according to claim 1 , wherein
the neural network has been updated, based on the reinforcement learning using the reward which is further for a level of burden on the at least one human when the at least one human and the at least one robot work cooperatively, to set the action of the first robot that leads to a smaller level of burden on the first human when the first human and the first robot work cooperatively, and
the level of burden on the at least one human includes at least one of
an unexpected contact to the at least one human,
an unexpected pressing to the at least one human, or
a level of burden on a surrounding object.
5. The robot system according to claim 1 , wherein
the reward is set based on information in relation to at least one of
a magnitude of a load sensed by the at least one robot,
a direction of the load sensed by the at least one robot,
a magnitude of a load sensed by the at least one human,
a direction of the load sensed by the at least one human,
a magnitude of a load sensed by a surrounding object,
a direction of the load sensed by the surrounding object,
a movement time of the at least one robot,
information from a camera, or
information from a sensor.
6. The robot system according to claim 1 , wherein
the neural network is located on a cloud server.
7. The robot system according to claim 1 , wherein
the at least one processor is configured to control, based on an identification of the first human, the action of the first robot.
8. The robot system according to claim 1 , wherein
the at least one processor is configured to control, based on the output from the neural network, an action of a hand of the first robot.
9. The robot system according to claim 1 , further comprising:
the first robot.
10. The robot system according to claim 1 , wherein
the neural network has been updated, based on the reinforcement learning using the reward which is further for a level of burden on the at least one human when the at least one human and the at least one robot work cooperatively, to set the action of the first robot that leads to a smaller level of burden on the first human when the first human and the first robot work cooperatively, and
the reward becomes greater for a smaller value of the level of burden on the at least one human.
11. The robot system according to claim 1 , wherein
the at least one robot includes the first robot and the second robot,
the at least one human includes the first human working cooperatively with the first robot and the second human working cooperatively with the second robot, and
the at least one processor is configured to
control the action of the first robot and an action of the second robot independently from each other, and
share information which are output from the neural network updated by the first robot and from the neural network updated by the second robot with each other for controlling the first robot and the second robot.
12. The robot system according to claim 1 , wherein the first human and the first robot work cooperatively in a working area of the first robot where the first human and the first robot exert forces cooperatively on a same object for performing a same task.
13. A robot system, comprising:
at least one memory; and
at least one processor configured to:
obtain a state variable representing a state of a robot when a human and the robot work cooperatively,
obtain an output from a neural network having the state variable as input information, and
control, based on the output from the neural network, an action of the robot when the human and the robot work cooperatively,
wherein the neural network has been updated, based on supervised learning using training data with labels, to set the action of the robot that leads to at least a higher working efficiency when the human and the robot work cooperatively, and
wherein the training data has been obtained from at least one of (1) simulation performed outside the robot system or (2) another robot system.
14. The robot system according to claim 13 , wherein
the state variable includes information from at least one of a rotation angle detector, a camera, a motion sensor, a pressure sensor, a torque sensor for a motor, or a contact sensor.
15. The robot system according to claim 13 , wherein
the state variable includes information in relation to at least one of a position, an orientation, a velocity, or an acceleration of the robot.
16. The robot system according to claim 13 , wherein
the neural network has been updated, based on the supervised learning using the training data with the labels, to set the action of the robot that leads to a smaller level of burden on the human when the human and the robot work cooperatively, and
the level of burden on the human includes at least one of
an unexpected contact to the human,
an unexpected pressing to the human, or
a level of burden on a surrounding object.
17. The robot system according to claim 13 , wherein
the training data includes data obtained from the robot.
18. The robot system according to claim 13 , wherein
the neural network is located on a cloud server.
19. The robot system according to claim 13 , wherein
the at least one processor is configured to control, based on an identification of the human, the action of the robot.
20. The robot system according to claim 13 , wherein
the at least one processor is configured to control, based on the output from the neural network, an action of a hand of the robot.
21. The robot system according to claim 13 , further comprising:
the robot.
22. The robot system according to claim 13 , wherein
the at least one robot includes the first robot and the second robot,
the at least one human includes the first human working cooperatively with the first robot and the second human working cooperatively with the second robot, and
the at least one processor is configured to
control the action of the first robot and an action of the second robot independently from each other, and
share information which are output from the neural network updated by the first robot and from the neural network updated by the second robot with each other for controlling the first robot and the second robot.
23. The robot system according to claim 13 , wherein the human and the robot work cooperatively in a working area of the robot where the human and the robot exert forces cooperatively on a same object for performing a same task.
24. A robot controlling method, comprising:
obtaining a state variable representing a state of a first robot when a first human and the first robot work cooperatively;
obtaining an output from a neural network having the state variable as input information; and
controlling, based on the output from the neural network, an action of the first robot when the first human and the first robot work cooperatively,
wherein the neural network has been updated, based on reinforcement learning using a reward for at least a working efficiency when at least one human and at least one robot work cooperatively, to set the action of the first robot that leads to at least a higher working efficiency when the first human and the first robot work cooperatively,
wherein the at least one human includes at least one of the first human or a second, different human,
wherein the at least one robot includes at least one of the first robot or a second, different robot, and
wherein the reward is set based on at least a value of the working efficiency when the at least one human and the at least one robot work cooperatively.
25. The robot controlling method according to claim 24 , wherein
the neural network has been updated, based on the reinforcement learning using the reward which is further for a level of burden on the at least one human when the at least one human and the at least one robot work cooperatively, to set the action of the first robot that leads to a smaller level of burden on the first human when the first human and the first robot work cooperatively, and
the reward becomes greater for a smaller value of the level of burden on the at least one human.
26. A robot controlling method, comprising:
obtaining a state variable representing a state of a robot when a human and the robot work cooperatively;
obtaining an output from a neural network having the state variable as input information; and
controlling, based on the output from the neural network, an action of the robot when the human and the robot work cooperatively,
wherein the neural network has been updated, based on supervised learning using training data with labels, to set the action of the robot that leads to at least a higher working efficiency when the human and the robot work cooperatively, and
wherein the training data has been obtained from at least one of (1) simulation performed outside the robot system or (2) another robot system.