Method for neural network control of motion using real-time environmental feedback

• US 5,673,367 A
• Filed: 09/15/1994
• Issued: 09/30/1997
• Est. Priority Date: 10/01/1992
• Status: Expired due to Fees

• Current Assignee: Theresa M. Buckley
• Original Assignee: Theresa M. Buckley
• Inventors: Buckley, Theresa M.
• Primary Examiner(s): Downs, Robert W.
• Assistant Examiner(s): Shah, Sanjiv
• Application Number:
  US08/307,388
• Time in Patent Office: 1,111 Days
• Field of Search: 395/22, 395/23, 395/24, 395/81, 395/94, 395/21, 364/167.01
• US Class Current: 706/23
• CPC Class Codes: B25J 9/1612 characterised by the hand, wrist, grip control
• CPC Class Codes: G05B 2219/33037 Learn parameters of network offline, not while controlling system
• CPC Class Codes: G05B 2219/39465 Two fingers each with 2-DOF

Abstract

A method of motion control for robotics and other automatically controlled machinery using a neural network controller with real-time environmental feedback. The method is illustrated with a two-finger robotic hand having proximity sensors and force sensors that provide environmental feedback signals. The neural network controller is taught to control the robotic hand through training sets using back- propagation methods. The training sets are created by recording the control signals and the feedback signal as the robotic hand or a simulation of the robotic hand is moved through a representative grasping motion. The data recorded is divided into discrete increments of time and the feedback data is shifted out of phase with the control signal data so that the feedback signal data lag one time increment behind the control signal data. The modified data is presented to the neural network controller as a training set. The time lag introduced into the data allows the neural network controller to account for the temporal component of the robotic motion. Thus trained, the neural network controlled robotic hand is able to grasp a wide variety of different objects by generalizing from the training sets.

Claims

1. A method for neural network control of a time-varying process, said time-varying process having an input variable capable of affecting an operating state of said time-varying process and an output variable indicative of said operating state of said time-varying process, said method comprising the steps of:

a) training a neural network controller by the steps of;

i) simulating said time-varying process and recording said input variable as a function of time and recording said output variable as a function of time to create a data set, said data set including input variable data and output variable data as a function of time;

ii) creating a training set from said data set by dividing said data set into increments of time and shifting the output variable data out of phase with the input variable data so that the output variable data lag at least one time increment behind the input variable data; and

iii) presenting said training set to said neural network controller so that said neural network controller learns a correlating relationship between said output variable and said input variable based on said training set;

b) subsequently controlling said time-varying process by the steps of;

j) receiving said output variable from said time-varying process as a feedback signal in said neural network controller;

jj) creating a control signal based on said feedback signal in accordance with said correlating relationship learned by said neural network controller from said training set; and

jjj) receiving said control signal from said neural network controller as the input variable of said time-varying process.

2. The method of claim 1 further comprising the steps of:

k) recording the input variable as a function of time and recording the output variable as a function of time during the operation of said time-varying process to create a supplemental data set, said supplemental data set including supplemental input variable data and supplemental output variable data as a function of time;

kk) creating a supplemental training set from said supplemental data set by dividing said supplemental data set into increments of time and shifting the supplemental output variable data out of phase with the supplemental input variable data so that the supplemental output variable data lag at least one time increment behind the supplemental input variable data; and

kkk) modifying said correlating relationship by presenting said supplemental training set to said neural network controller so that said neural network controller learns a modified correlating relationship between said output variable and said input variable based on said supplemental training set combined with said training set.

3. The method of claim 1, wherein:

step i) comprises simulating a time-varying process representing an operation of a machine, said input variable being capable of affecting an operating state of said machine, said output variable being indicative of said operating state of said machine, and recording said input variable as a function of time and recording said output variable as a function of time to create a data set, said data set including input variable data and output variable data as a function of time;

step j) comprises receiving said output variable from said machine as a feedback signal in said neural network controller; and

step jjj) comprises receiving said control signal from said neural network controller as the input variable of said machine.

4. The method of claim 1, wherein:

step i) comprises simulating a time-varying process representing a movement of a machine, said input variable being capable of affecting a position of said machine, said output variable being indicative of said position of said machine, and recording said input variable as a function of time and recording said output variable as a function of time to create a data set, said data set including input variable data and output variable data as a function of time;

step j) comprises receiving said output variable from said machine as a feedback signal in said neural network controller; and

step jjj) comprises receiving said control signal from said neural network controller as the input variable of said machine.

5. The method of claim 4 wherein step i) comprises simulating said movement of said machine by manually controlling said machine through said movement and recording said input variable as a function of time and recording said output variable as a function of time throughout said movement to create a data set.

6. The method of claim 1 wherein step i) includes the substep of simulating said time-varying process by operating a simulation means to produce simulated input variable data as a function of time to be recorded as said input variable data as a function of time and to produce simulated output variable data as a function of time to be recorded as said output variable data as a function of time.

7. A method for neural network control of a time-varying process, said time-varying process having an input variable capable of affecting an operating state of said time-varying process and an output variable indicative of said operating state of said time-varying process, said method comprising the steps of:

a) training a neural network controller by the steps of;

i) simulating said time-varying process and recording said input variable as a function of time and recording said output variable as a function of time to create a data set, said data set including input variable data and output variable data as a function of time;

ii) creating a training set from said data set by dividing said data set into increments of time and shifting the output variable data out of phase with the input variable data so that the output variable data lag at least one time increment behind the input variable data; and

iii) presenting said training set to an input layer of said neural network controller so that a hidden layer of said neural network controller learns a correlating relationship between said output variable and said input variable based on said training set within;

b) subsequently controlling said time-varying process by the steps of;

j) receiving said output variable from said time-varying process as a feedback signal in said input layer of said neural network controller;

jj) creating a control signal based on said feedback signal in accordance with said correlating relationship learned by said hidden layer of said neural network controller from said training set; and

jjj) receiving said control signal from an output layer of said neural network controller as the input variable of said time-varying process.

8. The method of claim 7 wherein step iii) further includes the substep of assigning a value to each of a plurality of connections between said input layer and said hidden layer, and assigning a value to each of a plurality of connections between said hidden layer and said output layer through back-propagation methods based upon said training set.

9. The method of claim 7, wherein:

step i) comprises simulating a time-varying process representing an operation of a machine, said input variable being capable of affecting an operating state of said machine, said output variable being indicative of said operating state of said machine, and recording said input variable as a function of time and recording said output variable as a function of time to create a data set, said data set including input variable data and output variable data as a function of time;

step j) comprises receiving said output variable from said machine as a feedback signal in said neural network controller; and

step jjj) comprises receiving said control signal from said neural network controller as the input variable of said machine.

10. The method of claim 7, wherein:

step i) comprises simulating a time-varying process representing a movement of a machine, said input variable being capable of affecting a position of said machine, said output variable being indicative of said position of said machine, and recording said input variable as a function of time and recording said output variable as a function of time to create a data set, said data set including input variable data and output variable data as a function of time;

step j) comprises receiving said output variable from said machine as a feedback signal in said neural network controller; and

step jjj) comprises receiving said control signal from said neural network controller as the input variable of said machine.

11. The method of claim 10 wherein step i) comprises simulating said movement of said machine by manually controlling said machine through said movement and recording said input variable as a function of time and recording said output variable as a function of time throughout said movement to create a data set.

12. The method of claim 7 wherein step i) includes the substep of simulating said time-varying process by operating a simulation means to produce simulated input variable data as a function of time to be recorded as said input variable data as a function of time and to produce simulated output variable data as a function of time to be recorded as said output variable data as a function of time.

13. A method for neural network motion control of a robotic end effector, said robotic end effector having an input variable capable of affecting a position of said robotic end effector and an output variable indicative of an operating state of said robotic end effector, said method comprising the steps of:

a) training a neural network controller by the steps of;

i) simulating a movement of said robotic end effector and recording said input variable as a function of time and recording said output variable as a function of time to create a data set, said data set including input variable data and output variable data as a function of time;

ii) creating a training set from said data set by dividing said data set into increments of time and shifting the output variable data out of phase with the input variable data so that the output variable data lag at least one time increment behind the input variable data; and

iii) presenting said training set to said neural network controller so that said neural network controller learns a correlating relationship between said output variable and said input variable based on said training set;

b) subsequently controlling the motion of said robotic end effector by the steps of;

j) receiving said output variable from said robotic end effector as a feedback signal in said neural network controller;

jj) creating a control signal based on said feedback signal in accordance with said correlating relationship learned by said neural network controller from said training set; and

jjj) receiving said control signal from said neural network controller as the input variable of said robotic end effector.

14. The method of claim 13 wherein step i) comprises simulating said movement of said robotic end effector by manually controlling said robotic end effector through said movement and recording said input variable as a function of time and recording said output variable as a function of time throughout said movement to create a data set.

15. The method of claim 13 wherein step i) includes the substep of simulating said movement of said robotic end effector by operating a simulation means to produce simulated input variable data as a function of time to be recorded as said input variable data as a function of time and to produce simulated output variable data as a function of time to be recorded as said output variable data as a function of time.

17. The method of claim 15, wherein:

step i) comprises simulating a time-varying process representing a trajectory of a machine with said operating state representing a position of said machine; and

step b) comprises controlling the trajectory of said machine in accordance with said correlating relationship learned by said neural network controller.

16. A method for neural network control of a time-varying process, said time-varying process having an input variable capable of affecting an operating state of said time-varying process and a time indication variable, said method comprising the steps of:

a) training a neural network controller by the steps of;

i) simulating said time-varying process and recording said input variable as a function of time and recording said time indication variable as a function of time to create a data set, said data set including input variable data and time indication variable data as a function of time;

ii) creating a training set from said data set by dividing said data set into increments of time and shifting the time indication variable data out of phase with the input variable data so that the time indication variable data lag at least one time increment behind the input variable data; and

iii) presenting said training set to said neural network controller so that said neural network controller learns a correlating relationship between said time indication variable and said input variable based on said training set;

b) subsequently controlling said time-varying process by the steps of;

j) receiving said time indication variable from said time-varying process as a feedback signal in said neural network controller;

jj) creating a control signal based on said feedback signal in accordance with said correlating relationship learned by said neural network controller from said training set; and

jjj) receiving said control signal from said neural network controller as the input variable of said time-varying process.