Controller for autonomous device

• US 5,642,467 A
• Filed: 01/31/1995
• Issued: 06/24/1997
• Est. Priority Date: 01/31/1995
• Status: Expired due to Term

• Current Assignee: Penn State Research Foundation
• Original Assignee: Penn State Research Foundation
• Inventors: Gibson, Ronald E., Stover, James A.
• Primary Examiner(s): Downs, Robert W.
• Assistant Examiner(s): CHUNG, JI YONG DAVID
• Application Number:
  US08/381,257
• Time in Patent Office: 875 Days
• Field of Search: 395/11, 395/22, 395/23, 395/51, 395/85
• US Class Current: 700/250
• CPC Class Codes: G05B 13/0285 using neural networks and fuzzy logic

Abstract

A controller for directing the actions of an autonomous device in response to the existence or actions of objects in the physical world comprises program means for fusing physical world data and inferred property confidence factors into a plurality of representational instances, at least one representational instance having an inferred behavior interface property for soliciting a response of the autonomous device, program means stored in a random access memory for implementing a plurality of independent behavior instances for carrying out response tasks, program means stored in the random access memory for implementing a mission control task by generating a task list of behavior instances to be performed and program means stored in the random access memory for implementing an execution engine for executing the behavior instance at the head of the task list.

Claims

1. A controller for directing the actions of an autonomous device in response to the existence or actions of objects in the physical world comprising:

a programmable computing system having at least one central processing unit, random access memory, and input/output channels, said input/output channels connected to data acquisition devices for gathering data about the physical world in the vicinity of the autonomous device and output effectors for outputting signals to effect the physical response of the autonomous device,

program means stored in said random access memory for inputting the data relating to said physical world objects in the form of a stream of data vectors,

program means stored in said random access memory for implementing a continuous inference network for generating inferred property confidence factors related to objects in the physical world,

program means stored in said random access memory for repetitively fusing physical world data and inferred property confidence factors into a plurality of Representational instances, said Representational instances comprising a list of physical and inferred properties and their values and confidence factors, respectively,

at least one Representational instance having an inferred behavior interface property for soliciting a response of the autonomous device,

program means stored in said random access memory for implementing a plurality of independent behavior instances for carrying out response tasks,

program means stored in said random access memory for implementing a mission control task by generating a task list of behavior instances to be performed wherein each said behavior instance includes at least one goal achievement function for testing for completion and/or modifying of the behavior response, and

program means stored in said random access memory for implementing an execution engine for executing the behavior instance at the head of the task list.

2. The controller according to claim 1 wherein the continuous inference network is a classification network based upon a fuzzy logic in which the output confidence factor of any node in the network is representative of an inferred property.

3. The controller according to claim 2 wherein the continuous inference network is a classification network in which the inputs to any node may be weighted by adjustable weighing factors.

4. The controller according to claim 3 wherein at least one response task includes an analysis function for adjusting at least one adjustable weighing factor in the continuous inference network.

5. The controller according to claim 1 wherein each said behavior instance includes at least one action function and a plan designing function for generating a task list of action functions to be performed to carry out a behavior response.

6. A controller for directing the actions of an autonomous device in response to mission orders and the existence of actions or objects in the physical world comprising:

a programmable computing system having at least one central processing unit, random access memory, and input/output channels, said input/output channels connected to data acquisition devices for gathering data about the physical world in the vicinity of the autonomous device and output effectors for outputting signals to effect the physical response of the autonomous device,

program means stored in said random access memory for inputting the data relating to said physical world objects in the form of a stream of data vectors,

program means stored in said random access memory for implementing a continuous inference network for generating inferred property confidence factors related to objects in the physical world,

program means stored in said random access memory for repetitively fusing physical world data and inferred property confidence factors into a plurality of Representational instances, said Representational instances comprising a list of physical and inferred properties and their values and confidence factors, respectively,

at least one Representational instance having an inferred behavior interface property for soliciting a response of the autonomous device,

program means stored in said random access memory for implementing a plurality of independent behavior instances for carrying out responses,

each said behavior response instance including at least one action function and associated goal achievement function for testing for completion of and/or modifying the action function,

a behavior response decision function for polling the behavior interface properties of at least one of said Representational instances, making a decision whether or not to request activation of the behavior instance and generating activation requests,

a plan designing function for generating a task list of action functions to be performed to carry out a behavior response, said action functions sending signals to said input/output channels, and

at least one goal achievement function for testing for completion or modifying of the behavior response,

program means stored in said random access memory for implementing a mission control task for defining said mission plan by generating a task list of behavior instances to be performed and arbitrating between activation requests from behavior instances, and

program means stored in said random access memory for implementing an execution engine which seriatim;

calls the current action function goal achievement function,

calls the current behavior response goal achievement function,

calls a response decision function that polls response behavior instances to determine if any instance requests activation based upon the observation of a behavior interface property and calls a mission response decision from the mission level,

calls a function to activate the mission planning function to order or reorder the mission plan, and

calls for the execution of the next behavior instance or the next action function, sets response directive and repeats.

7. The controller according to claim 6 wherein the continuous inference network is a classification network based upon a fuzzy logic in which the output confidence factor of any node in the network being representative of an inferred property is a continuous and discriminatory function of the inputs to the node with an output confidence factor between zero and unity.

8. The controller according to claim 7 wherein the continuous inference network is a classification network in which the inputs to any node may be weighted by adjustable weighing factors.

9. The controller according to claim 8 wherein at least one response task includes an analysis function for adjusting at least one adjustable weighing factor in the continuous inference network.

10. The controller according to claim 6 wherein the means for repetitively fusing determines which Representational instance is to be updated with a given input data vector in the input vector stream by comparing said data vector with the corresponding physical property vector associated with each Representational instance, applying each value of the comparison to a BLEND function wherein the upper and lower bounds are selected for each property to weigh each comparison and applying each BLEND output to a vague AND node to produce a confidence factor indicative of a match, and updating the physical property vector of the Representational instance so matched, but if no Representational instance is matched then adding a new Representational instance and if after a predetermined time a Representational instance is not matched then deleting that Representational instance.

11. The controller according to claim 6 wherein the data acquisition devices may be adjusted and at least one response task includes an action function for adjusting a data acquisition device to improve performance relative to the response task that has been initiated.

12. The controller according to claim 6 wherein the autonomous device is mobile and has a mission response task which includes a plan designing function for generating a list of behavior response tasks.

13. The controller according to claim 6 wherein the autonomous device is mobile and the behavior instances include a Collision-Avoidance task for directing the movement of the mobile device to avoid an obstacle in response to a high level confidence factor assigned to an "Obstacle" behavior interface property of a Representational instance.

14. The controller according to claim 6 wherein the autonomous device is mobile and the behavior instances include a Transit task for directing the movement of the mobile device to a location identified by the mission orders.

15. The controller according to claim 6 wherein the autonomous device is mobile and the behavior instances include a Pursue task for directing the movement of the mobile device to pursue an obstacle in response to a high level confidence factor assigned to a Pursue behavior interface property of a Representational instance.

16. The controller according to claim 6 wherein the programming means are implemented in an object oriented programming environment and the Representational instances are program objects of the class Representational and the Behavior instances are program objects of the class Behavior.

17. The controller according to claim 6 wherein the programming means are implemented in a list processing language enabling the plan designing functions to be redefined by the controller during operation.

18. The controller according to claim 6 wherein the continuous inference network is a classification network in which the output of an AND node is the product of factors each of which is a function of an input.

19. The controller according to claim 8 wherein the continuous inference network is a classification network in which the output of an OR node is the Euclidean norm mapped between zero and unity of factors each of which is a function of an input.

20. The controller according to claim 7 wherein the continuous inference network is a classification network in which the output of a NOT node is the product of two factors, one of which is the input to be negated subtracted from 1 and the other is a global knowledge confidence factor being representative of the completeness of available physical world data.

21. The controller according to claim 7 wherein the physical data inputs to the continuous inference network are scaled by a BLEND function to have values between zero and unity.