Self-correcting controller systems and methods of limiting the operation of neural networks to be within one or more conditions

• US 10,254,760 B1
• Filed: 06/04/2018
• Issued: 04/09/2019
• Est. Priority Date: 12/29/2017
• Status: Active Grant

• Current Assignee: Apex AI Industries LLC
• Original Assignee: Apex Artificial Intelligence Industries, Inc.
• Inventors: Abeloe, Kenneth A.
• Primary Examiner(s): Tzeng, Feng-Tzer
• Application Number:
  US15/997,192
• Time in Patent Office: 309 Days
• CPC Class Codes: B60Q 5/006 indicating risk of collision between vehicles or with pedestrians
• CPC Class Codes: B60R 11/0217 for loud-speakers
• CPC Class Codes: B60R 11/04 Mounting of cameras operative during drive; Arrangement of controls thereof relative to the vehicle
• CPC Class Codes: B60R 2300/102 using 360 degree surveillance camera system
• CPC Class Codes: B60R 2300/103 using camera systems provided with artificial illumination device, e.g. IR light source
• CPC Class Codes: B60W 30/0956 the prediction being responsive to traffic or environmental parameters
• CPC Class Codes: G05D 1/0088 characterized by the autonomous decision making process, e.g. artificial intelligence, predefined behaviours using knowledge based models G06N5/00
• CPC Class Codes: G05D 1/0221 involving a learning process
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/047 Probabilistic or stochastic networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: G06N 3/088 Non-supervised learning, e.g. competitive learning
• CPC Class Codes: G06N 5/046 Forward inferencing; Production systems
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G10L 13/00 Speech synthesis; Text to speech systems
• CPC Class Codes: G10L 13/047 Architecture of speech synthesisers
• CPC Class Codes: G10L 15/16 : using artificial neural networks
• CPC Class Codes: G10L 15/22 : Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 2015/223 : Execution procedure of a spoken command

Abstract

Systems and methods for automatically self-correcting or correcting in real-time one or more neural networks after detecting a triggering event, or breaching boundary conditions are provided. Such a triggering event may indicate incorrect output signal or data being generated by the one or more neural networks. In particular, machine controllers of the invention limit the operations of neural networks to be within boundary conditions. Autonomous machines of the invention can be self-corrected after a breach of a boundary condition is detected. Autonomous land vehicles of the invention are capable of determining the timing of automatic transition to the manual control from automated driving mode. The controller of the invention filters and saves input-output data sets that fall within boundary conditions for later training of neural networks. The controllers of the invention include security architectures to prevent damages from virus attacks or system malfunctions.

Claims

1. A controller for an autonomous machine having a plurality of sensors, the controller comprising:

a first neural network deployed on the autonomous machine, trained with a first training data set that includes training data generated by a sensor located remote from the autonomous machine, and configured to generate first output data after processing a set of input data;

a first processor coupled to the first neural network, including;

i) a detector adapted to receive the first output data and to determine whether the first output data breach a first predetermined condition; and

ii) a neural network manager coupled to the first neural network and adapted to re-train the first neural network using a second training data set if the detector determines the first output data breach the first predetermined condition; and

a second neural network structured and trained identical to the first neural network to generate a second output data by processing the set of input data, wherein the first and second neural networks are executed simultaneously,

wherein the first processor is further configured to operate the second neural network if the first output data breaches the first predetermined condition.

2. The controller of claim 1, wherein the second training data set includes data generated by a sensor on the autonomous machine.

3. The controller of claim 1, the controller further comprising:

a database management system coupled to the detector and constructed to receive and store the input data and the corresponding first output data if the detector determines the corresponding first output data does not breach the first predetermined condition, wherein the database management system is further adapted to retrieve the stored input data and the corresponding output data to form the second training data set.

4. The controller of claim 1, wherein the predetermined condition is defined with a machine recognizable human speech part.

5. The controller of claim 1, wherein the first predetermined condition is unchangeable after an initial installation onto the control system.

6. The controller of claim 1, further comprising a second processor configured to execute the first and second neural networks.

7. A method of controlling an autonomous machine using a controller configured with one or more processors, the method comprising:

generating first output data after processing a set of input data using a first neural network on the autonomous machine, wherein the first neural network is trained with a first training data set that includes training data generated by a sensor located remote from the autonomous machine;

processing the first output data and to determine whether the first output data breach a first predetermined condition;

re-training the first neural network using a second training data set if the first output data are determined to breach the first predetermined condition;

executing, simultaneously with generating the first output data using the first neural network, a second neural network structured and trained identical to the first neural network to generate a second output data by processing the set of input data vector; and

operating the second neural network if the first output data breaches the first predetermined condition.

8. The method of claim 7, wherein the second training data set includes data generated by a sensor on the autonomous machine.

9. The method of claim 7, further comprising:

storing the input data and the corresponding first output data if the detector determines the corresponding first output data does not breach the first predetermined condition, and

retrieving the stored input data and the corresponding output data to form the second training data set.

10. The method of claim 7, further comprising:

executing the first and second neural networks on a first thread on the one or more processors; and

executing the steps of processing the first output data and re-training the first neural network on a second thread of the one or more processors.

11. The method of claim 7, further comprising:

executing the first neural network on a first thread on the one or more processors; and

executing the second neural network on a second thread of the one or more processors.

12. The method of claim 7, wherein the first predetermined condition is defined with a machine recognizable human speech part.

13. The method of claim 7, wherein the first predetermined condition is unchangeable after an initial installation onto the control system.

14. A controller, for an autonomous machine having a plurality of sensors, the controller comprising:

a first neural network deployed on the autonomous machine, trained with a first training data set that includes training data generated by a sensor located remote from the autonomous machine, and configured to generate first output data after processing a set of input data;

a second neural network structured and trained identical to the first neural network, the second neural network configured to generate second output data by processing the set of input data, wherein the controller executes the first and second neural networks simultaneously; and

a first processor coupled to the first neural network and the second neural network, including;

i) a detector means for receiving the first output data and to determine whether the first output data breach a first predetermined condition;

ii) a neural network manager means coupled to the first neural network and for re-training the first neural network using a second training data set if the detector determines the first output data breach the first predetermined condition;

wherein the controller operates the second neural network if the first output data breaches the first predetermined condition.

15. The controller of claim 14, wherein the second training data set includes data generated by a sensor on the autonomous machine.

16. The controller of claim 14, the controller further comprises:

a database management system coupled to the detector, the database management system configured to receive and store the input data and the corresponding first output data if the detector determines the corresponding first output data does not breach the first predetermined condition, the database management system further configured to retrieve the stored input data and the corresponding output data to form the second training data set.