METHODS AND SYSTEMS FOR NONLINEAR ADAPTIVE CONTROL AND FILTERING

• US 20160147201A1
• Filed: 11/10/2015
• Published: 05/26/2016
• Est. Priority Date: 11/11/2014
• Status: Active Grant

• Current Assignee: Applied Brain Research, Inc.
• Original Assignee: Applied Brain Research, Inc.
• Inventors: Crncich-DeWolf, Travis, Stewart, Terrence Charles, Slotine, Jean-Jacques Emile, Eliasmith, Christopher David
• Granted Patent: US 10,481,565 B2
• Time in Patent Office: Days
• CPC Class Codes: G05B 13/0265 the criterion being a learning criterion
• CPC Class Codes: G05B 2219/41152 Adaptive filter
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

Methods, systems and methods for designing a system that provides adaptive control and adaptive predictive filtering using nonlinear components. A system design is described that provides an engineered architecture. This architecture defines a core set of network dynamics that carry out specific functions related to control or prediction. The adaptation systems and methods can be applied to limited areas of he system to allow the system to learn to compensate for unmodeled system dynamics and kinematics. Two types of adaptive modules are described which are configured to account for the unmodeled system dynamics and kinematics.

Claims

1. A system for providing at least one of adaptive control and adaptive filtering, the system comprising:

a command module configured to provide an initial output;

an output module coupled to the command module;

a feedback module configured to provide a system state input; and

an adaptive module coupled to the command module and the feedback module, the adaptive module comprising at least one adaptive sub-module coupled to the output module, each adaptive sub-module being configured to provide an output modifier to the output module;

wherein

each adaptive sub-module comprises a plurality of nonlinear components, wherein each nonlinear component is configured to generate a scalar output in response to the system state input and is coupled to the output module by at least one weighted coupling, each weighted coupling has a corresponding connection weight;

the scalar output from each nonlinear component is weighted by the connection weights of the corresponding weighted couplings and the weighted outputs are provided to the output module to form the output modifier;

each adaptive sub-module further comprises a learning module configured to update the connection weights for each weighted coupling in that adaptive sub-module based on the initial output; and

the output module is configured to generate a final output by modifying the initial output using the output modifiers.

2. The system of claim 1, wherein for each adaptive sub-module the output modifier is a sum of the weighted outputs from each of the nonlinear components.

3. The system of claim 1, wherein the at least one adaptive sub-module comprises a bias adaptation sub-module configured to provide a bias output modifier to the output module;

and

the output module is configured to generate the final output by adding the bias output modifier to the initial output.

4. The system of claim 3 wherein the learning module of the bias adaptation sub-module is configured to update connection weights based on the initial output and the outputs generated by the nonlinear components.

5. The system of claim 4 wherein the learning module of the bias adaptation sub-module is configured to update the connection weights based on an outer product of the initial output and the scalar outputs from the nonlinear components.

6. The system of claim 1, wherein the at least one adaptive sub-module comprises a transformative adaptation sub-module configured to provide a transformative output modifier to the output module;

and

the output module is configured to generate the final output by multiplying the transformative output modifier and the initial output.

7. The system of claim 6 wherein multiplying the initial output and the transformative output modifier transforms the initial output from a first system space to a second system space to generate the final output.

8. The system of claim 6, wherein the transformative adaptation sub-module is configured to provide the transformative output modifier directly, based on the weighted outputs from the nonlinear components.

9. The system of claim 6, wherein the learning module of the transformative adaptation sub-module is configured to update connection weights based on the initial output, the state input and the outputs from the nonlinear components.

10. The system of claim 9, wherein the learning module of the transformative adaptation sub-module is configured to update connection weights based on an outer product of the initial output and a product of the state input and the weighted outputs from the nonlinear components.

11. The system of claim 9, wherein the learning module of the transformative adaptation sub-module is configured to update connection weights based on an outer product of the initial output, the state input and the scalar outputs from the nonlinear components.

12. The system of claim 1, further comprising a motor module coupled to the output module;

wherein the final output is a control output configured to control operation of the motor module.

13. The system of claim 1, wherein the final output is a predictive output representing a prediction of a future system state.

14. The system of claim 1, wherein the nonlinear components are neurons.

15. The system of claim 1, wherein at least one of the command module, the output module and the feedback module are implemented in neurons.

16. The system of claim 15, wherein each of the command module, the output module and the feedback module is implemented in neurons.

17. The system of claim 14, wherein the neurons are spiking neurons.

18. The system of claim 1, wherein each nonlinear component has a tuning curve that determines the scalar output generated by the nonlinear component in response to any input and the tuning curve for each nonlinear component is generated randomly.

19. The system of claim 1, wherein the output modifier generated by at least one of the adaptive sub-modules is multidimensional.

20. The system of claim 19, wherein at least one of the nonlinear components in the adaptive sub-module that generates the multidimensional output modifier is coupled to the output module by a plurality of weighted couplings, one weighted coupling for each dimension of the multidimensional output modifier.

21. The system of claim 20, wherein each of the nonlinear components in the adaptive sub-module that generates the multidimensional output modifier is coupled to the output module by a plurality of weighted couplings, one weighted coupling for each dimension of the multidimensional output modifier.

22. A method for designing a system for providing at least one of adaptive control and predictive filtering, the method comprising:

defining a command module to provide an initial output, an output module to provide a final output, and a feedback module to provide a system state input;

defining an adaptive module comprising at least one adaptive sub-module, each adaptive sub-module being defined to provide an output modifier and comprising a learning module and a plurality of nonlinear components, each nonlinear component configured to generate a scalar output in response to the system state input; and

determining a plurality of initial couplings such that the adaptive module is coupled to the command module and the feedback module, the output module is coupled to the command module, and each nonlinear component is coupled to the output module by at least one weighted coupling;

wherein

each weighted coupling has a corresponding connection weight such that the scalar output generated by each nonlinear component is weighted by the corresponding connection weights to generate a weighted output and the weighted outputs from the nonlinear components combine to provide the output modifier;

the output module is configured to generate the final output by modifying the initial output using the output modifiers from each adaptive sub-module; and

each learning module is configured to update the connection weights for each weighted coupling in the corresponding adaptive sub-module based on the initial output.

23. The method of claim 22 wherein the initial couplings and connection weights are determined using a neural compiler.

24. A system for providing at least one of adaptive control and adaptive filtering, the system comprising:

a command module configured to provide an initial output;

an output module coupled to the command module;

a feedback module configured to provide a system state input; and

an adaptive module coupled to the command module and the feedback module, the adaptive module comprising at least one adaptive sub-module coupled to the output module, each adaptive sub-module being configured to provide an output modifier to the output module;

wherein

each adaptive sub-module comprises a plurality of nonlinear components, each nonlinear component is configured to generate a response to the system state input and is coupled to the output module by at least one weighted coupling, and each weighted coupling has a corresponding connection weight;

the response from each nonlinear component is weighted by the connection weights of the corresponding weighted couplings and the weighted responses are provided to the output module to form the output modifier; and

each adaptive sub-module further comprises a learning module configured to update the connection weights for each weighted coupling in that adaptive sub-module based on the initial output; and

the output module is configured to generate a final output by modifying the initial output using the output modifiers; and

further wherein

the at least one adaptive sub-module comprises a transformative adaptation sub-module configured to provide a transformative output modifier directly using the weighted responses from the nonlinear components.

25. The system of claim 24 wherein the output module is configured to generate the final output by multiplying the transformative output modifier and the initial output.