Synthesizing machine learning-based controllers for underactuated robotic manipulators
AssigneeSiemens Aktiengesellschaft
InventorsYingnan Cui, Yaofeng Zhong, Georgia Olympia Brikis, Amit Chakraborty, Biswadip Dey
Life of this patent
Estimating the term…
Abstract
A computer-implemented system and method for synthesizing a controller for an under actuated robotic manipulator includes a machine learning based model having a plurality of neural network modules. Each module is configured to approximate a function related to an underactuated controller for a robotic manipulator. Parameters of each function are learned during training of the model using a loss function that satisfies one or more conditions including structure preservation, integrability and equilibrium assignment.
Claims20
A computer-implemented system for generating a feedback controller input to control an underactuated robotic manipulator, the system comprising:
- a processor;
- and
- memory having modules stored thereon with instructions to be executed by the processor, the modules comprising:
- a machine learning based model comprising a plurality of neural network modules, each module configured to approximate one of a plurality of functions related to an underactuated controller for a robotic manipulator, wherein the plurality of functions includes a desired mass, a desired potential energy, and an interconnection matrix;
- wherein parameters of each function are learned during training of the model using a loss function that satisfies one or more conditions including structure preservation, integrability and equilibrium assignment;
- wherein the trained model generates a desired mass matrix and a desired potential energy matrix to enable synthesis of a feedback controller input.
A computer-implemented method for generating a feedback control input to control an under-actuated robotic manipulator, the method comprising:
- approximating, by each of a plurality of neural network modules, one of a plurality of functions related to an underactuated controller for a robotic manipulator, wherein the plurality of functions includes a desired mass, a desired potential energy, and an interconnection matrix;
- wherein parameters of each function are learned during training of the model using a loss function that satisfies one or more conditions including structure preservation, integrability and equilibrium assignment;
- wherein the trained model generates a desired mass matrix and a desired potential energy matrix to enable synthesis of a feedback controller input.
Dependencies are inferred from the claim text.