Granted Patent B2

US 12,198,274 · App. 17/618,783 · Granted Jan 14, 2025

Inventors: Demetri Terzopoulos (Los Angeles, CA); Masaki Nakada (Los Angeles, CA)

Assignee: The Regents of the University of California

G06T19/003A61B3/0025G06F30/20G06T7/0012G06T2207/30041

Abstract

Claims (39)

1. A method for operating a sensorimotor control framework, the method comprising:

determining a current state of a musculoskeletal model comprising a torso complex and a set of one or more limb complexes attached to the torso complex;

determining a desired state of the musculoskeletal model, wherein the current and desired states comprise at least one of a position, an orientation, a muscle strain, and a strain rate;

computing forces exerted by the set of limb complexes;

computing a set of muscle activations for the torso complex, based on the forces exerted by the set of limb complexes and a difference between the current and desired states of the musculoskeletal model; and

driving a musculoskeletal model based on the computed muscle activations.

2. The method of claim 1 , wherein the musculoskeletal model is a model of a human being and comprises a plurality of Hill-type actuators.

3. The method of claim 1 , wherein the musculoskeletal model comprises a plurality of complexes, computing the set of muscle activations comprises computing a first set of muscle activations for a first complex and computing a second set of muscle activations for a second complex.

4. The method of claim 3 , wherein the plurality of complexes comprises at least one of an arm complex, a leg complex, a cervicocephalic complex, and a torso complex.

5. The method of claim 3 , wherein computing the second set of muscle activations for the second complex comprises:

computing forces from the first set of muscle activations; and

computing the second set of muscle activations based on the computed forces.

6. The method of claim 1 further comprising receiving sensory inputs from a sensory element, wherein the sensory element is an eye model and the sensory inputs comprise a one-dimensional optic nerve vector (ONV).

7. The method of claim 6 , wherein the eye model has a non-uniform distribution of photoreceptors on the retina.

8. The method of claim 1 , wherein computing the set of muscle activations comprises:

computing a voluntary signal using a voluntary controller;

computing a reflex signal using a reflex controller; and

computing the set of muscle activations based on outputs from the voluntary controller and outputs from the reflex controller.

9. The method of claim 8 , wherein the voluntary controller takes as input a target orientation and output activations for muscles of an associated muscle group.

10. The method of claim 8 , wherein the reflex controller takes as input muscle strains and strain rates of muscles of an associated muscle group.

11. A method for training a torso controller of a torso complex, the method comprising:

receiving training data comprising a set of target inputs and a target state of the torso complex and limb complexes attached to the torso complex;

determining forces exerted by the limb complexes attached to the torso complex;

computing muscle activations for the torso complex based on the target inputs, the target state, and the determined forces;

computing differences between a result of the muscle activations and the target state; and

modifying parameters of the torso controller based on the differences between the target state and a current state of the torso complex and the limb complexes, wherein the current and target states comprise at least one of a position, an orientation, a muscle strain, and a strain rate.

12. The method of claim 11 further comprising synthesizing the training data.

13. The method of claim 12 , wherein synthesizing the training data comprises:

specifying a current position and a target position for a model;

determining discrepancies between the current position and the target position;

determining muscle activation adjustments based on the discrepancies and muscle activations of the current position;

determining discrepancies comprises solving an inverse kinematics problem.

14. The method of claim 13 , wherein the discrepancies comprise at least one of an angular discrepancy, a linear discrepancy, and an acceleration discrepancy.

15. The method of claim 13 , wherein determining muscle activation adjustments comprises performing a muscle optimization technique to compute the minimal muscle activations that can generate a desired torque.

16. The method of claim 12 , wherein synthesizing the training data comprises:

specifying current and desired muscle strains and strain rates for a model;

determining changes in muscle strain and in strain rate for muscles of the controlled group; and

determining muscle activation adjustments based on the changes in muscle strain and in strain rate.

17. The method of claim 12 , wherein determining forces exhibited by limb complexes comprises computing random forces for the set of limb complexes by sending activation signals to the limb complexes.

Assignments (1)

Continuity (2)

References Cited (87)