Granted Patent B2

US 12,611,767 · App. 19/322,446 · Granted Apr 28, 2026

Inventor: Jenna Reher (San Jose, CA)

Assignee: FIGURE AI INC.

B25J9/1605B25J13/003G06N3/008

Abstract

Claims (16)

1 . A method for controlling a humanoid robot, the method comprising:

identifying a task for the humanoid robot to perform;

establishing a safety-related limit for the humanoid robot;

constructing a phantom inverse kinematics task that is mathematically derived from the identified task, wherein the phantom inverse kinematics task is configured to regularize a pose of the humanoid robot;

formulating a combined inverse kinematics problem that incorporates: (i) the task, (ii) the phantom inverse kinematics task, and (iii) the safety-related limit;

solving the inverse kinematics problem to generate target joint parameters that both: (i) satisfy the task, and (ii) are within the established safety-related limit; and

providing the target joint parameters to one or more actuators to effect a physical movement of the humanoid robot.

2 . The method of claim 1 , wherein the safety-related limit comprises a kinematic control barrier function.

3 . The method of claim 2 , wherein the kinematic control barrier function is configured to enforce one or more of: a joint position limit, a torque limit, and a friction cone.

4 . The method of claim 1 , wherein the identified task is an existing motion task and the phantom inverse kinematics task is a phantom motion task paired to the existing motion task.

5 . The method of claim 1 , wherein the identified task is an existing contact task between the humanoid robot and an object in its environment, and the phantom inverse kinematics task is a phantom contact task paired to the existing contact task.

6 . The method of claim 5 , wherein the phantom motion task defines a cost function derived from a Jacobian and a set point of the existing motion task.

7 . The method of claim 1 , wherein the combined inverse kinematics problem is formulated as a quadratic program, and wherein decision variables for the quadratic program includes generalized velocities for the humanoid robot.

8 . The method of claim 7 , wherein solving the inverse kinematics problem comprises minimizing a cost associated with a set of stacked weighted objectives.

9 . The method of claim 7 , wherein the quadratic program is solved at a present time instance by a differential inverse kinematics solver without a predictive control horizon.

10 . The method of claim 1 , wherein the wherein the phantom inverse kinematics task regularizes the pose of the humanoid robot towards a reference pose, and wherein the reference pose is defined by a paired skeleton model.

Assignments (1)

Continuity (2)

References Cited (289)