IP Library Granted Patent US 12697720
Granted Patent B2
US 12697720 · App. 19/355,531 · Granted Aug 4, 2026

Dynamic task assignment amongst communicating humanoid robots

Inventors: Corey Lynch (San Jose, CA); Toki Migimatsu (San Jose, CA); Yevgen Chebotar (San Jose, CA); Michael Ahn (San Jose, CA); Ivan Babushkin (San Jose, CA)
Assignee: FIGURE AI INC.
B25J9/163B25J9/1661
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12697720
App. No.
19/355,531
Granted
Aug 4, 2026
Kind
B2
Abstract

The present disclosure provides a method for coordinating task execution among multiple humanoid robots, comprising receiving a high-level task command, decomposing it into sub-tasks, determining a cost-optimized assignment using a cost-optimized bipedal action model (CoBAM) based on energy consumption, time to completion, and robot capabilities, and transmitting the assignment to assigned robots. The CoBAM comprises a hierarchical architecture including an L2 beta model operating at 1-20 Hz for high-level planning and an L1 alpha model operating at 100-10,000 Hz for continuous control commands. The cost function considers battery levels, physical distances between robot and sub-task locations, and mechanical wear factors associated with specific joint movements.

Claims (17)

1 . A method for coordinating task execution among multiple humanoid robots, comprising:

receiving a high-level task command;

decomposing the high-level task command into a plurality of sub-tasks;

determining a cost-optimized assignment of the plurality of sub-tasks to at least one humanoid robot of the multiple humanoid robots based on a cost function that considers at least one of energy consumption, or time to completion; and

transmitting the cost-optimized assignment to the at least one humanoid robot for execution; and

using a bipedal action model to generate output continuous joint torques based in part on the cost-optimized assignment.

2 . The method of claim 1 , wherein the cost function further considers at least one of battery levels of the available humanoid robots, physical distances between robot locations or sub-task locations, or mechanical wear factors associated with specific joint movements required for each sub-task.

3 . The method of claim 1 , wherein the cost-optimized assignment is determined using a cost-optimized bipedal action model, and said cost-optimized bipedal action model includes: (i) a beta model is configured to operate at a frequency between 1-25 Hz and (ii) an alpha model is configured to operate at a frequency between 100-10,000 Hz.

4 . The method of claim 3 , wherein the beta model is deployed on a remote AI system and the alpha model is deployed locally on each humanoid robot of the multiple humanoid robots.

5 . The method of claim 1 , wherein the cost-optimized assignment is determined using a cost-optimized bipedal action model, and said cost-optimized bipedal action model is trained by:

collecting training data that includes human data;

annotating the training data with cost metrics associated with robot actions, the cost metrics include at least one of energy consumption or time to completion; and

training a hierarchical model architecture that includes a beta model and an alpha model using the annotated training data.

6 . The method of claim 5 , wherein the beta model and the alpha model are co-trained end-to-end using a supervised learning technique.

7 . The method of claim 5 , further comprising the step of: splicing the collected training data into segments; and using a transformer-based model to generate written text that describes at least one object or action associated with the segment.

8 . The method of claim 5 , wherein the cost metrics associated with robot actions are determined by a separate machine learning model.

9 . The method of claim 1 , wherein the continuous joint torques are provided as input to a reinforcement learning (RL)-based controller, and wherein said RL-based controller is configured to control at least one actuator contained in said at least one humanoid robot.