System and method for user intent recognition
Inventors
Swift, Timothy Alan • Cox, Nicolas • Kemper, Kevin Conrad
Assignees
Publication Number
US-11259979-B2
Publication Date
2022-03-01
Expiration Date
2038-02-02
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
A method of operating an exoskeleton system that includes determining a first state estimate for a current classification program being implemented by the exoskeleton system; determining a second state estimate for a reference classification program; determining that a difference between the first and second state estimate is greater than a classification program replacement threshold; generating an updated classification program; and replacing the current classification program with the updated classification program based at least in part on the determining that the difference between the first and second state estimates is greater than the classification program replacement threshold.
Core Innovation
The invention relates to a user intent recognition system for exoskeletons that uses adaptive classification programs to improve the accuracy and responsiveness of intent detection based on data collected from sensors. Unlike conventional finite state machines, which require significant expert input and manual tuning, the disclosed methods allow automatic and data-driven refinement of intent recognition programs, enabling them to adapt to both the general population and individual users over time.
This system determines state estimates using both the current and a reference classification program by processing sensor data to predict user intentions. If a significant difference is detected, surpassing a predefined threshold and indicating the reference program is more accurate, the current program is updated or replaced with an improved version. The update process can occur locally or remotely and can use data aggregated from multiple users or exoskeletons.
The invention solves problems in the prior art related to slow adaptation, lack of personalization, the need for manual supervision, and reduced classification speed or accuracy due to sensor noise or hardware changes. The new system also incorporates mechanisms for user feedback, programmatically adjusting adaptation rates, and can address sensor drift or failure by automatically tuning or updating classification programs accordingly. These innovations provide for more reliable, efficient, and user-tailored exoskeleton operation.
Claims Coverage
The independent claims disclose five main inventive features that define the scope and uniqueness of the invention.
Wearable pneumatic exoskeleton network with adaptive classification program updates
A network that includes a wearable pneumatic exoskeleton system having multiple pneumatic actuators, a pneumatic system, sensors, memory storing a classification program, and a processor executing the program to control actuation. The system is operably connected with a local user device and a remote classification server via the Internet. The inventive feature includes the ability to determine state estimates from current and reference classification programs, compare prediction accuracy using sensor data, and replace the current program with an updated version if the reference program is more accurate and exceeds a replacement threshold.
Remote updating of classification programs for multiple exoskeleton systems
A feature where a classification server communicates with a plurality of wearable pneumatic exoskeleton systems, each with their own sensors and actuators, and replaces the respective current classification program of each device with an updated program when a determined accuracy difference, calculated from successful and unsuccessful intended user action predictions, surpasses a replacement threshold.
Exoskeleton system with local processing for adaptive intent recognition
An exoskeleton system comprising actuators, sensors, memory, and processor, wherein the system determines state estimates from current and reference classification programs using sensor data, compares differences, and updates the classification program when differences exceed a threshold, enabling local adaptation without reliance on external servers.
Method for determining and updating exoskeleton classification programs based on state prediction differences
A method comprising the steps of determining first and second state estimates using current and reference classification programs with sensor data, determining when the difference surpasses a replacement threshold, generating an updated classification program, and replacing the current program when warranted. The method includes training or updating the program based on actual system performance data.
User feedback-driven modification and adaptation of exoskeleton classification programs
A feature allowing the system to sense exoskeleton state changes, determine classifications for these states, present classifications to users via a display, obtain user confirmation or rejection (including via touchscreens or buttons), and modify the classification program based on user responses. Additionally, the system selectively presents confirmation requests when predicted classification changes are above a performance difference threshold.
The inventive features cover a comprehensive networked and adaptive exoskeleton intent recognition system using sensor-based, data-driven classification updates, remote and local program management, and user-in-the-loop feedback for refinement.
Stated Advantages
Intent recognition programs can improve accuracy, reduce delay in recognition, and customize performance for each user.
The methods allow refinement of intent recognition without requiring supervisory experts, enabling unsupervised automatic learning and adaptation.
The system can automatically adjust to sensor or hardware changes, maintaining or improving classification accuracy over time.
Allows high frequency or user-tailored adaptation rates for program updates, balancing responsiveness and user acceptance.
Automated updating minimizes the need for user or administrator input, increasing operational efficiency.
Documented Applications
Operation and control of wearable exoskeleton systems that assist user movement by recognizing user intent for actions such as walking, running, jumping, climbing, lifting, throwing, squatting, and similar motions.
Implementation in exoskeletons for any body area, including legs, torso, arms, head, and complete body suits.
Application to non-human users, such as animals or robotic devices, for intent recognition and adaptive control.
Use in other worn devices using onboard sensors to recognize operator intent, including active footwear.
Interested in licensing this patent?