top of page
Copy of Vibrant IP.png

VIP-3644

Adaptive Intelligence and Assured Autonomy

An independent supervisory neural network checks AI outputs against fixed limits, blocks unsafe actions and hands control back in autonomous machines.

  • The Adaptive Intelligence & Assured Autonomy portfolio covers an architecture for keeping neural networks inside defined limits while they operate. A supervisory controller or second neural network evaluates an operating network's output and allows it to control an apparatus only when it satisfies a fixed condition or confidence threshold.


    Neural networks are validated before deployment and then trusted in operation. The specifications address what happens when a network meets unfamiliar inputs, malfunctions or is altered: runtime checks detect an out-of-bounds output, block it, hand control to a person or alternative controller, and keep only in-bounds data for retraining.


    Family 1: Neural-Network Runtime Supervision and Boundary Control. Twelve US grants and one pending continuation, with priority to December 29, 2017. The independent claims cover a second neural network that is prevented from retraining and checks a primary network's output against a condition that cannot be changed after installation; self-correcting control that retrains a network after a breach and can hand control to a substantially identical network; supervisory and redundant networks implemented on solid-state chips; and control that proceeds only when the confidence level of a network's output meets a threshold, including confidence generated by a second network. Dependent claims add separation across machines, virtual machines or memory spaces.


    Family 2: Adaptive, Interchangeable and Integrity-Checked Neural Networks. Six US grants and one pending continuation, with priority to November 26, 2019. The independent claims cover selecting neural-network coefficient sets by current or predicted operating setting, such as time of day, location or weather; comparing simultaneously executed networks to confirm integrity; and moving lower-performing machines in a group onto a better-performing coefficient set while they continue operating. One specification works through a group of drones coordinated to intercept another vehicle.


    The owner publishes ten interactive simulations at apexai.biz, with 65 test scenarios across both families: runtime supervision in driving, maritime, air, robotics and manufacturing, plus model orchestration, swarm policy transfer, sensor-fusion integrity, a generative AI assistant and network defense. Scenarios include injected faults, fine-tuning regressions, model switching, twin-integrity checks and shared sensor errors that supervision cannot correct. In the driving simulation, supervision kept 20 of 20 faulted episodes safe, against 0 of 20 without it. All are synthetic, not field tests. Google Patents lists 236 publications citing the portfolio, including Micron, NVIDIA, Waymo and Toyota; a citation does not indicate use. Anticipated expirations run from 2038 to 2042.


    The specifications name autonomous land vehicles, drones, robots, wind turbines and IoT-controlled appliances. Potential commercial applications include safety supervisors for automated driving systems, run-time assurance for uncrewed aircraft, coordinated drone groups and counter-UAS interceptors, fleet control for warehouse robots, and confidence-gated control and integrity monitoring in AI accelerators and edge hardware.


    The portfolio is available for acquisition or license.

  • The IP and assets available for sale or license include:

    Neural-Network Runtime Supervision and Boundary Control (Family 1)

    62/612,008 (Provisional), 62/630,596 (Provisional), 62/659,359 (Provisional), US10324467, US10242665, US10672389, US10254760, US10620631, US10802488, US10802489, US10795364, US10627820, US11366472, US11815893, US12174631, US20250224724 (Pending, application 18/951,121)

    Adaptive, Interchangeable and Integrity-Checked Neural Networks (Family 2)

    62/940,762 (Provisional), US10691133, US11366434, US10956807, US12081646, US11367290, US11928867, US20240203132 (Pending, application 18/594,532)

  • Other Public Data Room Materials Available at VIP-3644 Public Data Room for Adaptive Intelligence & Assured Autonomy IP:

    • Introductory Materials

    • Patents and Patent Applications

    • Market Analysis

    • Technology Demonstrations

    Additional Materials (available upon signing a mutual non-disclosure agreement):

    • Due Diligence Review

    • Patent File Histories

  • Pricing guidance for this transaction may be available. All parties interested in the IP above are authorized to contact Radu Stoica at Vibrant IP, LLC at radu@vibrantip.com

bottom of page