Aircraft with modeled GPS

Abstract

An aircraft has an inertial measurement unit (IMU) configured to detect real-time inertial information relating to the aircraft, a receiver configured to receive global positioning system (GPS) signals, and a navigation controller operatively coupled with the IMU and the receiver. The navigation controller is configured to navigate the aircraft as a function of the inertial information received from the IMU and a GPS signal received by the receiver. Importantly, the navigation controller also is configured to navigate as a function of a simulated GPS signal produced by a reinforcement learning policy model.

Description

[0001] 126933-10102-7362487

[0002] 8 / 5 / 2025

[0003] 1

[0004] AIRCRAFT WITH MODELED GPS

[0005] PRIORITY

[0006] This patent application claims priority from provisional United States patent application number 63 / 680,334, filed August 7, 2024, entitled, "GPS- ALTERNATIVE UAV NAVIGATION SYSTEM," and naming Jesse Hamel and Jordan Hamel as inventors, the disclosure of which is incorporated herein, in its entirety, by reference.

[0007] GOVERNMENT RIGHTS

[0008] None

[0009] FIELD

[0010] Illustrative embodiments of the invention generally relate to aircraft and, more particularly, various embodiments of the invention relate to managing navigation of aircraft.

[0011] BACKGROUND

[0012] Modern aircraft— whether manned or unmanned— routinely combine inertial measurement units (IMUs) and Global Positioning System (GPS) receivers in their navigation systems to achieve reliable, accurate positioning. An IMU uses inertial sensors (and sometimes magnetometers) to measure changes in motion and orientation at high frequency, providing precise shortterm tracking of the aircraft's attitude, velocity, and relative position.

[0013] Meanwhile, GPS offers absolute, globally referenced positioning data that corrects the drift inherent in integrating IMU outputs over time. This combination is important because while the IMU delivers smooth, rapid updates needed for control loops, it suffers from error accumulation over time. GPS, by contrast, provides stable long-term position fixes, but at lower rates and with susceptibility to noise.126933-10102-7362487

[0014] 8 / 5 / 2025

[0015] 2

[0016] For manned aircraft, this fusion enables pilots and autopilot systems to maintain accurate navigation even in poor visibility or when flying complex instrument approaches. Unmanned systems— like drones— rely even more heavily on this pairing, as they often need to navigate autonomously with high precision. The fusion of IMU and GPS data allows these vehicles to follow preplanned routes, maintain stable flight in wind or turbulence, and avoid obstacles, all while providing ground controllers or onboard autonomy software with accurate situational awareness. This integration is typically managed by sensor fusion algorithms, such as Kalman filters, which intelligently combine the complementary strengths of both sensor types.

[0017] Undesirably, these navigation systems face challenges when GPS signals are unavailable or compromised. For example, in urban canyons, dense forests, or indoor environments, GPS signals can be blocked or degraded, leaving the IMU to estimate position alone— a task at which it will drift increasingly over time. Worse, GPS signals can be deliberately jammed or spoofed (falsely modified to mislead the receiver), potentially tricking the navigation system into believing it is somewhere it is not.

[0018] SUMMARY OF VARIOUS EMBODIMENTS

[0019] In accordance with one embodiment of the invention, an aircraft has an IMU configured to detect real-time inertial information relating to the aircraft, a receiver configured to receive GPS signals, and a navigation controller operatively coupled with the IMU and the receiver. The navigation controller is configured to navigate the aircraft as a function of the inertial information received from the IMU and a GPS signal received by the receiver. Importantly, the navigation controller is also configured to navigate as a function of a simulated GPS signal produced by a reinforcement learning policy model.

[0020] The aircraft may have any of a number of other components, including a propulsion system operatively coupled with the navigation controller. The126933-10102-7362487

[0021] 8 / 5 / 2025

[0022] 3

[0023] propulsion system thus may be controlled by the navigation controller. Among other ways, the navigation controller may be configured to use a simulated GPS signal when the GPS signal is detected to be unavailable, spoofed, or corrupted.

[0024] Preferably, the policy model includes an artificial intelligence model pre-trained in a simulated environment relating to a prescribed environment of flight of the aircraft. In that case, the policy model may be pre-trained to learn a navigation policy that maximizes a cumulative reward function. In addition, the policy model may be configured to predict relative position changes of the aircraft as a function of the inertial information.

[0025] In illustrative embodiments, the navigation controller uses factor graphing to estimate positions based on previously determined positions and known or estimated speeds and courses over time. Additionally, the navigation controller may execute sensor fusion to continuously integrate the inertial information from the IMU to maintain ongoing position updates during flight.

[0026] The navigation controller may include an autonomous navigation controller. Moreover, the aircraft also may have a body forming one of a drone, plane, helicopter, or airship. Regardless of the form factor, the navigation system, receiver, and IMU may be onboard the aircraft.

[0027] The navigation controller may be configured to detect corrupted or spoofed GPS signals by comparing received GPS signals to a predicted position produced by the model. If spoofed or corrupted, then the navigation controller may use the simulated GPS signal.

[0028] In accordance with other embodiments, a method of navigating an aircraft uses an IMU onboard the aircraft to detect real-time inertial information relating to the aircraft and receives a GPS signal from an external source. After determining whether the received GPS signal is sufficient for navigation, the method navigates the aircraft as a function of the inertial information received from the IMU and the received GPS signal when the126933-10102-7362487

[0029] 8 / 5 / 2025

[0030] 4

[0031] received GPS signal is determined to be sufficient for navigation. However, the method may alternatively navigate the aircraft as a function of the inertial information received from the IMU and a simulated GPS signal when the received GPS signal is determined to be insufficient for navigation. In either case, the simulated GPS signal is produced by a reinforcement learning (RL) policy model executing onboard the aircraft.

[0032] Illustrative embodiments of the invention are implemented as a computer program product having a computer usable medium with computer readable program code thereon. The computer readable code may be read and utilized by a computer system in accordance with conventional processes.

[0033] BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Those skilled in the art should more fully appreciate advantages of various embodiments of the invention from the following "Description of Illustrative Embodiments," discussed with reference to the drawings summarized immediately below.

[0035] Figure 1 schematically shows an aircraft with various components of illustrative embodiments.

[0036] Figure 2 schematically shows more details of the various components of illustrative embodiments.

[0037] Figure 3 schematically shows various data processing engines configured to facilitate navigation with or without a simulated GPS signal.

[0038] Figure 4 shows a high-level process of controlling navigation of the aircraft in accordance with illustrative embodiments.

[0039] Figure 5 shows more details of a process of navigating the aircraft of Figure 1 in accordance with illustrative embodiments.

[0040] Figure 6 shows more details of the RL policy model making real-time inferences while navigating the aircraft of Figure 1 in accordance with illustrative embodiments of the invention.126933-10102-7362487

[0041] 8 / 5 / 2025

[0042] 5

[0043] Figure 7 shows a process for training the RL policy model in accordance with illustrative embodiments.

[0044] DESCRIPTION OF ILLUSTRATIVE EMBODIMENTS Illustrative embodiments provide a robust aircraft navigation system capable of maintaining reliable navigation even when actual GPS signals are lost or spoofed. To that end, the system has a navigation controller that employs a reinforcement learning policy model to generate simulated GPS data when actual GPS signals are unavailable or unreliable. In such cases, the navigation controller substitutes the simulated GPS signal for the degraded / absent one, enabling continued navigation to the desired destination until the actual GPS signal is restored and usable. Details of illustrative embodiments are discussed below.

[0045] Figure 1 schematically shows an aircraft 104 with various components of illustrative embodiments. The aircraft 104 may include manned or unmanned aircraft, including fixed-wing airplanes, helicopters, drones, guided missiles, airships, and other forms of unmanned aerial vehicles (UAVs).

[0046] Although manned aircraft 104 may control navigation manually, in some cases aircraft 104 may utilize autopilot devices that navigate the aircraft 104 in lieu of a human pilot.

[0047] The aircraft 104 may include a navigation control system that controls the speed, altitude, and bearing to navigate from a starting location to an ending location. The navigation control system may include a navigation controller 112 coupled to an inertial measurement unit (IMU) 116 and a global positioning system (GPS) receiver 108. The navigation control system includes a number of data processing engines to autonomously obtain sensor data and make real-time calculations to synthesize GPS coordinates.

[0048] As known by those in the art, the IMU 116 is an electronic device that measures and reports a body's (e.g., the aircraft 104) specific force, angular rate, and sometimes the orientation of the body, using a combination of one126933-10102-7362487

[0049] 8 / 5 / 2025

[0050] 6

[0051] or more of accelerometers, gyroscopes, and magnetometers (preferably a six- degree of freedom IMU, or 6- DOF IMU). The IMU 116 measures the kinematic values not via absolute coordinates, but by using acceleration, which is a relative value. Undesirably, over time, the IMU 116 experiences error accumulation. IMU 116 error accumulation refers to a tendency of IMUs 116 to gradually drift from their true orientation or position due to the accumulation of small measurement errors over time. This drift, particularly noticeable with gyroscopes, leads to inaccuracies in velocity and position calculations, impacting the IMU's 116 overall performance.

[0052] The GPS receiver 108 receives GPS signals, which provide the location of a body (aircraft 104) in latitude, longitude, and altitude. The GPS receiver 108 also may provide the accurate time. GPS includes a number of geosynchronous satellites that circle the Earth in precise orbits. Each satellite typically makes a full orbit of Earth every 12 hours.

[0053] As noted, illustrative embodiments mitigate or overcome the challenges associated with GPS-denied environments, which can occur in urban canyons, dense forests, indoor settings, or during GPS failure or jamming scenarios. Traditional aircraft 104 navigation systems heavily rely on GPS signals for positioning and navigation, making them vulnerable in such conditions. To obviate this issue, illustrative embodiments integrate multiple sensors and advanced algorithms to provide robust and reliable navigation capabilities, ensuring navigation success even when GPS signals may not be available and / or inconsistently available.

[0054] Although not explicitly shown in Figure 1, it is generally understood that the aircraft 104 includes a propulsion system. This system typically comprises one or more engines or electric motors, depending on the aircraft's configuration, which may include turbofan, turboprop, distributed electric propulsion units, or rocket engines. These engines are responsible for converting stored energy— such as aviation fuel or battery power— into mechanical force that propels the aircraft forward through the atmosphere.126933-10102-7362487

[0055] 8 / 5 / 2025

[0056] 7

[0057] In addition to the thrust-generating components, the propulsion system is integrated with various aerodynamic control surfaces, such as rudders, ailerons, elevators, and flaps. These surfaces are actuated either mechanically or electronically to manage the aircraft's orientation and stability during different phases of flight, including takeoff, cruising, and landing. The coordination between the propulsion elements and control surfaces is often governed by the navigation control system, which, in addition to compensating for poor GPS signals / reception, also may optimize performance, fuel efficiency, and maneuverability.

[0058] Figure 2 schematically shows more details of the various components of the navigation controller 112, according to illustrative embodiments. It should be reiterated that the representation of Fig. 2 is a simplified representation of various components of the aircraft 104. Those skilled in the art should understand that such the components shown may have other physical and functional components, such as central processing units, other packet processing modules, and various forms of memory. This also should be noted for other figures, such as Figs. 1 and 3-5.

[0059] Each of these components is operatively connected by any conventional interconnect mechanism. Moreover, it should be noted that Fig.

[0060] 2 only schematically shows components. Those skilled in the art should understand that each component can be implemented in a variety of conventional manners, such as by using hardware, software, or a combination of hardware and software, across one or more other functional components. For example, one or more of those components may be implemented using a plurality of microprocessors executing firmware, one or more application specific integrated circuits (i.e., "ASICs") and related software, or a combination of ASICs, discrete electronic components (e.g., transistors), and microprocessors. Accordingly, the representation of the components in a single box is for simplicity purposes only. In fact, in some components may be126933-10102-7362487

[0061] 8 / 5 / 2025

[0062] 8

[0063] distributed across a plurality of different machines— not necessarily within the same housing or chassis.

[0064] The components may include a processor 204, one or more memories 208, and various input / output (I / O) interfaces. The processor 204 may include one or more physical processing devices that execute stored programs in the memory 208. The processor 204 controls navigation of the aircraft 104 based on the stored programs and inputs from the I / O interfaces. The processor 204 may include one or more microprocessors, microcontrollers, state machines, and field programmable gate arrays (FPGAs).

[0065] The memory 208 stores software / firmware applications and data / data structures. To that and other ends, the memory 208 may include one or more applications 212 that the processor executes (i.e., stored programs) and images, maps, and data 216. The data 216 may include historical data for the aircraft 104, including but not limited to previous GPS coordinates, locations where GPS signals were compromised, starting locations, destination locations, waypoints, locations / coordinates of known obstacles, and data structures associated with the reinforcement learning (RL) policy model. The memory 208 may include both volatile and non-volatile memory devices. The applications 212 may include the RL policy model and include associated modules shown in Fig. 3 for navigating the aircraft 104 by simulated GPS coordinates. Processes used by such applications 212 for navigation using simulated GPS coordinates are discussed below.

[0066] One or more high-resolution cameras 224 capture detailed visual data 232 of the surrounding environment. These cameras 224 provide the primary data for visual odometry and landmark recognition by capturing high-resolution images of the aircraft's surroundings. The cameras may include multiple fixed cameras 224 or movable cameras 224. Fixed cameras 224 may face to the front, rear, sides, above, and below the aircraft 104. Movable cameras 224 may be rotated, elevated, or depressed within arcs limited to position on the aircraft 104 and view or movement limitations. The cameras126933-10102-7362487

[0067] 8 / 5 / 2025

[0068] 9

[0069] 224 are calibrated during an initialization process 304 to ensure accurate image capture, with adjustments made for focus, exposure, and other parameters. The captured images provide the primary data for visual odometry and landmark recognition, which are essential for position estimation and navigation.

[0070] As noted, the IMU 116 provides real-time data 236 on the aircraft's movements, which are integrated with the visual data 232 from the camera(s) 224 to improve the accuracy of the position estimate. The IMU 116 is calibrated during the initialization process 304 to ensure accurate measurements and produces real-time data 236 to the processor 204.

[0071] Depth sensors 220 measure distances to objects and surfaces, creating a detailed 3D map of the environment that aids in obstacle detection and avoidance. Depth sensors 220 may include LiDAR, other depth cameras, and / or a fully software-based implementation of depth estimation. Depth sensors 220 provide object distances 228 to the processor 204.

[0072] Environmental sensors 240 measure environmental parameters such as temperature, barometric pressure, daylight / darkness, and wind, and produce environmental data 244.

[0073] Figure 3 schematically shows various data processing engines configured to facilitate navigation with or without a simulated GPS signal in accordance with illustrative embodiments. These embodiments are but examples of some implementations those skilled in the art may use.

[0074] Moreover, the data processing engines may operate as one or more applications 212 stored in the memory 208. Each of these engines is discussed below.

[0075] A visual odometry module 308 is responsible for estimating the aircraft's motion by analyzing sequential image frames captured by the onboard cameras 224, collectively referred to as visual data 232. This module 308 operates by detecting and extracting salient visual features— such as corners, edges, and textured regions— from each frame using feature126933-10102-7362487

[0076] 8 / 5 / 2025

[0077] 10

[0078] detection algorithms like FAST, ORB, or SIFT. Once these key points are identified, the module tracks their displacement across successive frames using techniques such as optical flow or descriptor matching. By measuring how these features shift over time, the system infers the aircraft's relative translation and rotation, effectively reconstructing its trajectory in three-dimensional space.

[0079] To ensure robustness and accuracy, the visual odometry module may incorporate outlier rejection methods (e.g., RANSAC) to filter erroneous matches and apply geometric constraints such as epipolar geometry to validate motion hypotheses. The resulting motion estimates are typically expressed as a transformation matrix or pose vector, which can be integrated over time to yield a continuous estimate of the aircraft's position and orientation. In some implementations, the visual odometry output 328 is fused with data from inertial measurement units (IMUs) 116 or GPS 108 to enhance stability and correct for drift, particularly in environments with poor visual texture or rapid motion.

[0080] A sensor fusion module 312 is configured to continuously receive and process heterogeneous real-time data streams from multiple onboard sensing subsystems, including inertial measurements from the IMU 116, visual data 232 from the cameras 224, and object distance measurements 228 from the depth sensors 220. This module performs multi-modal data integration to generate high-fidelity estimates of the aircraft's position, velocity, and orientation, collectively referred to as position estimates 332. As known by those in the art, the fusion process 312 leverages complementary characteristics of each sensor modality: the IMU 116 provides high-frequency motion data, the cameras 224 contribute rich spatial and contextual information, and the depth sensors 220 offer direct range measurements to nearby surfaces and objects.

[0081] In dynamic or visually degraded environments— such as low-light conditions, rapid maneuvers, or feature-sparse terrain— the IMU's data 236126933-10102-7362487

[0082] 8 / 5 / 2025

[0083] 11

[0084] becomes particularly useful. It supplies continuous acceleration and angular velocity readings that enable dead-reckoning and short-term motion prediction, even when visual tracking is unreliable or temporarily unavailable. The sensor fusion module can apply filtering techniques such as extended Kalman filters (EKF), unscented Kalman filters (UKF), or factor graph optimization to reconcile discrepancies between sensor inputs, correct for drift, and maintain globally consistent navigation estimates. In some embodiments, the module may also incorporate probabilistic models or machine learning-based estimators to adaptively weight sensor contributions based on environmental context and system confidence levels.

[0085] A path planning module 316 is configured to compute and refine an optimal flight route 356 for the aircraft 104 based on a diverse set of input parameters. These inputs include the starting and ending locations 336, which are typically determined pre-flight via a GPS receiver 108 or retrieved from pre-programmed GPS coordinates. The module also ingests map data 340 generated from three-dimensional environmental reconstructions produced by the onboard depth sensors 220. This 3D map provides spatial context and topological detail essential for obstacle detection and avoidance, particularly in cluttered or dynamically changing environments. Additionally, the module utilizes distance data 344 and obstacle data 348, which are derived from object distance measurements 228 captured by the depth sensors 220, to identify and localize potential hazards along the projected route.

[0086] Environmental conditions data 244, sourced from onboard environmental sensors 240 such as temperature sensors, wind sensors, and barometers, is also incorporated into the planning process. This data enables the module to account for external factors that may influence flight dynamics, such as wind shear, thermal gradients, or pressure variations. The path planning algorithm may employ graph-based search techniques (e.g., A*, D*, or RRT), optimization frameworks, or machine learning-based heuristics to determine a route that minimizes cost functions such as travel time, energy126933-10102-7362487

[0087] 8 / 5 / 2025

[0088] 12

[0089] consumption, or risk exposure. The resulting optimal route 356 preferably is not static; it is regularly, periodically, aperiodically, or continuously updated (e.g., in real time) during flight to reflect changes in environmental conditions, obstacle configurations, or mission objectives.

[0090] The computed optimal route 356 includes a sequence of waypoints, each annotated with navigational parameters such as velocity, altitude, and azimuth (bearing) between waypoints. These parameters are dynamically adjusted to ensure compliance with flight constraints and to maintain safe separation from obstacles and restricted zones. In some embodiments, the path planning module 316 may interface with a higher-level mission management system to support adaptive rerouting, contingency handling, and coordination with other autonomous agents or air traffic control systems.

[0091] The flight controller 320 is configured to execute real-time control of the aircraft's propulsion and actuation systems based on navigation commands 368 received from the navigation control module 324. These commands include target velocities, orientations, and control surface deflections necessary to follow the optimal route 356 generated by the path planning module 316. The flight controller 320 interfaces directly with the aircraft's motors and actuators 372, issuing low-level control signals to adjust thrust, pitch, roll, yaw, and other flight dynamics parameters. This closed-loop control ensures that the aircraft 104 adheres to the planned trajectory while maintaining stability, responsiveness, and compliance within flight constraints.

[0092] The navigation control module 324 serves as an intermediary between high-level path planning outputs and low-level flight control execution. It receives GPS-based position estimates 360 and path planning results 364— including the optimal route 356 and associated waypoints— and translates these into actionable navigation commands 368. These commands are dynamically updated to reflect real-time changes in the aircraft's state or environment, such as deviations from the planned path, newly detected obstacles, or shifts in environmental conditions. The module may employ126933-10102-7362487

[0093] 8 / 5 / 2025

[0094] 13

[0095] predictive control algorithms, trajectory tracking controllers, or model predictive control (MPC) frameworks to generate smooth and feasible command sequences that respect the aircraft's kinematic and dynamic limits.

[0096] The ability of the flight controller 320 to accurately interpret and execute the navigation commands 368 facilitates safe and efficient operation of the aircraft 104. In some embodiments, the system may include feedback mechanisms that monitor execution fidelity and report deviations back to the navigation control module 324 for corrective action. This hierarchical control architecture enables robust and adaptive navigation, allowing the aircraft 104 to autonomously complete complex missions in uncertain or dynamic environments.

[0097] In one embodiment, the navigation control module 324 is configured to detect corrupted or spoofed GPS signals by leveraging an RL policy model trained to predict the aircraft's expected position based on recent sensor inputs and control actions. The RL policy model continuously estimates the aircraft's position by analyzing historical state-action trajectories, including IMU data 236, control commands, and environmental context. These predicted positions are then compared in real time to the GPS-derived coordinates. If the deviation between the predicted and received GPS positions exceeds a predefined threshold— indicating a potential anomaly— the navigation control module flags the GPS signal as potentially corrupted or spoofed. This comparison may be further refined using statistical confidence bounds or anomaly detection thresholds learned during training. Upon detection of such discrepancies, the system may initiate corrective actions such as rejecting the GPS input, switching to inertial or visual navigation modes, or triggering alerts for operator review, thereby enhancing the aircraft's resilience to GPS spoofing attacks and ensuring continued navigational integrity.

[0098] The navigation control module 324 also preferably uses factor graphing to estimate positions based on previously determined positions and known or126933-10102-7362487

[0099] 8 / 5 / 2025

[0100] 14

[0101] estimated speeds and courses over time. As known by those in the art, a factor graph is a bipartite graph representing the factorization of a probabilistic function. It expresses how a global function (e.g., the joint probability over positions, speeds, and headings over time) decomposes into a product of simpler "factors" that each depend only on a subset of variables. Factor graphing refers to a graph-based probabilistic modeling technique used primarily in simultaneous localization and mapping (SLAM), sensor fusion, and robotic navigation.

[0102] The GPS estimates 360 are provided by an RL policy model executed by the processor 204 and stored / updated in the memory 208. The RL policy model is shown and described in more detail with respect to Figs. 6 and 7.

[0103] In one embodiment, the system may not include a GPS receiver 108. In another embodiment, the system may include a GPS receiver 108 but the GPS receiver 108 may not provide GPS coordinates because of a malfunction, damage, spoofing, or radio frequency (RF) jamming.

[0104] Figure 4 shows a process 400 of controlling navigation of the aircraft 104 in accordance with illustrative embodiments. It should be noted that this process (as well as those of Figs. 5 and 6) is substantially simplified from a longer process that normally would be used to control aircraft navigation. Accordingly, the process may have additional steps that those skilled in the art likely would use. In addition, some of the steps may be performed in a different order than that shown, or at the same time. Those skilled in the art therefore can modify the process as appropriate. Moreover, as noted herein, the structures noted are but one of a wide variety of different structures that may be used. Those skilled in the art can select the appropriate structures depending upon the application and other constraints. Accordingly, discussion of specific structures is not intended to limit all embodiments.

[0105] The process 400 begins at step 404, in which the aircraft 104 detects inertial information using an onboard IMU 116. As noted, the IMU 116 has one or more accelerometers and gyroscopes configured to measure linear126933-10102-7362487

[0106] 8 / 5 / 2025

[0107] 15

[0108] acceleration and angular velocity along multiple axes. These measurements are used to determine the aircraft's motion state, including its velocity, orientation, and displacement over time. The IMU 116 generates this inertial information as real-time data 236 and transmits it to the processor 204 for further processing.

[0109] In some embodiments, the IMU 116 may also include a magnetometer to assist with heading estimation. The real-time data 236 may be preprocessed by the processor 204 to correct for known sensor biases, temperature drift, or noise artifacts. This inertial information forms the foundational input for downstream modules, including the sensor fusion module 312 and the navigation control module 324, which rely on high-frequency motion data to maintain accurate and continuous navigation, particularly in environments where GPS signals are degraded or unavailable. Upon completion of this step, flow proceeds to step 408.

[0110] At step 408, the processor 204 receives a GPS signal from the GPS receiver 108. The GPS signal includes geographic coordinates— typically expressed in terms of latitude, longitude, and optionally altitude— that represent the current estimated position of the aircraft 104. These coordinates are derived from satellite-based triangulation and may also include metadata such as signal strength, timestamp, and satellite identifiers. The GPS signal is parsed and stored in memory 208 for use by the navigation control module 324 and other downstream modules.

[0111] In some embodiments, the processor 204 may also assess the quality of the received GPS signal by evaluating parameters such as dilution of precision (DOP), number of visible satellites, or signal-to-noise ratio. This quality assessment may be used in subsequent decision step 412 to determine whether the GPS signal is sufficiently reliable for navigation. Upon successful receipt and parsing of the GPS signal, flow proceeds to decision step 412.126933-10102-7362487

[0112] 8 / 5 / 2025

[0113] 16

[0114] At step 412, the processor 204 determines whether the GPS signal received by the GPS receiver 108 is sufficient for reliable navigation. A GPS signal may be considered sufficient if, for example, it is stable, consistent, and within acceptable accuracy thresholds for the aircraft's current operational context. The processor 204 may evaluate signal sufficiency based on a combination of factors, including signal strength, satellite geometry (e.g., dilution of precision), number of satellites in view, and the temporal consistency of the reported coordinates. In some embodiments, the processor 204 may also compare the received GPS coordinates to a predicted position generated by a RL policy model to detect anomalies indicative of spoofing or degradation.

[0115] A GPS signal may be deemed insufficient under a variety of conditions. For example, in urban environments with tall buildings ("urban canyons"), GPS signals may reflect off surfaces, resulting in multipath errors that cause the reported position to jump erratically. In rural or forested areas, signal blockage may occur due to dense canopy cover. In indoor or subterranean environments, GPS signals may be entirely unavailable. Additionally, the signal may be actively spoofed— where a malicious actor transmits false GPS data to mislead the receiver— or jammed, where the signal is overwhelmed by noise. Even in open environments, a weak satellite constellation (e.g., fewer than four satellites in view) or high dilution of precision may render the signal unreliable.

[0116] If the processor 204 determines at step 412 that the GPS signal is insufficient— due to being unavailable, degraded, spoofed, or otherwise unreliable— then the process continues to step 416. At this step, the processor 204 initiates navigation of the aircraft 104 using a simulated GPS signal generated by a RL policy model. The simulated GPS signal is produced by the RL policy model, which may use a combination of real-time inertial data 236 from the IMU 116, historical flight data, velocity estimates, and contextual126933-10102-7362487

[0117] 8 / 5 / 2025

[0118] 17

[0119] environmental inputs 244, as described in greater detail with respect to Figs.

[0120] 3, 5, and 6.

[0121] The RL policy model is pre-trained in a simulated environment to learn a navigation policy that maximizes a cumulative reward function, enabling it to predict position deltas or absolute coordinates in the absence of valid GPS input. The simulated GPS signal 360 is then used by the navigation control module 324 to generate navigation commands 368, which are executed by the flight controller 320 to control the aircraft's motors and actuators 372. This allows the aircraft 104 to continue along the optimal route 356, maintain stability, and avoid obstacles, even in GPS-denied or contested environments. Flow proceeds from this step and terminates if the mission is complete. If the processor 204 determines at step 412 that the GPS signal is sufficient, then the process continues to step 420. At this step, the processor 204 navigates the aircraft 104 using the received GPS signal provided by the GPS receiver 108. The GPS signal includes geographic coordinates (e.g., latitude, longitude, and altitude) that are used by the navigation control module 324 to generate navigation commands 368. These commands are then executed by the flight controller 320 to control the motors and actuators 372, effectively guiding the aircraft 104 along the optimal route 356.

[0122] Steps 404 through 420 may be repeated at regular intervals, such as every 500 milliseconds, or upon receipt of each GPS coordinate update.

[0123] During flight, if the processor 204 detects that the GPS signal has become unstable, corrupted, or spoofed, the system may transition to step 416, where navigation is performed using a simulated GPS signal generated by the RL policy model. If reliable GPS coordinates are subsequently restored, the processor 204 may resume navigation using the received GPS signal at step 420. In some embodiments, the aircraft 104 may continue using simulated GPS coordinates even after GPS recovery, depending on mission parameters or signal confidence thresholds.126933-10102-7362487

[0124] 8 / 5 / 2025

[0125] 18

[0126] Figure 5 shows details of a related process 500 of navigating the aircraft 104 of Figure 1 in accordance with illustrative embodiments. Although the aircraft 104 may navigate using stable GPS signals, in the absence of stable signals the processor 204 executes the following steps to successfully navigate the aircraft 104 to a destination. Flow begins at step 504.

[0127] At step 504, the processor 204 generates navigational commands 368. Navigational commands 368 may include new velocity, altitude, and bearing. Flow proceeds to steps 508 and 512.

[0128] At step 508, the processor 204 (specifically, the flight controller 320) receives the navigation commands 368 and directs the flight controller 320 to adjust motor and actuator controls 372 to account for changes in velocity, altitude, and bearing. The new navigation commands 368 are compared to the current navigation commands 368. Differences between them result in changes to the motor and actuator controls 372. Flow proceeds to steps 516 and 520.

[0129] At step 512, the processor converts GPS estimates and path planning results to match the updated navigation commands 368.

[0130] At step 516, the processor 204 executes real-time adjustments. The flight controller 320 translates the new navigation commands 368 into updated motor and actuator controls 372 and saves the new navigation commands 368 (and possibly the changes to motor and actuator controls 372) in the memory 208. The real-time adjustments 516 may take into account collision avoidance with detected objects (trees, power lines, etc.). Flow proceeds to steps 524 and 528.

[0131] At step 520, the processor 204 adjusts speed, direction, and altitude by controlling the motors and actuators 372.

[0132] At step 524, the processor 204 monitors the aircraft 104. The processor 204 may perform internal testing to verify the integrity and performance of computing resources, interfaces, sensors, motors, and actuators. Flow proceeds to steps 532 and 536.126933-10102-7362487

[0133] 8 / 5 / 2025

[0134] 19

[0135] At step 528, the processor 204 continuously updates the current position of the aircraft 104. For example, the processor 204 may update the position based on the latest navigation commands 368, collision avoidance, and the time since the last update to the navigation commands 368. Flow proceeds to step 540.

[0136] At step 532, the processor 204 makes necessary adjustments based on results of monitoring the aircraft 104. For example, if a sensor is producing varying readings, it may be helpful to average readings over a most recent time period to "smooth out" the variations and avoid abrupt changes to motors and actuators 372. Flow proceeds back to step 524 to continue to monitor the aircraft 104.

[0137] At step 536, the processor 204 continuously monitors performance of the aircraft 104. Performance may include reaching a waypoint at a specific time, achieving a desired speed or altitude, or other performance measure. If the aircraft 104 is meeting desired performance or not meeting desired performance, the results are stored and adjusted to by the processor 204. For example, if the aircraft is unable to reach a waypoint or the destination at a desired time due to strong headwinds, the processor 204 may adjust an expected arrival time in the memory 208 and / or increase a speed of the aircraft 104 to compensate. Flow proceeds back to step 524 to continue to monitor the aircraft 104.

[0138] At step 540, the processor 204 adjusts a trajectory of the aircraft 104 to avoid obstacles. In one embodiment, obstacles may be sensed by one or more cameras 224 coupled with image recognition and obstacle detection. The processor 204 may execute an image recognition application 212 in the memory 208 and identify one or more objects. The processor 204 may determine a current trajectory may intercept one or more objects and calculates a temporary change in speed, altitude, and bearing (or a series of speed / altitude / bearing commands at different times) to avoid the object and return to the desired navigation commands 368.126933-10102-7362487

[0139] 8 / 5 / 2025

[0140] 20

[0141] Figure 6 shows a process 600 showing more details of the execution of the RL policy model making real-time inferences during flight in accordance with illustrative embodiments. Specifically, the RL policy model generates simulated global positioning system (GPS) signals, preferably in real time, particularly in environments where GPS signals are unavailable, degraded, or unreliable (e.g., urban canyons, tunnels, or during jamming or spoofing events). As noted, when executing during flight, the RL policy model collects input data representing the current state of the aircraft 104, including realtime data 236 (e.g., acceleration, angular rate) from the IMU 116, historical or recent position coordinates, aircraft velocity estimates, and environmental / contextual info (e.g., GPS signal strength, time since last GPS lock).

[0142] The process begins at step 604, in which the processor 204 collects input data. The input data may be sourced from a variety of onboard systems, including the GPS receiver 108, the IMU 116, one or more cameras 224, depth sensors 220, environmental sensors 240, and aircraft-specific sensors such as propulsion sensors, fuel level sensors, and control surface sensors. Flow proceeds to step 608.

[0143] At step 608, the processor 204 preprocesses raw sensor data. The raw sensor data is preprocessed or transformed into a set of features compatible with the trained policy model. This could involve noise filtering, normalization, and aggregation of temporal data (e.g., through recurrent units or sliding windows). Flow proceeds to step 612.

[0144] At step 612, the processor 204 determines a policy of a best navigation action. The RL policy model evaluates the current state using a trained policy network (e.g., a deep neural network) to predict the best navigation action or the simulated GPS coordinates. In one embodiment, the RL policy model may include a neural network with multiple layers that maps observed aircraft 104 states to navigation actions or positional estimates. If the policy is value-126933-10102-7362487

[0145] 8 / 5 / 2025

[0146] 21

[0147] based (e.g., Q-learning), the RL policy model selects an action with highest predicted value.

[0148] At step 616, the processor 204 predicts simulated GPS coordinates. The action or output from the RL policy model is converted into a simulated GPS signal. Depending on implementation, this may involve generating coordinate estimates (latitude, longitude, altitude), formatting an output to match NMEA strings or other GPS-compatible formats, or emitting pseudosignals or positioning data compatible with the aircraft's navigation system.

[0149] At step 620, the processor 204 combines the simulated GPS coordinates with IMU real-time data 236 to obtain navigation commands for the aircraft 104. The simulated GPS signal 360 is fed into the navigation controller 324, which uses it in combination with IMU real-time data 236 to continue navigating the aircraft 104.

[0150] At step 624, the processor 204 navigates the aircraft 104 using the navigation commands 368. Navigation commands 368 are converted into motor and actuator controls 372 by the flight controller 320. Flow proceeds to step 604 to collect more input data. Although represented serially in the flowchart, it should be understood that input data is collected constantly, and the other steps of the process also execute constantly and in parallel.

[0151] In more advanced systems, the RL policy model may continue to refine its policy during operation by receiving a reward signal based on how well the navigation matches ground truth or desired path, updating the policy parameters incrementally (e.g., using gradient descent or experience replay), and logging experience data for future offline retraining. As noted above at step 616, the exemplary RL policy model executes a sequence of steps to generate a simulated or synthetic GPS signal. To that end, in illustrative embodiments, the system uses state observation, in which the model receives input signals representing the current state of the aircraft 104. These signals may include real-time inertial data 236 (e.g., linear acceleration and angular126933-10102-7362487

[0152] 8 / 5 / 2025

[0153] 22

[0154] velocity), velocity estimates, historical position coordinates, elapsed time since the last GPS acquisition, and environmental data 244.

[0155] Once collected, the input signals are preprocessed into a standardized feature vector. This preprocessing may involve normalization, filtering, and dimensionality reduction to ensure compatibility with the input structure of the trained model. The refined feature vector is then passed to the RL policy model for inference. Based on the current state, the model predicts a positional output, which may correspond to estimated GPS coordinates (latitude, longitude, altitude) or a delta position relative to the last known location.

[0156] The inferred position is formatted into a simulated GPS signal. In some embodiments, this signal complies with standard GPS output protocols, such as NMEA 0183, allowing it to be seamlessly integrated into downstream systems like the navigation control module 324. The navigation system treats this simulated signal as if it were a live GPS input, enabling continued operation in GPS-denied or contested environments.

[0157] In certain implementations, the RL policy model may also support feedback and adaptation. For example, the model may receive a reward signal based on navigation accuracy, allowing for online learning during flight. Alternatively, flight data may be logged for offline retraining to improve future performance.

[0158] The simulated GPS signal 360 is ultimately provided to the navigation control module 324 as noted above (e.g., see step 620). When the received GPS signal falls below a reliability threshold, the system transitions to using the simulated GPS output to guide the aircraft 104. This approach ensures robust and uninterrupted navigation, even in the presence of GPS signal degradation, spoofing, or complete loss.

[0159] Figure 7 shows a process 700 for training the RL policy model in accordance with illustrative embodiments. At a high level, the training process for the RL policy model begins with knowledge of the start and end points of126933-10102-7362487

[0160] 8 / 5 / 2025

[0161] 23

[0162] a representative flight path. The model is trained through repeated simulations under varying environmental conditions to develop a navigation policy capable of inferring position estimates in the absence of reliable GPS. In some embodiments, the RL policy model is implemented using machine learning techniques such as Proximal Policy Optimization (PPO), Deep Q-Learning, or Actor-Critic methods, and is trained to predict GPS-like outputs based on historical flight data, IMU 116 signals, and environmental inputs 244.

[0163] At step 704, the system defines the training environment and mission objectives. This includes specifying the geographic region, terrain type, and environmental conditions (e.g., wind, lighting, GPS availability) that the aircraft 104 may encounter. The training environment also includes the start and end points of the flight path, as well as any constraints or goals such as minimizing energy consumption, avoiding obstacles, or maintaining altitude. These parameters form the basis for constructing a simulated environment in which the RL model will be trained.

[0164] At step 708, the system generates synthetic flight scenarios using a high-fidelity simulation engine. These scenarios emulate real-world conditions and sensor outputs, including IMU data 236 (e.g., acceleration and angular velocity), GPS signals (including degraded or spoofed signals), environmental sensor data 244 (e.g., barometric pressure, wind speed), and optionally visual and depth data from simulated cameras and LiDAR. The aircraft 104 is virtually flown through these scenarios using randomized or scripted control policies, and the resulting state-action trajectories are logged to form the training dataset.

[0165] At step 712, the system preprocesses the collected data into structured input features suitable for training. This may include normalization of sensor values, temporal aggregation using sliding windows or recurrent encoders, and dimensionality reduction for high-resolution visual data. The system also encodes contextual information such as time since last GPS lock, signal126933-10102-7362487

[0166] 8 / 5 / 2025

[0167] 24

[0168] strength, and environmental variability. The goal is to produce a consistent and information-rich input representation that captures both short-term dynamics and long-term context.

[0169] At step 716, the RL policy model is trained using the preprocessed data. The model receives the encoded state as input and outputs either a navigation action (e.g., velocity vector) or a simulated GPS coordinate. The training process optimizes a cumulative reward function that may include terms for positional accuracy, trajectory smoothness, obstacle avoidance, and energy efficiency. The model is updated iteratively using RL algorithms such as PPO, Deep Q-Learning, or Actor-Critic, and is evaluated periodically on held-out validation scenarios to assess generalization and robustness.

[0170] At step 720, the trained model is validated and calibrated. The system evaluates the model's performance in previously unseen environments, measuring metrics such as drift over time, deviation from ground truth, and resilience to sensor noise or GPS spoofing. If necessary, the model is finetuned or retrained to improve stability and accuracy. Calibration may also involve adjusting reward weights or pruning unstable behaviors. As part of the validation, the model is serialized and stored in memory 208 for deployment.

[0171] At step 724, the trained RL policy model is packaged for onboard inference. This includes exporting the model in a format compatible with the aircraft's processor 204 (e.g., TensorFlow Lite, ONNX), integrating it with the navigation control module 324, and ensuring compatibility with real-time data streams 236 from IMU 116, environmental sensors 240, and other onboard systems. The model is configured to generate simulated GPS signals 360 in real time during flight, particularly when actual GPS signals are unavailable, degraded, or spoofed.

[0172] This training process 700 enables the RL policy model to operate autonomously (e.g., an autonomous navigation control module 324) and126933-10102-7362487

[0173] 8 / 5 / 2025

[0174] 25

[0175] reliably in GPS-denied environments, providing the aircraft 104 with a robust alternative to conventional satellite-based navigation.

[0176] Exemplary Use:

[0177] Consider an unmanned aerial vehicle (UAV) tasked with flying from a launch point at a coastal base (Point A) to a designated landing zone located 50 kilometers inland (Point B), traversing a route that includes urban environments, forested terrain, and a tunnel-like canyon known for GPS signal degradation.

[0178] Prior to deployment, the RL policy model is trained in a simulated environment that replicates the mission route, including terrain features, expected weather conditions, and known electromagnetic interference zones. The model is trained to predict positional deltas and generate synthetic GPS signals based on IMU data, velocity estimates, and environmental context.

[0179] Once deployed, the UAV begins its mission using standard GPS and IMU data. As it enters the canyon, the GPS signal becomes unreliable. The navigation controller detects the degradation and seamlessly transitions to using the simulated GPS signal generated by the onboard RL policy model. The model, having been trained on similar conditions, produces accurate positional estimates that allow the UAV to maintain its trajectory and avoid obstacles.

[0180] Upon exiting the canyon and reacquiring a reliable GPS signal, the navigation controller resumes standard GPS-based navigation. The UAV continues to Point B, completing the mission with uninterrupted navigation despite the temporary loss of GPS.

[0181] Illustrative embodiments improve the state of the art in aircraft navigation by enabling more reliable operation in GPS-denied or spoofed environments. Unlike conventional systems that rely solely on GPS and Kalman-filtered IMU data, various embodiments use a reinforcement learning126933-10102-7362487

[0182] 8 / 5 / 2025

[0183] 26

[0184] policy model trained in mission-specific simulated environments to generate synthetic GPS signals in real time. This approach allows the aircraft 104 to maintain accurate navigation without requiring external positioning updates, enhancing resilience, autonomy, and operational flexibility. By integrating factor graphing and adaptive learning mechanisms, the system offers effective drift correction and robustness compared to traditional sensor fusion techniques, making it particularly valuable for unmanned aerial vehicles 104 operating in contested or degraded signal conditions.

[0185] Various embodiments of the invention may be implemented at least in part in any conventional computer programming language. For example, some embodiments may be implemented in a procedural programming language (e.g., "C"), or in an object oriented programming language (e.g., "C++"). Other embodiments of the invention may be implemented as a preconfigured, stand-alone hardware element and / or as preprogrammed hardware elements (e.g., application specific integrated circuits, FPGAs, and digital signal processors), or other related components.

[0186] In an alternative embodiment, the disclosed apparatus, and methods (e.g., see the various flow charts described above) may be implemented as a computer program product for use with a computer system. Such implementation may include a series of computer instructions fixed either on a tangible, non-transitory medium, such as a computer readable medium (e.g., a diskette, CD-ROM, ROM, or fixed disk). The series of computer instructions can embody all or part of the functionality previously described herein with respect to the system.

[0187] Those skilled in the art should appreciate that such computer instructions can be written in a number of programming languages for use with many computer architectures or operating systems. Furthermore, such instructions may be stored in any memory device, such as semiconductor, magnetic, optical, or other memory devices, and may be transmitted using126933-10102-7362487

[0188] 8 / 5 / 2025

[0189] 27

[0190] any communications technology, such as optical, infrared, microwave, or other transmission technologies.

[0191] Among other ways, such a computer program product may be distributed as a removable medium with accompanying printed or electronic documentation (e.g., shrink wrapped software), preloaded with a computer system (e.g., on system ROM or fixed disk), or distributed from a server or electronic bulletin board over the network (e.g., the Internet or World Wide Web). In fact, some embodiments may be implemented in a software-as-a-service model ("SAAS") or cloud computing model. Of course, some embodiments of the invention may be implemented as a combination of both software (e.g., a computer program product) and hardware. Still other embodiments of the invention are implemented as entirely hardware, or entirely software.

[0192] The embodiments of the invention described above are intended to be merely exemplary; numerous variations and modifications will be apparent to those skilled in the art. Such variations and modifications are intended to be within the scope of the present invention as defined by any of the appended claims.

Claims

126933-10102-73624878 / 5 / 202528What is claimed is:

1. An aircraft comprising:an inertial measurement unit (IMU) configured to detect real-time inertial information relating to the aircraft;a receiver configured to receive global positioning system (GPS) signals; anda navigation controller operatively coupled with the IMU and the receiver, the navigation controller configured to navigate the aircraft as a function of the inertial information received from the IMU and a GPS signal received by the receiver,the navigation controller further configured to navigate as a function of a simulated GPS signal produced by a reinforcement learning policy model.

2. The aircraft of claim 1, further comprising a propulsion system operatively coupled with the navigation controller, the propulsion system being controlled by the navigation controller.

3. The aircraft of claim 1, wherein the policy model comprises an artificial intelligence model pre-trained in a simulated environment relating to a prescribed environment of flight of the aircraft, the policy model pre-trained to learn a navigation policy that maximizes a cumulative reward function.

4. The aircraft of claim 3, wherein the policy model is configured to predict relative position changes of the aircraft as a function of the inertial information.

5. The aircraft of claim 1, wherein the navigation controller is configured to use the simulated GPS signal when the GPS signal is detected to be unavailable, spoofed, or corrupted.126933-10102-73624878 / 5 / 2025296. The aircraft of claim 1, wherein the navigation controller includes an autonomous navigation controller.

7. The aircraft of claim 1, further comprising a body forming one of a drone, plane, helicopter, or airship.

8. The aircraft of claim 1, further wherein the navigation controller uses factor graphing to estimate positions based on previously determined positions and known or estimated speeds and courses over time.

9. The aircraft of claim 1, wherein the navigation controller, receiver, and IMU are onboard the aircraft.

10. The aircraft of claim 1, wherein the navigation controller is configured to detect corrupted or spoofed GPS signals by comparing received GPS signals to a predicted position produced by the model.

11. The aircraft of claim 1, wherein the navigation controller executes sensor fusion to continuously integrate the inertial information from the IMU to maintain ongoing position updates during flight.

12. A method of navigating an aircraft, the method comprising:detecting, using an inertial measurement unit (IMU) onboard the aircraft, real-time inertial information relating to the aircraft;126933-10102-73624878 / 5 / 202530receiving a global positioning system (GPS) signal from an external source;determining whether the received GPS signal is sufficient for navigation;navigating the aircraft as a function of the inertial information received from the IMU and the received GPS signal when the received GPS signal is determined to be sufficient for navigation; andnavigating the aircraft as a function of the inertial information received from the IMU and a simulated GPS signal when the received GPS signal is determined to be insufficient for navigation,the simulated GPS signal being produced by a reinforcement learning policy model,the reinforcement learning policy model executing onboard the aircraft.

13. The method of claim 12, wherein said determining comprises determining whether the received GPS signal is corrupted, spoofed, unavailable, or weaker than a prescribed amount sufficient for navigation.

14. The method of claim 12, wherein the policy model comprises an artificial intelligence model pre-trained in a simulated environment relating to a prescribed environment of flight of the aircraft, the policy model pre-trained to learn a navigation policy that maximizes a cumulative reward function.

15. The method of claim 14, wherein the policy model is configured to predict relative position changes of the aircraft as a function of the inertial information.

16. The method of claim 12, the aircraft comprise one of a drone, plane, helicopter, or airship.126933-10102-73624878 / 5 / 20253117. The method of claim 12, further comprising using factor graphing to estimate positions based on previously determined positions and known or estimated speeds and courses over time.

18. The method of claim 12, further comprising executing sensor fusion to continuously integrate the inertial information from the IMU to maintain ongoing position updates during flight.

19. The method of claim 12, wherein the aircraft is navigating between two known geographic points, the policy model being trained using information relating to the two geographic points and the region between the two geographic points.

20. A computer program product for use on a computer system for navigating an aircraft, the computer program product comprising a tangible, non-transient computer usable medium having computer readable program code thereon, the computer readable program code comprising:program code for receiving, from an inertial measurement unit (IMU) onboard the aircraft, real-time inertial information relating to the aircraft; program code for receiving a global positioning system (GPS) signal from an external source;program code for determining whether the received GPS signal is sufficient for navigation;program code for implementing and executing a reinforcement learning policy model onboard the aircraft to produce a simulated GPS signal;program code for controlling navigation of the aircraft as a function of the inertial information received from the IMU and the received GPS signal when the received GPS signal is determined to be sufficient for navigation; and126933-10102-73624878 / 5 / 202532program code for controlling navigation of the aircraft as a function of the inertial information received from the IMU and the simulated GPS signal when the received GPS signal is determined to be insufficient for navigation.

21. The computer program product of claim 20, further comprising program code for navigating the aircraft as a function of the simulated GPS signal.

22. The computer program product of claim 20, wherein the policy model comprises program code implementing an artificial intelligence model pretrained in a simulated environment relating to a prescribed environment of flight of the aircraft, the policy model pre-trained to learn a navigation policy that maximizes a cumulative reward function.

23. The computer program product of claim 20, wherein the policy model is configured to predict relative position changes of the aircraft as a function of the inertial information.

Images