Anti-theft system for autonomous work vehicle

Abstract

An autonomous work vehicle includes a controller that determines that the vehicle has been manually stopped, and in response thereto, starts a delay timer that runs for a predetermined time. During the predetermined time, a motion measurement unit of the autonomous work vehicle is monitored to detect a motion event that indicates a movement of the autonomous work vehicle. In response to the event, an alert that the motion event has occurred is wirelessly communicated.

Description

ANTI-THEFT SYSTEM FOR AUTONOMOUS WORK VEHICLERELATED PATENT DOCUMENTS

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 619,082, filed on January 9, 2024, which is incorporated herein by reference in its entirety.SUMMARY

[0002] The present disclosure is directed to an anti-theft system for autonomous / robotic ground care machine. In one embodiment, a method involves determining that an autonomous work vehicle has been manually stopped, and in response thereto, starting a delay timer that runs for a predetermined time. During the predetermined time, a motion measurement unit of the autonomous work vehicle is monitored to detect a motion event that indicates a movement of the autonomous work vehicle. In response to the event, an alert that the motion event has occurred is wirelessly communicated.

[0003] In another embodiment, a method is performed while an autonomous work vehicle is performing autonomous work in a work region. The method involves monitoring a gyroscope of an autonomous work vehicle for an attitude profile of the autonomous work vehicle that does not conform to an autonomous work profile of the autonomous work vehicle. In response to detecting the non-conforming attitude profile, the autonomous work vehicle wirelessly communicates that a non-conformal attitude profile has been detected

[0004] In another embodiment, a method involves receiving two or more time varying signals from respective two or more sensors of an autonomous work vehicle that is located in a work region. A first sensor of the two or more sensors comprises a motion sensor. The two or more time varying signals are input into a machine learning model that classifies patterns of the two or more time varying signals into a first class that indicates a theft of the autonomous work vehicle is occurring and a second class that indicates a theft of the autonomous work vehicle is not occurring. In response to detecting a first event that corresponds to the first class, the vehicle wirelessly communicates that the event has occurred. The alert is not communicated responsive to a second event that corresponds to the second class.

[0005] These and other features and aspects of various embodiments may be understood in view of the following detailed discussion and accompanying drawings.BRIEF DESCRIPTION OF THE DRAWINGS

[0006] The discussion below makes reference to the following figures, wherein the same reference number may be used to identify the similar / same component in multiple figures. The drawings are not necessarily to scale.

[0007] FIG. l is a block diagram of an autonomous vehicle according to various example embodiments;

[0008] FIG. 2 is a block diagram of a machine learning architecture according to an example embodiment;

[0009] FIG. 3 is a flow diagram of a theft-prevention scheme according to an example embodiment;

[0010] FIGS. 4-6 are flowcharts showing methods according to example embodiments.DETAILED DESCRIPTION

[0011] In the following detailed description of illustrative embodiments, reference is made to the accompanying figures of the drawing which form a part hereof. It is to be understood that other equivalent embodiments, which may not be described and / or illustrated herein, are also contemplated.

[0012] The present disclosure relates generally to ground care machines, which may be variously referred to herein as ground care vehicles, ground maintenance machines, ground maintenance vehicles, and the like. Ground care machines, such as lawn and garden machines, are known for performing a variety of tasks. For instance, powered lawn mowers are used by both homeowners and professionals alike to maintain grass areas within a property or yard. The same or different machines may be used for maintenance on the turf areas (and sometimes away from turf), performing operations such as debriscollection, spraying, dethatching, edging, rolling, towing, snow / ice treatment and removal, etc.

[0013] Embodiments of the present disclosure relate to features of an autonomous work vehicle, such as an autonomous ground maintenance machine. Generally, an autonomous machine may perform a defined set of operations without human input. The autonomous inputs are generated by a computer processor and cause and / or effect a physical action performed by the machine. One example of autonomous operation is autonomous navigation, where the machine can maneuver around a work region without user input, or with minimal user input (e.g., initial placement and initiating a start command). Robotic work vehicles do not need to be fully autonomous, e.g., may be remote controlled by a human, however some actions may still be autonomously guided or halted based on sensor inputs to the machine. For example, a bump sensor may detect an obstacle that may not be visible by the operator, and the machine can autonomously take an action based on this, e.g., stop moving, alert the operator, back up, re-route, etc.

[0014] Autonomous work vehicles are often designed for unattended operation in an outdoor environment. Autonomous vehicles can be conspicuous in such environments, e.g., they may be brightly colored for easy visibility, and the machinery is often audible from a distance. As such, an autonomous work vehicle is an attractive target for thieves, vandals, and the like. The present disclosure relates to devices and methods to detect theft or other illicit activity affecting an autonomous work vehicle. If such activity is detected, the vehicle can take actions in an attempt to mitigate the effects of the activity.

[0015] Theft prevention means have been developed for vehicles such as cars and bicycles, and some of those techniques, such as locks, alarms, location tracking, etc., can be adapted for similar use with an autonomous work vehicle. Unlike a piloted vehicle, autonomous work vehicles may be susceptible to being taken or damaged while in operation, as there is often no person around to discourage such behavior. On the other head, autonomous vehicles may have a sophisticated suite of sensors that monitor the vehicle’s location and orientation for navigation purposes. These sensors can also be used to detect unusual movements that are indicative of a theft or some other unwanted interactions in the work environment.

[0016] One issue that is sometimes seen with theft prevention systems (e.g., car alarms) is false alarms that are set off by innocuous activities (e.g., being bumped, loud noises nearby). False alarms make users less likely to employ or have confidence in theft prevention systems. An automotive theft alarm may use simple motion or sound detector triggers, as the most likely scenario for theft is when the vehicle is parked and so any motion may assumed to be the result of a break in or the like. Such sensors would be less effective use in a moving autonomous vehicle, as sounds and motion are to be expected while working. A system is described herein that can sense events that would not be expected to occur during normal operation of the autonomous work vehicle. In some embodiments, machine learning can be used to train a model that can make distinctions between normal and abnormal sensor readings during operation of an autonomous work vehicle.

[0017] In FIG. 1, a diagram shows an autonomous work vehicle 100 according to one or more embodiments. The autonomous work vehicle 100 includes a chassis 104 with a motor coupled to drive wheels 101 that propels the work vehicle 100. A shroud 102 or other body covering encloses a top part the chassis 104. In some embodiments, the vehicle 100 is steered by differential rotation of the drive wheels 101 such that front wheels 103 may freely rotate, e.g., caster wheels. In other embodiments, the front wheels 103 that can be used for steering, e.g., actively controlled. A work implement (not seen in this view) performs work on or near the ground such as cutting, scraping, spraying, aerating, etc. In this example, the work implement may include a rotary, turf-cutting, blade underneath the chassis 104 that faces the ground and is driven by a rotating electric motor.

[0018] Block 106 represents electrical and electronic components of the autonomous work vehicle 100. Generally, the autonomous work vehicle 100 includes power circuitry 108, a user interface 110, a navigation system 112, and an anti -theft system 114. While these functional systems are illustrated as separate modules, they may be integrated into a common hardware architecture, e.g., a system-on-a-chip, sensor interfaces, sensors. The modules may also share peripheral devices, such as a network interface, persistent data storage, etc.

[0019] The power circuitry 108 may include components such as main power battery 115, backup or processing subsystem batteries, charging circuitry, powerconditioning circuitry, and power monitoring. While the drive system of the autonomous work vehicle 100 may be powered by a source other than a battery (e.g., internal combustion engine, fuel cell, photovoltaic cells) generally at least one battery 115 may be used, e g., to provide a reliable source of power for controller electronics even when motive power is unavailable. A main power switch 116 is shown that connects and disconnects the battery 115 from the rest of the vehicle 100, e.g., from main power distribution bus or line. The main power switch 116 may be used for a number of purposes, e.g., to facilitate maintenance, reboot the controller, etc.

[0020] The user interface 110 includes an external shutoff switch 118. In some embodiments, the external shutoff switch 118 may disconnect power similar to the power switch 116 (e.g., may be the same switch). In other cases, the external shutoff switch 118 may leave the battery 115 connected to the main power distribution , but will disconnect motors, activate brakes, and the like to stop the vehicle from moving as well as stopping a work implement such as a blade. In this way, the external shutoff switch 118 can stop the autonomous work vehicle 100 while allowing the main controller circuits to continue running. The shutoff switch 118 allows a bystander to readily stop the vehicle 100, e.g., if it appears that is may cause damage to itself or something else. Accordingly, the external shutoff switch 118 is generally displayed in a prominent and easily accessible location, e.g., top and center on the shroud 102. In contrast, the power switch 116 (if provided separately) may be in a less accessible location, e.g., under the shroud 102.

[0021] Communications modules 120, 122 are shown under both the user interface 110 and navigation system 112. The user interface communications module 120 may interact locally with a user device such as a mobile phone, laptop computer, and may use local networking protocols such as Bluetooth® and WiFi®. Generally, the user interface communications module 120 can be used for vehicle setup / training, remote control, troubleshooting, etc.

[0022] The navigation communications module 122 may also communicate with a local device and / or a remote computing entity 124, such as an Internet-based security service. In the latter case, the communications module 122 may communicate via wide- area network infrastructure such as a cellular network base station 127. Generally, the navigation communications module 122 may access map data, user settings, updates, etc.A separate geolocation module 123 is shown that also engages in wireless communications, e.g., receives wireless signals that assist in geolocation of the autonomous work vehicle 100 via a global navigation satellite system (GNSS), real-time kinematics (RTK), and the like.[0023J The navigation system 112 can also utilize other sensors used for sensing movement, which can be combined with other sensor data for navigation. This is represented by motion measurement unit 125, which includes and / or interfaces with sensors such as accelerometers and gyroscopes, and is often packaged together into a unit described as an inertial measurement unit (IMU). The motion measurement unit 125 can assist in navigation using such techniques as inertial navigation, in which the acceleration over time is integrated to estimate a velocity vector, which can be used to estimate a change in position from a known start point. A wheel encoder 143 can also provide an estimate in distance traversed from a known point. The motion measurement unit 125 can have a number of other uses besides navigation, such as identifying vehicle tilt that exceeds a limit, terrain mapping, etc. For example, a three-axis gyroscope can provide time varying signals representative of different angles of rotation relative to the earth, e.g., yaw, pitch, and roll of the vehicle 100. Other sensors that can measure angles / tilt include a mechanical tilt sensor or switch (e.g., ball switch, liquid switch).

[0024] Other sensors may also be included that detect relative motion between machine parts, such as a lift sensor 144 that detects a displacement between the shroud 102 and the chassis 104. The easiest way to lift the vehicle 100 in some embodiments is by lifting it by the shroud 102, and therefore this could be the most immediate indicator that the machine is being moved. A wheel encoder 143 also operates in this way, e.g., detecting rotation of a rotating part such as an axle relative to a non-rotating part of the vehicle frame. This rotation can be converted to a traversal distance based on known parameters such as wheel diameter, differences between left and right speeds, etc. Wheel encoder measurements can have errors due to, among other things, wheel slippage, and so wheel encoder measurements are often relied upon over relatively short distances and checked against other distance or location sensors.

[0025] The navigation system 112 may also have a visual navigation module 126 that process images from cameras 140 mounted on the autonomous work vehicle 100. Thevisual navigation module 126 can recognize fixed objects (e.g., walls, sidewalks), determine a relative distance and bearing to the objects based on the size and location in the image, and determine a location based on the relative distance and bearing. The visual navigation module 126 may be configured to recognize other non-fixed objects, such as people and animals, in order to manage proximity to the objects while working. A description of example vision-based navigation for autonomous work vehicles can be found in U.S. Patent 11,334,082, dated May 17, 2022, which is incorporated herein by reference.

[0026] The anti -theft system 114 gathers data from other functional modules and sensors to detect events that are outside what is expected in normal operation, and therefore the events may be due to a theft attempt or other illicit activity. Because the autonomous work vehicle 100 operates in a dynamic environment, and theft (or other malicious events) may occur while working, the detection of theft-type events can more challenging than with a static (e.g., parked) vehicle. Two functional modules 130, 132 are shown that deal with specific use cases. A third functional module, event handler module 134, can be used to set the thresholds of suspected events, and define actions to deal with events that meet the thresholds.

[0027] A power shutoff monitoring module 130 works in a scenario where, if the autonomous work vehicle was in operation, that a thief might first manually stop the vehicle 100, by using one or both of the power switch 116 and shutoff switch 118. Thereafter, the thief would attempt to move the vehicle, e.g., pick it up, push it, drag it, tip it, etc. In some embodiments, the module 130 will determine that the work vehicle 100 has been manually stopped, and in response thereto, start a delay timer that runs for a predetermined time, e.g., 10 seconds to a minute. During the predetermined time, the motion measurement unit 125 is monitored to detect a motion event that indicates a movement of the work vehicle 100. In response to the event, the vehicle 100, e.g., via the event handler 134 and communications module 120, wirelessly communicates an alert to a user of the work vehicle 100 that the motion event has occurred. This could be done directly, e.g., via a text message, or via an Internet monitoring service 138. Presumably, the user would know about the delay timer, and so would wait at least until the delay timer has expired before disturbing the vehicle.

[0028] Other alerts could be generated in response to the event, such as sounding an alarm, flashing vehicle lights, locking vehicle wheels, initiating location tracking, etc. In some embodiments, one or more of the cameras 140 could record and communicate images, e.g., images of a human detected via visual machine learning. Other cameras (e.g., camera 145) away from the vehicle could also receive the alert message. The alert could be sent to at least one camera located on at least one of base station used for charging the autonomous work vehicle and / or a building or vehicle associated with the work region. The event and resulting alert triggering the at least one camera to record an image of the work region in response. This could capture other details away from the work vehicle, such as license numbers of nearby parked cars. The cameras 140, 145 may operate in any spectral range, including infrared, visible, and ultraviolet. Having infrared detection may also be useful, for example, for detection of theft events, e.g., identifying humans as described elsewhere herein.

[0029] The motion event detected by the module 130 may include a change in rotation along a ground-parallel axis that exceeds a threshold, which can be measured by a gyroscope of the motion measurement unit 125 or a separate gyroscope. The change in rotation may be measured relative to a rotation (e.g., due to ground slope) that was measured just before the vehicle stopped. This reflects a use case where the thief would lift the machine from one end. Other motion events that may occur includes a vertical translation relative to ground and / or any significant movement from rest parallel to the ground, which can be measured by one or more of the motion measurement unit 125, the geolocation module 123, and the visual navigation module 126.

[0030] In cases where it is the user or another authorized person who stops the vehicle, it may be advantageous to suppress alerts if a motion event occurs even if the delay timer is used but has not expired. For example, a user may use stop the machine if it appears stuck on an obstacle, and then nudge the vehicle to free it from the obstacle without wanting to wait. The work vehicle 100 may include a proximity detector 136 such as a radio-frequency identification (RFID) reader, Bluetooth or WiFi radio, that can detect proximity of a device 142 that was previously registered as belonging to the user. The proximity detector 136 may also include non-user interface proximity sensors, such as touch sensors and distance sensors used with navigation and anti-theft systems 112, 114.Such proximity sensors may detect proximity and / or contact via ultrasonic waves, optical waves, radio waves, capacitance, temperature changes, mechanical switches, etc. In this example, the device 142 is shown as a mobile phone, however other devices such as key fobs, smart watch, may be used. If the device 142 is detected before or during the predetermined time period when the vehicle is manually stopped, the alert such as wireless communication to the user (or other alert or action) is suppressed.

[0031] The module 130 focuses on events that occur after a manual stop has been initiated. In contrast, module 132 monitors events that occur while the autonomous work machine 100 is performing autonomous work in a work region. In such a scenario, the vehicle sensors will be actively registering a number of different signals which may indicate movement under normal conditions. Thus a task performed by the module 132 is to continually classify one or more sensor signals over a time period, and see if they are operating outside of a normal operational envelope. In addition to detecting a sensor signal is outside of normal operation, it may want to focus on particular patterns that indicate the vehicle 100 is being stolen or otherwise externally moved.

[0032] In one embodiment, the module 132 monitors a gyroscope of the motion measurement unit 125 to determine a time-dependent an attitude profde of the autonomous work vehicle 100 that does not conform to an known autonomous work profile. For purposes of this disclosure the attitude profile is a measurement of a rotation angle of the machine around any axis, and may capture ascent / descent (pitch) angles, roll angles, tilt angles (a combination of roll and pitch), etc. The attitude profile may also include rates of change (e.g., first and / or second derivatives with respect to time) and integration over time. In some embodiments, the angle may not conform if it exceeds angles that have been experienced in previous traversals of the work region, or if it indicates a condition that the autonomous vehicle is unable to perform autonomously. For example, the work machine may compile a profile of tilt angles of work over time, and be able to build a confidence score that a currently measure tilt angle is outside the envelope of past tilt angles (either overall or in a particular region) and act accordingly as described elsewhere in that case.

[0033] The determination of non-conformal profile might involve fusing inputs from multiple sensors, e g., sensing a sudden decrease in wheel load together with a sharp change in attitude. In another example, a roll angle that would result in one wheel being offof the ground would not be consistent with simultaneous forward movement. In response to detecting the attitude profde that is non-conformal and indicates possible theft or the like the vehicle 100, e.g., via the event handler 134 and communications module 120, triggers an alarm and related actions as previously described, e.g., wirelessly communicates an alert to a user. Because the vehicle 100 is performing work when the event is detected, it could perform other actions in response to this detection, such as a motor shutdown to avoid a risk of injury.

[0034] It is likely that parallel processing of multiple sensor signals will provide a more accurate result regarding an event being anomalous than can be obtained from a single signal. Multiple sensors can also be useful to accurately estimate a likelihood that an anomaly is theft-related other than some other anomalous event, e.g., encounter with an newly encountered obstacle. In one or more embodiments, the in-use monitoring module 132 receives two or more time varying signals from respective two or more sensors of the work vehicle 100 while it is located in the work region. A first sensor of the two or more sensors includes a motion sensor, such as the motion measurement unit 125, cameras 140, wheel encoders 143, etc.

[0035] The module 132 may include one or more machine learning models that are empirically trained to detect anomalous events based on sensor inputs. The two or more time varying signals are input to the machine learning model or models, which classifies patterns of the two or more time varying signals. These classifications at least include a first class that indicates a theft of the autonomous machine is occurring and a second class that indicates a theft of the autonomous machine is not occurring. In response to detecting a first event that corresponds to the first class, the module 132 triggers an alarm and / or performs related actions as previously described. Because the vehicle 100 is performing work when the event is detected, it could perform other actions in response to this detection, such as a motor shutdown, activate brakes, lock down the user interface.

[0036] There are a wide variety of machine learning models that can be used to classify sensor signals in an autonomous work vehicle. Two types of machine learning models that can be used to classify data include supervised learning and unsupervised learning models. Unsupervised classifiers are trained on data without any labels, and use pattern recognition or the like to create classifications. Supervised and semi-supervisedclassifiers train on labeled data, e.g., data which has been previously identified and labeled with a classification of interest. There are advantages and disadvantages to both supervised and unsupervised classifiers.

[0037] Unsupervised classifiers may be less intensive to train because no labeling is needed, and sometimes can find pattems / classes that are not contemplated by the system architect. A downside is that the resulting classes identified may not correspond to the classes of interest to the end-user. Supervised learning can be directed to look for specific classes and therefore may provide more relevant results. The downside is that labeling datasets can be cost intensive, e.g., requiring humans to do relatively boring and repetitive tasks to build a large enough dataset.

[0038] One commonly used machine learning classifier is a Support Vector Machine (SVM), which can provide complex distinctions between classes. Other machine learning classifiers include neural networks, decision trees, and Bayesian classifiers. For purposes of the following examples, neural network classifiers are shown, however these may be replaced or augmented with other types of machine learning algorithms, such as SVMs.

[0039] In FIG. 2, a block diagram illustrates an architecture that can be used for machine learning classification of sensor data for an autonomous work vehicle according to an example embodiment. The sensors in this example are divided into multidimensional sensors (one or more cameras 200, which may include cameras on the work vehicle or cameras separate from the work vehicle; and LIDAR / RADAR 222) and onedimensional sensors 202-208 and 220-221. For purposes of this example, the term “onedimensional” refers to the sensor providing one or more electrical signals that each correspond to a single dimensional measurement. For example, while accelerometer 202 has three outputs corresponding to x-, y-, and z-axes, each output is a one-dimensional signal that is orthogonal to the others. The other sensors include gyrosc ope / tilt sensor 203, compass 204, wheel encoders 205, proximity / bump sensors 206, lift sensor 207, geolocation sensor 208, microphone 220, and communications link 221 (e.g., radio, infrared, ultrasonic). Features, capabilities, operation, and configurations of these sensors can be found in the description of FIG. 1 above.

[0040] This list of sensors is not exhaustive, and video signals from the cameras 200 and LIDAR / RADAR 222 can be used for motion detection as well, as described below. While some sensors such as microphone 220 communications link 221 may not directly measure motion, they may detect secondary effects of motion. For example communications link 221 sensors (e.g., radio receiver) can detect loss of signal strength, Doppler effects, etc. due to motion and / or relocation of the vehicle. The microphone 220 can detect approaching footstep, and acoustic signature consistent with being placed in a vehicle, etc.

[0041] The signals from one-dimensional sensors 202-208, 220, 221 are shown input to one or more neural networks 209, which may be recurrent neural networks (RNNs) and / or convolutional neural networks (CNNs). An RNN is a network with feedback elements that can capture time dependent aspects of a stream of data. A CNN uses convolutions or filters to analyze patterns in data, and have been found to be useful in image processing and the convolutional algorithms can be readily adapted for time series analysis. The neural networks 209 can process the sensor data separately or jointly to identify movement classes 210. The classes can represent a class of operation such as “normal,” “abnormal,” “unknown” that occurred over a finite time period, e.g., the last 1-5 seconds.

[0042] The video images 201 that are produced by the cameras 200 and / or 3D point clouds produced by LIDAR / RADAR 222 are input to one or more CNNs 212. The CNNs 212 may provide a number of different functions, such as segmentation and object classification. Segmentation involves identifying distinct parts of a scene, such as moving objects, static objects, ground, sky, etc. This is a way to filter out parts of a scene so that only regions of interest are further analyzed. Object classification involves identifying objects within a scene, e.g., person, tree, dog, bicycle, and may be used with or without prior segmentation. The CNNs 212 output visual classes, which as pertains to theft, may at least include an indicator as of human presence nearby, and the number of humans. Larger machines may require a two-person lift and so two or more people nearby may be more indicative of a theft than that of a single person. If network 209 is implemented as an RNN, it can also be trained to classify signals such as footstep patterns, voices, etc. that are familiar or outside what has normally been encountered over time. This may involve anunsupervised learning stage in the vehicle’s work environment that proceeds over a long enough time so that sufficient data can be gathered and processed to provide a threshold level confidence in the classifications.

[0043] In this example, the neural networks 209, 212 output distinct classifications, and so may be further processed by one or more feedforward neural networks 216. These networks 216 can be trained to jointly analyze motion and visual classes 210, 214 to identify theft events, as indicated by output classes 218. Note that the classes 210, 214 may be time dependent, e.g., exist over finite time periods, and the relative speed of processing of the networks 209, 212 may be significantly different. Thus there may be some timedependency between the classes 210, 214, e.g., class 214 indicates a human has been nearby in for last 30 seconds and class 210 indicates unusual movement over the last 5 seconds. Therefore, neural network 216 may also or instead a use a type of network that can detect and correlate time-varying phenomena, such as RNN and CNN.

[0044] The training for the various machine learning models involves collecting large amounts of data, e.g., during device testing. Generally a device will undergo many hours of testing under real-world conditions in which performance can be evaluated, outlier conditions can identified and remedied, etc. It is relatively straightforward to collect large amounts of data during this testing, e.g., by installing a local data storage device and / or network interface for collecting the data. Most of this data will reflect normal operating conditions or anomalies due to reasons other than theft, e.g., vehicle getting stuck, vehicle’s navigation getting confused, low traction conditions, etc. As pertains to theft detection, there may be specific case where the vehicle actually gets stolen and the data is recorded. In other cases, testers can manually move test machines from a test region in under various conditions to simulate a theft.

[0045] Once large amounts of data are collected, a first step is to generally analyze the data to see if trends can be detected without significant human analysis. For example, a data set comprising only normal operations can be readily obtained, and this can be used as a baseline that a classifier can use in attempt to classify data segments as normal or abnormal. Each sensor could be analyzed separately to see if any sensors are particularly indicative of abnormal behavior of interest, and further efforts can be focused on these sensors, either individually or jointly.

[0046] Once a classifier is created that can at least identify theft-predictive anomalies, a more refined analysis may involve running the data using labeled data, e.g., non-theft anomalies and theft-caused anomalies (either real or simulated), to see of a classifier can provide statistically reliable prediction of whether a model can reliably predict theft-related anomalies as compared to non-theft-related anomalies. These models can be specific to a particular vehicle (e.g., model number) or class of vehicles (e.g., electric consumer mowers). The former may be more accurate however the latter may be easier to deploy, e.g., don’t require gathering new data and training new models for each product version or redesign.

[0047] In FIG. 3, a flowchart illustrates an example of states and events which may be considered by an autonomous vehicle theft detection system according to an example embodiment. Block 300 detects whether the vehicle is in a charging state (which may also include being docked in a charger without any battery charging occurring_ , during which any unauthorized movement will trigger an alert at block 301. In some embodiments, this block 300 could apply to an idle state, in which the vehicle is parked or otherwise immobilized, but not necessarily charging or in a charging dock. A number of considerations may be made to determine whether a movement is authorized, such as identification of an unknown person via a camera in the vehicle and / or charger, proximity detection of owner’s mobile device, power off or shutoff manually activated, signals from an external monitoring system such as a perimeter detector, etc.

[0048] Block 302 detects whether the vehicle is in a moving / working state. If so, and a manual stop is detected at block 304, then a delay timer is used and unauthorized movement during a delay tinier triggers an alert as indicated at block 306. When the vehicle is moving / working but no manual stop is detected, then the vehicle will signal an alert if it detects movement (and / or other activity) outside of a normal operating profile at block 305.

[0049] In some cases, the autonomous work vehicle may be stationary outside of its charger. This could be due to a number of reasons, such as low battery, pause in work due to conditions (e.g., low light, rain), being stuck, etc. The autonomous work vehicle may still be in a work mode even if not currently working, and so the vehicle may still monitor for manual power off at block 307, which will result in a detection of movementduring a delay period (block 308) if power off is detected, or unauthorized movement if not (block 309).

[0050] In FIG. 4, a flowchart shows a method according to an example embodiment. The method is performed continually while an autonomous work vehicle is performing autonomous work 400, and involves determining 401 that the vehicle has been manually stopped. In response thereto, a delay timer is started 402 that runs for a predetermined time. During the predetermined time (while block 403 returns ‘no’), an inertial measurement unit of the autonomous work vehicle is monitored 404 to detect a motion event that indicates a movement of the autonomous work vehicle. In response to the event (block 405 returns ‘yes’), an alert is wirelessly communicated 406 that the motion event has occurred. If and when the timer expires (block 403 returns ‘yes’), the motion event monitoring is stopped 407. Generally, the user will be instructed to wait for the predetermined time period after performing the manual stop to prevent triggering the alert. A user interface indicator may also be used to indicate the expiration of the timer, e.g., a flashing LED that stops flashing when the timer expires.

[0051] In FIG. 5, a flowchart shows a method according to another example embodiment. The method is performed while an autonomous work machine is performing autonomous work 500 in a work region. A gyroscope of an autonomous work vehicle is monitored 501 for an attitude profile of the autonomous work vehicle. If it is detected at block 502 that the attitude profile does not conform to an autonomous work profile of the autonomous work machine, an alert is wirelessly communicated 503 that a non-conformal attitude profile has been detected.

[0052] In FIG. 6, a flowchart shows a method according to another example embodiment. The method is performed while an autonomous work machine is performing autonomous work 600 in a work region. The vehicle receives 601 two or more time varying signals from respective two or more sensors of the autonomous work vehicle while the vehicle is located in a work region. A first sensor of the two or more sensors comprises a motion sensor. The two or more time varying signals are input 602 into a machine learning model that classifies patterns of the two or more time varying signals into a first class that indicates a theft of the autonomous machine is occurring and a second class that indicates a theft of the autonomous machine is not occurring. In response to detecting 603a first event that corresponds to the first class, an alert is wirelessly communicated 604 that the event has occurred.

[0053] In view of the above, it will be readily apparent that the functionality of the controllers of the system may be implemented in any manner known to one skilled in the art. For instance, the memory may include any volatile, non-volatile, magnetic, optical, and / or electrical media, such as a random-access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), electrically-erasable programmable ROM (EEPROM), flash memory, and / or any other digital media.

[0054] The processors used in the controllers may include any one or more of a microprocessor, a controller, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), and / or equivalent discrete or integrated logic circuitry. In some embodiments, the processor may include multiple components, such as any combination of one or more microprocessors, one or more controllers, one or more DSPs, one or more ASICs, and / or one or more FPGAs, as well as other discrete or integrated logic circuitry. The functions attributed to the controller and / or processor herein may be embodied as software, firmware, hardware, or any combination of these. Certain functionality of the controller may also be performed in the cloud or other distributed computing systems operably connected to the processor.

[0055] While the present disclosure is not so limited, an appreciation of various aspects of the disclosure will be gained through a discussion of the specific illustrative aspects provided below. Various modifications of the illustrative aspects, as well as additional aspects of the disclosure, will become apparent herein.

[0056] Example Al is method, comprising: determining that an autonomous work vehicle has been manually stopped, and in response thereto, starting a delay timer that runs for a predetermined time; during the predetermined time, monitoring an inertial measurement unit of the autonomous work vehicle to detect a motion event that indicates a movement of the autonomous work vehicle; and in response to the motion event, wirelessly communicating an alert that the motion event has occurred.

[0057] Example A2 includes the method of example Al, wherein determining that the autonomous work vehicle has been manually stopped comprises detecting activation of an external shutoff switch. Example A3 includes the method of example Al or A2,wherein determining that the autonomous work vehicle has been manually stopped comprises detecting shutoff of main power.

[0058] Example A4 includes the method of any of examples Al -A3, wherein the motion event comprises an attitude of the autonomous work vehicle detected by a gyroscope of the inertial measurement unit. Example A5 includes the method of any previous ‘A’ example, wherein the motion event comprises a change rotation along a ground-parallel axis that exceeds a threshold. Example A6 includes the method of the method of any previous ‘A’ example, wherein the wirelessly communicating the alert comprises sending a message to a user that a possible theft of the autonomous work vehicle has occurred.

[0059] Example A7 includes the method of the method of any previous ‘A’ example, further comprising detecting, via a proximity detector, a device that was previously registered as belonging to a user, and wherein if the device is detected before or during the delay timer expires, suppressing the wireless communication. Example A8 includes the autonomous work vehicle as set forth in any previous ‘A’ example, having one or more controllers operable to perform the method of any previous ‘A’ example.

[0060] Example B9 is method, comprising: while an autonomous work vehicle is performing autonomous work in a work region, monitoring a gyroscope of an autonomous work vehicle for an attitude profde of the autonomous work vehicle that does not conform to an autonomous work profile of the autonomous work vehicle; and in response to detecting the attitude profile, causing the autonomous work vehicle to wirelessly communicate that a non-conformal attitude profile has been detected.

[0061] Example B10 includes the method of example B9, wherein wirelessly communicating that the non-conformal attitude profile has been detected comprises sending a message to a user that a possible theft of the autonomous work vehicle has occurred. Example Bl 1 includes the method of example B9 or B10, wherein the attitude profile comprises an ascent or descent angle that exceeds a limit. Example B12 includes the method of any previous ‘B’ example, wherein the attitude profile comprises a tilt angle that exceeds what has been experienced in previous traversals of the work region. Example B13 includes the method of any previous ‘B’ example, wherein the attitude profile comprises a change in tilt angle that exceeds a limit. Example B14 includes the method ofany previous ‘B’ example, wherein one of a speed or an acceleration of the autonomous machine is monitored in addition to the gyroscope, wherein the attitude profde indicates a condition that the autonomous vehicle is unable to perform autonomously. Example Bl 5 includes the autonomous work vehicle as set forth in any previous ‘B’ example, having one or more controllers operable to perform the method of any previous ‘B’ example.

[0062] Example C16 is method, comprising: receiving two or more time varying signals from respective two or more sensors of an autonomous work vehicle that is located in a work region, wherein a first sensor of the two or more sensors comprises a motion sensor; inputting the two or more time varying signals into a machine learning model that classifies patterns of the two or more time varying signals into a first class that indicates a theft of the autonomous machine is occurring and a second class that indicates a theft of the autonomous machine is not occurring; and in response to detecting a first event that corresponds to the first class, causing the autonomous work vehicle to wirelessly communicate an alert that the event has occurred, wherein the alert is not communicated responsive to a second event that corresponds to the second class.

[0063] Example C17 includes the method of example Cl 6, wherein the event is communicated via a network to a device of a user of the autonomous work vehicle. Example C18 includes the method of example C 16 or C 17, wherein the event is communicated via a network to a security service. Example C19 includes the method of example Cl 6, wherein the event is communicated to at least one camera located on at least one of base station used for charging the autonomous work vehicle and a building or vehicle associated with the work region, the event triggering the at least one camera to record an image of the work region in response thereto.

[0064] Example C20 includes the method of any previous ‘C’ example, wherein a second sensor of the two or more sensors comprises a camera mounted on the autonomous work vehicle. Example C21 includes the method of example C20, wherein the machine learning model detects a human presence from an image taken from the camera. Example C22 includes the method of any previous ‘C’ example, wherein the first sensor comprises a first gyroscope element that senses rotation about a first axis. Example C23 includes the method of example C22, wherein a second sensor of the two or more sensors comprises asecond gyroscope element that senses a second rotation about a second axis different from the first axis.

[0065] Example C24 includes the method of any previous ‘C’ example, wherein the machine learning model comprises a neural network. Example C25 includes the autonomous work vehicle as set forth in any previous ‘C’ example, having one or more controllers operable to perform the method of any previous ‘C’ example.

[0066] It is noted that the terms “have,” “include,” “comprises,” and variations thereof, do not have a limiting meaning, and are used in their open-ended sense to generally mean “including, but not limited to,” where the terms appear in the accompanying description and claims. Further, “a,” “an,” “the,” “at least one,” and “one or more” are used interchangeably herein. Moreover, relative terms such as ’’left,” “right,” “front,” “fore,” “forward,” “rear,” “aft,” “rearward,” “top,” “bottom,” “side,” “upper,” “lower,” “above,” “below,” “horizontal,” “vertical,” and the like may be used herein and, if so, are from the perspective shown in the particular figure, or while the machine is in an operating configuration. These terms are used only to simplify the description, however, and not to limit the interpretation of any embodiment described. As used herein, the terms “determine” and “estimate" may be used interchangeably depending on the particular context of their use, for example, to determine or estimate a position or pose of a vehicle, boundary, obstacle, etc.

[0067] Further, it is understood that the description of any particular element as being connected to or coupled to another element can be directly connected or coupled, or indirectly coupled / connected via intervening elements.

[0068] Unless otherwise indicated, all numbers expressing feature sizes, amounts, and physical properties used in the specification and claims are to be understood as being modified in all instances by the term “about.” Accordingly, unless indicated to the contrary, the numerical parameters set forth in the foregoing specification and attached claims are approximations that can vary depending upon the desired properties sought to be obtained by those skilled in the art utilizing the teachings disclosed herein. The use of numerical ranges by endpoints includes all numbers within that range (e.g., 1 to 5 includes within that range.

[0069] The various embodiments described above may be implemented using circuitry, firmware, and / or software modules that interact to provide particular results. One of skill in the arts can readily implement such described functionality, either at a modular level or as a whole, using knowledge generally known in the art. For example, the flowcharts and control diagrams illustrated herein may be used to create computer-readable instructions / code for execution by a processor. Such instructions may be stored on a non- transitory computer-readable medium and transferred to the processor for execution as is known in the art. The structures and procedures shown above are only a representative example of embodiments that can be used to provide the functions described hereinabove.

[0070] Note that any components described herein using terms such as “processor,” “controller,” “logic circuit,” “CPU,” or the like may be implemented using a plurality of discrete units operating together. For example, a processer that performs a series of steps or operations may be construed as two or more processors operating cooperatively to perform the steps. Similarly, other processing hardware such as memory and input-output may perform the described functions with multiple discrete units operating cooperatively or being coordinated by another unit, e.g., by a central processor or processors.

[0071] The foregoing description of the example embodiments has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the embodiments to the precise form disclosed. Many modifications and variations are possible in light of the above teaching. Any or all features of the disclosed embodiments can be applied individually or in any combination and are not meant to be limiting, but purely illustrative. It is intended that the scope of the invention be limited not with this detailed description, but rather determined by the claims appended hereto.

Claims

CLAIMS:

1. A method, comprising: determining that an autonomous work vehicle has been manually stopped, and in response thereto, starting a delay timer that runs for a predetermined time; during the predetermined time, monitoring an motion measurement unit of the autonomous work vehicle to detect a motion event that indicates a movement of the autonomous work vehicle; and in response to the motion event, wirelessly communicating an alert that the motion event has occurred.

2. The method of claim 1, wherein determining that the autonomous work vehicle has been manually stopped comprises detecting activation of an external shutoff switch.

3. The method of claim 1 or 2, wherein determining that the autonomous work vehicle has been manually stopped comprises detecting shutoff of main power.

4. The method of any previous claim, wherein the motion event comprises an attitude of the autonomous work vehicle detected by a gyroscope of the autonomous work vehicle.

5. The method of any previous claim, wherein the motion event comprises a change in rotation along a ground-parallel axis that exceeds a threshold.

6. The method of any previous claim, wherein wirelessly communicating the alert comprises sending a message to a user that a possible theft of the autonomous work vehicle has occurred.

7. The method of any previous claim, wherein wirelessly communicating the alert comprises communicating the alert to at least one camera located separate from the autonomous work vehicle, the alert triggering the at least one camera to record an image of a work region at or near the autonomous work vehicle in response thereto.

8. The method of any previous claim, further comprising detecting, via a proximity detector, a device that was previously registered as belonging to a user, and wherein if the device is detected before or during the delay timer expires, suppressing the wireless communication.

9. The method of any previous claim, further comprising, while the autonomous work vehicle is performing autonomous work in a work region: monitoring the motion measurement unit for an attitude profde of the autonomous work vehicle that does not conform to an autonomous work profile of the autonomous work vehicle; and in response to detecting the attitude profile, causing the autonomous work vehicle to wirelessly communicate that a non-conformal attitude profile has been detected.

10. The method of claim 9, wherein the attitude profile comprises an ascent or descent angle that exceeds a first limit, or a change in tilt angle that exceeds a second limit.

11. The method of claim 9, wherein the attitude profile comprises a tilt angle that exceeds what has been experienced in previous traversals of the work region.

12. The method of any one of claims 8-11, wherein one or both of a speed and an acceleration of the autonomous work vehicle is monitored in addition to the attitude profile, wherein the attitude profile and the one or both of the speed and the acceleration indicates a condition that the autonomous vehicle is unable to perform autonomously.

13. The method of any previous claim, further comprising: receiving a time varying signal from the motion measuring unit; inputting the time varying signal into a machine learning model that classifies patterns of the time varying signal into a first class that indicates a theft of the autonomous work vehicle is occurring and a second class that indicates a theft of the autonomous work vehicle is not occurring; andin response to detecting a first event that corresponds to the first class, causing the autonomous work vehicle to wirelessly communicate the alert that the first event has occurred, wherein the alert is not communicated responsive to a second event that corresponds to the second class.

14. The method of claim 13, further comprising: receiving a second time varying signal from a second sensor different from the motion measuring unit; inputting the second time varying signal into the machine learning model, wherein the machine learning model that jointly classifies patterns of the time varying signal and the second time varying signal into the first and second classes.

15. The method of claim 14, wherein the second sensor comprises a first camera mounted on the autonomous work vehicle or a second camera separate from the autonomous work vehicle, and wherein the machine learning model detects a human presence from an image taken from the camera.

16. The method of any one of claims 13-15, wherein the machine learning model comprises a neural network.

17. The autonomous work vehicle as set forth in any previous claim, having one or more controllers operable to perform the method of any previous claim.

Images