Autonomous precision pruning of fruit bearing plants

Abstract

A method and system for autonomously pruning, thinning, and shoot and leaf removal of fruit plants is presented. The system includes an unmanned aerial vehicle (UAV) equipped with a robotic arm capable of translation along three orthogonal directions and rotation about three orthogonal axes. At the end of the arm is a cutting mechanism for executing canopy-management tasks. The UAV incorporates one or more cameras sensitive to the visible spectrum, enabling real-time imaging of surrounding plants, along with processing circuitry and memory. The memory stores instructions that, when executed, cause the UAV to navigate to a target plant, capture image data, and generate a three-dimensional semantic model of the canopy and adjacent structures. Using this model, the system computes a pruning, thinning, or shoot and leaf removal plan according to horticultural policies and operator objectives. The UAV controls its arm and cutting mechanism to execute the plan in real time.

Description

AUTONOMOUS PRECISION PRUNING OF FRUIT BEARING PLANTSCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 692,081 filed on September 7, 2025, entitled “Autonomous System for Precision Pruning of Fruit Trees,” the contents of which are hereby incorporated by reference in their entirety.TECHNICAL FIELD

[0002] The present disclosure relates generally to unmanned aerial vehicles for autonomous precision pruning of fruit bearing plants, like trees and vineyards.BACKGROUND

[0003] Precision pruning of fruit bearing trees and vineyards, unlike general plant pruning, is a professional farming technique of accurately cutting the branches and restructuring the whole branch system. Precision thinning of fruit bearing trees and vineyards is another professional farming technique of accurately detaching some particular fruits at a specific time of a growing season. Precision shoot and leaf removal of fruit bearing trees and vineyards is still another professional farming technique of accurately detaching some particular shoots and leaves at specific times during a growing season.

[0004] These foregoing professional farming techniques are each directed to increase fruit quality and yield, and, in turn, increase the marketability of harvested fruit and revenue.

[0005] These foregoing professional farming techniques each rely on a thorough and comprehensive perception of geometrical and biological features. For example, for fruit bearing trees, geometrical and biological features include those of each tree branch, bud, and other wood features, such size, location, angle, and quality. It also includes collective features of branches and buds and other wood elements such as local density in different parts of the tree including the global density of a particular tree, as a whole.

[0006] Few workers possess the skills of these foregoing professional farming techniques. There is an increasing cost in the labor of the few workers who do possess these skills. In view of the scarcity of skilled workers in these foregoing professional farmingtechniques and the rising labor costs, it would be advantageous to provide a solution that would overcome these challenges.SUMMARY

[0007] A summary of several example embodiments of the disclosure follows. This summary is provided for the convenience of the reader to provide a basic understanding of such embodiments and does not wholly define the breadth of the disclosure. This summary is not an extensive overview of all contemplated embodiments, and is intended to neither identify key or critical elements of all embodiments nor to delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more embodiments in a simplified form as a prelude to the more detailed description that is presented later. For convenience, the term “some embodiments” or “certain embodiments” may be used herein to refer to a single embodiment or multiple embodiments of the disclosure.

[0008] A system of one or more computers can be configured to perform particular operations or actions by virtue of having software, firmware, hardware, or a combination of them installed on the system that in operation causes or cause the system to perform the actions. One or more computer programs can be configured to perform particular operations or actions by virtue of including instructions that, when executed by data processing apparatus, cause the apparatus to perform the actions.

[0009] In one general aspect, a system may include an unmanned aerial vehicle (UAV). The system may also include a robotic arm attached to the UAV, the robotic arm configured to be moved in three orthogonal directions and to be rotated about three orthogonal axes. The system may furthermore include a cutting mechanism attached to an end of the robotic arm. The system may in addition include the unmanned aerial vehicle, may include: one or more cameras sensitive to the visible spectrum, configured to view in real time the surroundings of the UAV and one or more plants; a processing circuitry; a memory, the memory containing instructions that, when executed by the processing circuitry, configure the system to: navigate the UAV to a target plant; capture image data of the target plant; generate, in real time, a three-dimensional semantic model of a target plant, the three-dimensional semantic model including features of the target plant and features of structures adjacent to the target plant based on the image data; compute, in real time, any of a pruning plan, a thinning plan, and a shoot and leaf removal plan, basedon the three-dimensional semantic model, horticultural policies and operator-specified objectives for the target plant; and control, in real time, the UAV, the robotic arm, and the cutting mechanism to execute the any of a pruning plan, a thinning plan, and a shoot and leaf removal plan for the target plant. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

[0010] Implementations may include one or more of the following features. The system where the cutting mechanism is based on any of shears, a saw, and a heating element. The system where the system is powered by any of electricity from one or more batteries, a gaseous fossil fuel, and a liquid fossil fuel. The system where the memory contains further instructions that, when executed by the processing circuitry, configure the system to: isolate some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV. The system where propulsion of the UAV isolates some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV. The system where an assembly including a series-elastic element isolates some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV. The system where propulsion of the UAV isolates the UAV during any of pruning, thinning, and shoot and leaf removal of the target plant. The system where the system further may include: a second robotic arm attached to the unmanned aerial vehicle, where the further instructions that, when executed by the processing circuitry, configure the system to: control, in real time, the second robotic arm to attach to the target plant. The system where the system further may include wireless communication circuitry, where the memory contains further instructions that, when executed by the processing circuitry, configure the system to: collect images of the target plant; digitally tag the target plant for identification; and wirelessly communicate, through the wireless communication circuitry, the tag and archived images of the target plant to a remote data center. The system may include: controlling, in real time, a second robotic arm, connected to the UAV, to attach to the target plant. The system may include: collecting images of the target plant; digitallytagging the target plant for identification; and wirelessly communicating, through a wireless communication circuitry, the tag and archived images of the target plant to a remote data center. Implementations of the described techniques may include hardware, a method or process, or a computer tangible medium.

[0011] In one general aspect, a non-transitory computer-readable medium may include one or more instructions that, when executed by one or more processors of a device, cause the device to: navigate an UAV to a target plant; capture image data of the target plant; generate, in real time, a three-dimensional semantic model of a target plant, the three- dimensional semantic model including features of the target plant and features of structures adjacent to the target plant based on the image data; compute, in real time, any of a pruning plan, a thinning plan, and a shoot and leaf removal plan, based on the three-dimensional semantic model, horticultural policies and operator-specified objectives for the target plant; and control, in real time, the UAV, a robotic arm attached to the UAV, and a cutting mechanism attached to the robotic arm to execute the any of a pruning plan, a thinning plan, and a shoot and leaf removal plan for the target plant. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

[0012] In one general aspect, the method may include navigate an UAV to a target plant. The method may also include capture image data of the target plant. The method may furthermore include generate, in real time, a three-dimensional semantic model of a target plant, the three-dimensional semantic model including features of the target plant and features of structures adjacent to the target plant based on the image data. The method may in addition include compute, in real time, any of a pruning plan, a thinning plan, and a shoot and leaf removal plan, based on the three-dimensional semantic model, horticultural policies and operator-specified objectives for the target plant; and control, in real time, the UAV, a robotic arm attached to the UAV, and the cutting mechanism attached to the robotic arm to execute the any of a pruning plan, a thinning plan, and a shoot and leaf removal plan for the target plant. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on oneor more computer storage devices, each configured to perform the actions of the methods.

[0013] Implementations may include one or more of the following features. The method where the cutting mechanism is based on any of shears, a saw, and a heating element. The method where the UAV, the robotic arm, and the cutting mechanism are powered by any of electricity from one or more batteries, a gaseous fossil fuel, and a liquid fossil fuel. The method may include: isolating some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV. The method may include: isolating some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV, by a propulsion of the UAV. The method may include: isolating some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV, by an assembly including a series-elastic element. The method may include: isolating the UAV during any of pruning, thinning, and shoot and leaf removal of the target plant by propulsion of the UAV. Implementations of the described techniques may include hardware, a method or process, or a computer tangible medium.BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The subject matter disclosed herein is particularly pointed out and distinctly claimed in the claims at the conclusion of the specification. The foregoing and other objects, features, and advantages of the disclosed embodiments will be apparent from the following detailed description taken in conjunction with the accompanying drawings.

[0015] Figure 1 illustrates example autonomous pruning system utilized to describe the various disclosed embodiments.

[0016] Figure 2 illustrates example autonomous pruning system utilized to describe the various disclosed embodiments.

[0017] Figure 3 is an example flow diagram of a pruning task, implemented in accordance with an embodiment.

[0018] Figure 4 is an example isometric view of an autonomous unmanned aerial vehicle for performing agronomical functions, implemented in accordance with an embodiment.

[0019] Figure 5 is an example schematic diagram of a plurality of autonomous UAVs deployed in a field, implemented in accordance with an embodiment.

[0020] Figure 6 is an example schematic diagram of a control station according to an embodiment.

[0021] Figure 7 is an example flowchart of a method for deploying a plurality of autonomous agronomical UAVs in a single location, implemented in accordance with an embodiment.DETAILED DESCRIPTION

[0022] It is important to note that the embodiments disclosed herein are only examples of the many advantageous uses of the innovative teachings herein. In general, statements made in the specification of the present application do not necessarily limit any of the various claimed embodiments. Moreover, some statements may apply to some inventive features but not to others. In general, unless otherwise indicated, singular elements may be in plural and vice versa with no loss of generality. In the drawings, like numerals refer to like parts through several views.

[0023] FIG. 1 illustrates example autonomous pruning system 100 utilized to describe the various disclosed embodiments. In autonomous pruning system 100, unmanned aerial vehicle 102 may be a system to place the pruning capabilities adjacent to a tree or vine.

[0024] Autonomous pruning system 100 may additionally comprise six degree of freedom (6DOF) robotic arm 104. 6DOF robotic arm 104 may be attached to unmanned aerial vehicle 102. 6DOF robotic arm 104 may be fastened to unmanned aerial vehicle 102 by one or more bolts 114, as illustrated in FIG. 1 , or by other means. At the end of 6DOF robotic arm 104, manipulator 106 is attached to implement pruning, thinning, shoot and leaf removal, and the like. As shown in FIG. 1 , manipulator 106 are illustrated resembling pruning shears. In alternative embodiments, manipulator 106 may be shears, a saw, heating element, a combination, and the like. Manipulator 106 may be fastened to a joint at the end of 6DOF robotic arm by a bolt as shown in FIG. 1 or by other means. 6DOF robotic arm 104 may have two or more sections connected by joints, which may be bolted, and gimbals to provide translational motion of electrical shears 106 along the x, y, and z directions and rotation of pitch, yaw, and roll about the x, y, and z axes, respectively. In an embodiment, manipulator 106 may be placed any point in space between the longestextension and shortest contraction of 6DOF robotic arm 104, from unmanned aerial vehicle 102. Manipulator 106 may also be oriented in any direction slightly less than 4TT steradians from a normal to an end of 6DOF robotic arm 104. This arrangement gives rise to manipulator 106 being placed and oriented in a position and direction, respectively, for a particular task, such as pruning, thinning, and shoot and leaf removal. Optionally, autonomous pruning system 100 may have a back-force isolation assembly (not shown) to stabilize unmanned aerial vehicle 102 during any of a pruning, thinning, and shoot and leaf removal. Additionally, autonomous pruning system 100 may have an arm (not shown) that temporarily attaches to a branch or trellis to stabilize unmanned aerial vehicle 102 particularly during heavier procedures.

[0025] Autonomous pruning system 100 may additionally include a variety of sensors. In an embodiment, ground camera 108 may be placed at the bottom of unmanned aerial vehicle 102. Ground camera 108 may be mounted on a gimbal with one or more degrees of rotation to orient ground camera 108 in a direction less than 2TT steradians from normal to the bottom of unmanned aerial vehicle 102. A three dimensional red, green, and blue (i.e. , visible light) sensitive (3D + RGB) camera 110 may be attached to unmanned aerial vehicle 102 near where 6DOF robotic arm 104 is attached. Ground camera 108 may view the ground over which autonomous pruning system 100 is traveling as well as trees and vines intended to be pruned. Like ground camera 108, 3D + RGB camera 110 may be mounted on a gimbal with one or more degrees of rotation to orient 3D + RGB camera 110 in a direction less than 2TT steradians from normal to the mounting surface of unmanned aerial vehicle 102. 3D + RGB camera 110 may be sensitive to ultraviolet and infrared light in addition to the visible light. 3D + RGB camera 110 may view a portion of a tree or vine to be pruned. Arm camera 112 is mounted at the end of 6DOF robotic arm 104. Arm camera 112 may view a pruning task being performed. Autonomous pruning system 100 may also have one or more 360 degree obstacle sensors to detect obstacles interfering with the operation of autonomous pruning system 100. Autonomous pruning system 100 may also have an inertial measurement unit (I MU) to measure how fast autonomous pruning apparatus 100 is accelerating and rotating. Autonomous pruning system 100 may additionally include light detection and ranging (LIDAR) sensor for navigation.

[0026] FIG. 2 illustrates example autonomous pruning system 200 utilized to describe the various disclosed embodiments. Energy source 205 may also include one or more batteries to provide electric power. The one or more batteries may be based for high energy / weight electrical storage. For example, the one or more batteries may be lithium polymer (LiPo) batteries. Alternatively, energy source 205 may be powered by one or more liquid or gaseous fossil fuels, such as oil, gasoline, kerosene, natural gas, propane, or butane. In this embodiment, energy source 205 may include one or more generators to convert energy from one or more fossil fuels to electrical energy.

[0027] FIG. 2 additionally illustrates controller 210 including various components. Controller 210 may include processing circuitry 212 comprising a computer with one or more graphical processing units (GPUs) for artificial intelligence (Al) edge computing, autonomous Al-controlled navigation, machine vision, three dimensional (3D) reconstruction, and payload control. Controller 210 may also include memory 214 comprising any of static random access memory (SRAM) for access for high speed applications and dynamic random access memory (DRAM) to provide high storage density. Controller 210 may additionally include storage 216 for long-term memory, such as for images of ground camera 108, (3D + RGB) camera 110, and arm camera 112.

[0028] FIG. 2 moreover illustrates controller 210 including wireless communication 218. Wireless communication 218 may include various components for wireless connectivity including a global position satellite (GPS) receiver for location tracking. Wireless communication 218 may additionally include a 900 m antenna for communications in difficult environments with obstacles to wireless transmission, such as hills, ridges, mountains, and buildings, a network interface, including wireless cellular / wireless fidelity (WiFi) antenna for wireless communications and networking.

[0029] FIG. 2 moreover illustrates actuators for controlling 6DOF robotic arm 104 and propellers 116A, 116B. Arm actuator 220 may interface processing circuitry 212 and 6DOF robotic arm 104 and manipulator 106 to control 6DOF robotic arm 104 and manipulator 106 for any of pruning, thinning, and shoot and leaf removal. Correspondingly, propeller actuator 230 may interface processing circuitry 212 and propellers 116A, 116B. Propeller actuator 230 may include one or more flight control I electronic speed control (FC / ESC) stacks to control flight operation. Propeller actuator230 may moreover include one or more brushless direct current (BLDC) motors to operate one or more propellers for flight operation. Alternatively, propellers 116A, 116B may be operated by motors powered by a liquid or gaseous fossil fuel.

[0030] FIG. 2 additionally illustrates sensor interface 240. Sensor interface 240 may allow processing circuitry 212 to control ground camera 108, (3D + RGB) camera 110, arm camera 112, the one or more 360 degree obstacle sensors, an inertial measurement unit, and the LIDAR sensor. Sensor interface 240 facilitates initial inspection of a target plant, then more detailed images taken of the target plant, and then subsequently collecting sensor data during performing any of pruning, thinning, and shoot and leaf removal.During any of pruning, thinning, and shoot and leaf removal, autonomous pruning system 100 may undergo movement from a fixed location caused by the back force of 6DOF robotic arm 104 and manipulator 106. Unmanned aerial vehicle 102 may be isolated fully, or in part, from the back force of 6DOF robotic arm 104 and manipulator 106, by propulsion of the one or more propellers of unmanned aerial vehicle 102. Isolation may also be provided by an assembly including a series-elastic element (not shown). Additionally, isolation may also be provided by a second robotic arm (not shown) that attaches to a plant undergoing a pruning task.

[0031] FIG. 3 is an example flow diagram of a pruning task, implemented in accordance with an embodiment. In some embodiments, the operations depicted may be performed by specially configured autonomous pruning system 100.

[0032] At S310, autonomous pruning system 100 is configured to navigate UAV 102 to a target plant, such as a tree or vine. After the target plant is recognized as a tree or vine to be treated, autonomous pruning system 100 moves UAV 102 toward the target plant.

[0033] At S320, in close proximity to the target plant, autonomous pruning system 100 is configured to capture image data of the target plant. The image data of the target plant may be stored and may undergo further processing as explained below.

[0034] At S330, autonomous pruning system 100 is configured to generate, in real time, a three-dimensional semantic model of features of the target plant and features of structures adjacent to the target plant based on the image data. For example, features of the target plant may include branches, buds, spurs, fruit, one or more nuts, one or more leaves, and the like. Features of structures adjacent to the target plant may include wires,poles, trellis members, and the like. The foregoing features may be labelled. There may be updates at greater than or equal to 10 Hz with less than 150 ms planning latency. Autonomous pruning system 100 is configured to implement a semantic segmentation network trained to label buds, spurs, fruit, leaves, branches, and wires.

[0035] At S340, autonomous pruning system 100 is configured to compute, in real time, any of a pruning plan, a thinning plan, and a shoot and leaf removal plan, based on the three- dimensional semantic model, horticultural policies, and operator specified-objectives for the target plant. Computing any of a pruning plan, a thinning plan, and a shoot and leaf removal plan based on the three-dimensional semantic model may involve a library of images of various tree and vine species at different angles and proximities to an imaging device. Computing may involve feeding the library of images and the three dimensional semantic model through a machine learning algorithm to recognize the plant species from the three-dimensional semantic model. Computing may further include a policy / planning engine encoded with species / training rules from the horticultural policies. Specifically, an embodiment may include a parameterized horticultural policy table mapping species and training systems to rule weights. Computing may moreover involve the operator-specified objectives, such as crop-load targets, The computed plan of any of pruning, thinning, and shoot and leaf removal may include candidate cuts / thins, selection of tools in manipulator 106, and optimized trajectories. Autonomous pruning system 100 may implement three- dimensional (3D) skeletonization by fitting a minimum-spanning tree or Steiner tree over point-cloud data with local radius estimation.

[0036] At S350, autonomous pruning system 100 is configured to control, in real time, UAV 102, 6DOF robotic arm 104, and manipulator 106 to execute the any of a pruning plan, a thinning plan, and a shoot and leaf removal plan. Cuts performed during execution of any of these plans may be verified to a tolerance from 2 to 15 mm. Cuts may also be logged on a per-tree / vine digital record for pose, tool, force / torque, and imagery. Autonomous pruning system 100 is further configured to implement anomaly detection. For example, during approach for cutting and cutting, autonomous pruning system 100 is configured to prompt replanning for any of pausing, retreating, and re-approaching for cutting.

[0037] Embodiments of tasks may include certain fruits, such as apple, peach, and grapes. In an embodiment, for apple, a shoot and leaf removal plan may include removingcrossing / upward shoots, maintaining spacing along leaders, renewing overstretched spurs, enforcing a greater than 8 cm wire standoff, and verifying a less than 10 mm cut error. In a separate embodiment for peach, autonomous pruning system 100 may detect blossom clusters, and use a soft thinning tool with spur-guard and fruit exclusion masks of greater than 20 mm. In still another embodiment concerning a grapevine, autonomous pruning system 100 may identify renewal spurs and intermode lengths, cut with a microsaw while maintaining trellis standoff. In this embodiment, per-node cuts may be logged.

[0038] Figure 4 is an example isometric view of an autonomous unmanned aerial vehicle for performing agronomical functions, implemented in accordance with an embodiment. In an embodiment, a UAV includes a body 401 which houses various control circuitries, sensor arrays, wireless communication circuitry, various combinations thereof, and the like, discussed in more detail herein.

[0039] In an embodiment, the UAV includes a power source, such as battery 410. In an embodiment, the battery 410 is detachable. In some embodiments, the UAV includes a plurality of power sources, such that when a first battery 410 is detached, the UAV retains some power to certain mission critical circuitries, RAM, etc.

[0040] According to an embodiment, the UAV includes a plurality of arms 402-1 through 402- N, where ‘N’ is an integer having a value of ‘4’ or greater. In an embodiment, each arm is connected to a motor, actuator, and the like, such as motors 405-1 through 405-N, which correspond respectively to arms 402-1 through 402-N.

[0041] In an embodiment, the UAV includes a manipulator receptacle 415 which is configured to connect the UAV to a manipulator, such as a robotic arm. In an embodiment, the manipulator receptacle 415 is utilized by the UAV as a temporary attachment to a plant, for example by providing weight support to a UAV. In an embodiment, the manipulator is connected via a hinge 422 which offers a first arm portion 425 with a degree of a freedom. In an embodiment, the arm is connected via shaft 432 to a second servo 440, which is configured to provide motion via rotational axis 442 to an arm fixture which includes a first grapple arm 451 and a second grapple arm 452. In an embodiment, an actuator is connected to the grapple arms to move them between a first position (e.g., an open position), and a second position (e.g., a closed position), such that in the open position an end of the first grapple arm 451 is farther from an end of the second grapple arm 452than in the closed position. In an embodiment, the robotic arm includes a sensor array 450. In some embodiments, the sensor array includes a LiDAR sensor, a camera sensor, an IMU, various combinations thereof, and the like.

[0042] Figure 5 is an example schematic diagram of a plurality of autonomous UAVs deployed in a field, implemented in accordance with an embodiment. In an embodiment, a control station 530 is configured to communicate with a plurality of UAVs such as first UAV 520-1 and second UAV 520-2. In an embodiment, the control station 530 includes a wireless network interface 532 including short range wireless communication (e.g., Bluetooth, Zigbee, Wi-Fi, NFC, UWB, a combination thereof, and the like), medium-long range wireless communication (e.g., cellular networks, LoRa, Sigfox, Iridium, Starlink, a combination thereof, and the like), specialized interfaces, various combinations thereof, and the like.

[0043] In an embodiment, the control station 530 configures the first UAV 520-1 to navigate a first terrain 512-1 defined between a first trellis 510-1 and a second trellis 510-2. In an embodiment, each trellis 510-1 through 510-N is associated with a single UAV of a plurality of UAVs 520. For example, the first UAV 520-1 is configured to navigate through the first terrain 512-1 and initiate agronomical actions only on the first trellis 510-1 , while the second UAV 520-2 is configured to navigate through a second terrain 512-2 defined between the second trellis 510-2 and a third trellis 510-3 and operate only the second trellis 510-2. According to an embodiment, a control station 530 is provided with a digital terrain map which is utilized to configure a navigation path for each UAV 520, the digital terrain map including a plurality of rows (such as trellis 510-1 through 510-N, where ‘N’ is an integer having a value of ‘2’ or greater) and a plurality of terrains defined between the plurality of rows (such as terrains 512-1 through 512-M, where ‘M’ is an integer having a value ‘T less than ‘N’).

[0044] In an embodiment, each UAV is configured to operate autonomously of each other, such that a single UAV is assigned a single terrain 512. In some embodiments, a plurality of UAVs are assigned a single terrain 512.

[0045] Figure 6 is an example schematic diagram of a control station 530 according to an embodiment. The control station 530 includes, according to an embodiment, a processing circuitry 610 coupled to a memory 620, a storage 630, and a network interface 640. In anembodiment, the components of the control station 530 are communicatively connected via a bus 650.

[0046] In certain embodiments, the processing circuitry 610 is realized as one or more hardware logic components and circuits. For example, according to an embodiment, illustrative types of hardware logic components include field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), Application-specific standard products (ASSPs), system-on-a-chip systems (SOCs), graphics processing units (GPUs), tensor processing units (TPUs), Artificial Intelligence (Al) accelerators, general-purpose microprocessors, microcontrollers, digital signal processors (DSPs), and the like, or any other hardware logic components that are configured to perform calculations or other manipulations of information.

[0047] In an embodiment, the memory 620 is a volatile memory (e.g., random access memory, etc.), a non-volatile memory (e.g., read only memory, flash memory, etc.), a combination thereof, and the like. In some embodiments, the memory 620 is an on-chip memory, an off-chip memory, a combination thereof, and the like. In certain embodiments, the memory 620 is a scratch-pad memory for the processing circuitry 610.

[0048] In one configuration, software for implementing one or more embodiments disclosed herein is stored in the storage 630, in the memory 620, in a combination thereof, and the like. Software shall be construed broadly to mean any type of instructions, whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise. Instructions include, according to an embodiment, code (e.g., in source code format, binary code format, executable code format, or any other suitable format of code). The instructions, when executed by the processing circuitry 610, cause the processing circuitry 610 to perform the various processes described herein, in accordance with an embodiment.

[0049] In some embodiments, the storage 630 is a magnetic storage, an optical storage, a solid-state storage, a combination thereof, and the like, and is realized, according to an embodiment, as a flash memory, as a hard-disk drive, another memory technology, various combinations thereof, or any other medium which can be used to store the desired information.

[0050] The network interface 640 is configured to provide the control station 530 with communication with, for example, the UAVs 520-1 and 520-2, according to an embodiment.

[0051] It should be understood that the embodiments described herein are not limited to the specific architecture illustrated in Fig. 6, and other architectures may be equally used without departing from the scope of the disclosed embodiments.

[0052] Figure 7 is an example flowchart of a method for deploying a plurality of autonomous agronomical UAVs in a single location, implemented in accordance with an embodiment.

[0053] At S710, a digital terrain map is received. In an embodiment, the digital terrain map includes a plurality of paths, each path including a plurality of plants. For example, in an embodiment, a path includes a terrain defined by a first trellis on one side, and a second trellis on a second side. In an embodiment, the plurality of plants include a plurality of grape vines.

[0054] In one embodiment, the system receives a digital terrain map that represents the layout of an agricultural block. The terrain map includes a plurality of paths, each path corresponding to a navigable corridor containing a sequence of plants. For example, a path may be defined by a first trellis structure on one side and a second trellis structure on the opposite side, thereby forming a bounded lane suitable for UAV navigation and canopy management operations.

[0055] According to an embodiment, each path includes a plurality of plants, such as grapevines, that are arranged along the trellis in accordance with a bilateral cordon, vertical shoot positioning, or other viticultural training system. The digital terrain map may further encode geospatial information such as elevation changes, row orientation, row length, and the relative positions of poles, wires, and other supporting infrastructure. By incorporating this terrain data, the system can align flight planning, obstacle avoidance, and task scheduling with the physical structure of the vineyard. In an embodiment, the map thus provides both structural boundaries and plant-level references, enabling precise execution of pruning, thinning, or inspection activities within each designated path.

[0056] At S720, each path is assigned to an autonomous UAV. In an embodiment, each path defined within the digital terrain map is assigned to a single autonomous UAV, such that the UAV operates exclusively within its designated corridor. The UAV is responsible forcompleting all canopy-management tasks along that path, including pruning, thinning, inspection, and data collection, without relying on another UAV for assistance. This independence simplifies coordination, prevents collisions, and ensures that each UAV can function even in the event of limited or delayed communication with other units. The UAV leverages the terrain map, a portion of the map, etc., and onboard sensors to navigate, localize plants, and execute operations in real time while maintaining safe standoff from trellis structures and wires.

[0057] In some cases, however, a UAV may be unable to complete its assigned path, for example due to battery depletion, unexpected mechanical issues, or adverse environmental conditions. In such situations, the UAV is configured to communicate with the control station to initiate remediation. In an embodiment, remediation includes navigating to a predefined location for a battery replacement. In certain embodiments, the control station is configured to initiate a handoff procedure in which the unfinished portion of the path is reassigned to another available UAV. The reassigned UAV resumes execution at the precise interruption point, using the shared terrain map and per-plant digital records to ensure continuity and prevent redundancy. This approach balances localized autonomy with flexible reallocation, thereby maximizing overall fleet efficiency and reliability.

[0058] At S730, each autonomous UAV is configured to execute an agronomy objective. Each UAV assigned to a path is configured to autonomously perform a wide range of canopy-management tasks along the plants contained within that path. As the UAV advances, its integrated sensor suite continuously generates a semantic three- dimensional model of the canopy and nearby infrastructure. This model allows the UAV to localize itself, identify key plant features, and compute intervention points with high precision. For example, in a vineyard row the UAV may identify renewal spurs and internode lengths, then carry out selective cuts with a micro-saw while maintaining a defined clearance from trellis wires. In an apple orchard, the UAV may remove crossing shoots, renew overstretched spurs, or thin blossoms using a compliant gripper, ensuring optimal branch spacing and spur density. For peach trees, the UAV may detect blossom clusters and selectively thin them while masking developing fruit larger than a set threshold.

[0059] Beyond pruning and thinning, each UAV may also execute leaf removal to improve light penetration, estimate local vigor through multispectral sensing, or detect early signs of disease. These broader functions expand canopy management from purely structural interventions to health and productivity monitoring. All actions are logged, including endeffector states, positional accuracy, and before-and-after imagery, creating a persistent digital record that supports individualized, multi-season plant management.

[0060] At S740, the UAVs are continuously deployed. In an embodiment, the system is designed for continuous deployment, allowing each UAV to operate over extended periods without interruption to overall task execution. As a UAV progresses along its assigned path, it continuously monitors its remaining energy through onboard sensors and compares this against the estimated energy required to complete the remaining portion of the path. When the energy level falls below a defined threshold, the UAV autonomously disengages from active canopy operations, retreats safely from the plants, and navigates to a designated control or docking station. At the station, the UAV either exchanges its depleted battery for a fully charged unit or connects to a charging interface.

[0061] For example, during pruning of a vineyard row, a UAV may identify that its battery charge has dropped to a level insufficient to complete the row. In response, the UAV withdraws from the trellis corridor, travels to the dock, and performs an automated battery swap. Once fully powered, the UAV returns to the precise location where it left off, guided by its semantic map and digital task log. From there, it resumes pruning without redundancy or omission, continuing along the path until all planned actions are completed. This process enables uninterrupted productivity across the orchard.

[0062] In some embodiments, the system supports continuous deployment not only at the level of a single UAV but also across an entire fleet, ensuring uninterrupted operations throughout the orchard or vineyard. Each UAV monitors its remaining energy and autonomously decides when to withdraw for recharging or battery replacement. To maximize efficiency, the fleet employs scheduling logic that staggers docking cycles so that no two UAVs assigned to adjacent paths retreat at the same time. This coordination prevents operational gaps and ensures that canopy-management activities continue seamlessly across multiple rows.

[0063] For example, in a vineyard block where several UAVs are assigned to neighboring trellis corridors, one UAV may detect that its battery is nearing depletion. Rather than risk an incomplete operation, the UAV disengages from its row, safely retreats from the plants, and travels to the control station for an automated battery swap. During this interval, UAVs operating in adjacent paths continue uninterrupted, maintaining full system productivity. Once its energy is restored, the redeployed UAV returns precisely to the point in its row where it left off, guided by its semantic model and logged digital record. By resuming from the interruption point without redundancy, the UAV completes its assigned path, while the staggered redeployment ensures that fleet-level coverage is continuous and efficient.

[0064] Various embodiments include an autonomous aerial canopy-management system. The system includes a UAV-mounted manipulator equipped with cutting and thinning endeffectors, along with a back-force isolation assembly that reduces reaction loads. A sensor suite maintains a real-time semantic three-dimensional model of each plant as well as nearby wires and trellis structures.

[0065] Some embodiments includes a policy and planning engine configured to compute candidate cuts and thins according to species-specific rules and operator objectives, and control UAV and manipulator trajectories while enforcing safety constraints. The engine is further capable of replanning in real time in response to branch deflection or canopy occlusion. For additional stability during cutting, the system may employ a temporary attachment mechanism. Each plant-specific action is logged into a digital record, enabling traceability and personalization based on before-and-after scans. The system also manages energy autonomously by returning for battery swapping or charging, and it supports coordination of multiple UAVs operating together across orchard rows.

[0066] Certain embodiments include an autonomous robotic system that is configured to autonomously carry out precise tasks in agriculture and beyond, including autonomous independent UAV-borne precision pruning, thinning, etc., which in some embodiments is based on a professional farming technique. In an embodiment, the system is configured to perform on-the-fly planning and replanning. In some embodiments, the system is configured to generate back-force isolation. For example, in an embodiment the system is configured to optionally attach to a stable portion of a plant, such as a tree branch, trunk, etc. According to an embodiment, the system is configured to generate a per-treedigital identity, log, and the like. In an embodiment, the system is configured to initiate autonomous battery swap, recharge, etc. In some embodiments, a control system is configured to coordinate a multi-UAV fleet.

[0067] Some embodiments includes an unmanned aerial vehicle (UAV) system, a device, and methods that perceive, plan, and execute precision pruning, fruit thinning, blossom thinning, shoot thinning, selective leaf removal in fruit orchards, nut orchards, vineyards, various combinations thereof, and the like.

[0068] According to an embodiment, a UAV is configured to include a manipulator. In an embodiment, the manipulator includes an arm equipped with cutting effectors, thinning end-effectors, a back-force isolation (BFI) assembly, a combination thereof, and the like. In some embodiments, the UAV further includes an attaching arm, a plurality of attachment arms, which are configured to temporarily attach to wood, trellis, etc., to stabilize heavier cuts, e.g., cuts of branches having a diameter greater than a predefined threshold.

[0069] In an embodiment, the UAV is configured to generate a fused real-time semantic 3D model. In certain embodiments, the model includes labeled branches, buds (e.g., floral, vegetative, etc.), spurs, fruit, leaves, wires, trellis, various combinations thereof and the like. In an embodiment, the UAV is configured to generate each label for a feature of the semantic 3D model. In some embodiments, the UAV is configured to update the model at a rate of >10 Hz with <150 ms planning latency.

[0070] In certain embodiments, the system includes a policy engine, a planning engine, and the like, which is configured to encode species, training rules, operator objectives (e.g., crop-load targets, spur renewal, etc.,) to generate candidate cuts, to generate candidate thins, choose tools, optimize trajectories, various combinations thereof, and the like.

[0071] In some embodiments, the system includes a control circuitry which is configured to enforce wire / trellis standoff, attenuate reaction forces through admittance and impedance, react to deflections, occlusions, and the like, utilizing on-the-fly replanning.

[0072] As used herein, real-time refers to the processing time of software, including artificial intelligence (Al) model systems, and additional software configured to execute on-the-fly, and not generating any delay to the robotic operation. As used herein, some embodiments include an attachment device which includes a quick-release, temporaryanchoring (e.g., micro-spikes, micro-claws, vacuum cups, dry-adhesive pads, trellis hooks, etc.), any combination thereof, and the like. As used herein, a semantic model is a fused 3D map labeling plant organs, infrastructure, branch skeleton model (order, diameter, angle, internode length, spur density, etc.), various combinations thereof, and the like.

[0073] In an embodiment, a multirotor UAV carries a two to six degrees of freedom (2-6- DOF) manipulator with cutting / detaching end-effectors. In some embodiments, the system includes a sensor array, including an RGB-D sensor. In an embodiment, the system includes a LiDAR. In certain embodiments, the system includes an inertial measurement unit (IMU). In an embodiment, the IMU includes GPS-RTK. In some embodiments, the sensor array periodically (e.g., multiple times in 1 second) feeds a fusion stack providing visual odometry with trellis loop closures. In some embodiments, the system is configured to generate semantic segmentation labels based on the received telemetry, including labeling plant organs, infrastructure (such as wires, poles, etc.); skeletonization (to yield branch order, diameter, angle, internode length, spur density, etc.), and the like. In an embodiment, wire detection, trellis detection, and the like, utilizes multi-view I ine / cyl inder fitting.

[0074] In certain embodiments, the system is configured to apply farming expert rulesets for fruit trees and nut trees such as, and not limited to, apple tai l-spindle, pear bi-axis, peach open-vase, cherry central-leader, almond hedgerow, grapevine bilateral cordon, and the like.

[0075] In an embodiment, candidate cuts, candidate thins, a combination thereof, and the like, are generated from the branch graph constrained by local and global tree spur / bud density and fruit proximity, multi-objective scoring based on crop-load targets, spur density minima, light proxies, safety margins, various combinations thereof, and the like. In an embodiment, trajectories are generated which utilize minimum-snap with model predictive control (MPC) under jerk / acceleration bounds and a wind-oriented safe halfspace. In an embodiment, tool selection (e.g., shear, micro-saw, heated, thinning, etc.) depends on branch diameter, bud proximity, and spur density.

[0076] According to an embodiment, during approach and cutting operations, deflectionbased triggers (e.g., exceeding 5-20 mm) initiate replanning of the cutting path.Admittance control, impedance control, etc., is employed to reduce reaction loads on the UAV and cutting mechanism. Where an occlusion is detected, the system is configured to pause, retreat, and then re-approach the target.

[0077] In an embodiment, fruit (or other objects) to be excluded from cutting are masked out if they are >20 mm in size (or other predetermined size) during pruning operations. Each cutting action is validated within a specified tolerance and logged with associated data, including tool pose, applied force / torque, and image records.

[0078] In an embodiment, the system includes a generative Al model which is trained based on a plurality of farming techniques, growing styles, etc. In an embodiment, the Al model is configured to receive pre / post scans for multiple trees of a particular grower, and learn growing techniques based on the received input.

[0079] In some embodiments, the system is configured to detect rows, trellises, etc., and supports autonomous coverage. In certain embodiments, which include deployment of a UAV fleet, a dock and central control is configured to perform a battery swap. Multiple UAVs are coordinated with predefined landing spots, dynamic landing spot allocation, etc., in an embodiment.

[0080] For example, for apple trees trained to a tall-spindle system during dormant pruning, the method involves selectively removing shoots that cross or grow upward, while maintaining consistent spacing along the leader branches. Spurs that have become overstretched are renewed to sustain productivity, and a minimum standoff distance of 8 cm from supporting wires is enforced. Each pruning cut is verified for accuracy, ensuring deviations do not exceed 10 mm.

[0081] As another example, in the case of peach trees during bloom thinning, blossom clusters are first detected, after which a compliant thinning tool equipped with a spur guard is applied. To prevent accidental removal of developing fruit, an exclusion mask is implemented for objects with a size of at least 20 mm, thereby ensuring selective and precise thinning operations.

[0082] As a further example, for grapevines managed under a bilateral cordon training system, renewal spurs are identified and internode lengths are measured to guide the pruning process. Cuts are executed using a micro-saw, while maintaining properclearance from the trellis structure. Each cut performed at a node is logged, capturing relevant operational data to ensure accuracy and traceability.

[0083] In an embodiment, real-time simultaneous localization and mapping (SLAM) combined with three-dimensional modeling is executed on an edge computer on the UAV, where data from multiple visual sensors is continuously fused to achieve a precision exceeding 5 mm. The system is configured to incorporate trellis loop closures to maintain accuracy and employs dynamic calibration along with a re-scanning procedure to ensure reliable operation under changing conditions.

[0084] According to an embodiment, skeletonization and detection are performed to enable real-time, detailed perception and modeling of the tree structure. The system provides complete and precise identification of tree features, including buds and branches, while also detecting surrounding elements such as poles and wires. Based on this structural understanding, cutting-point estimation is generated and adapted from expert policies developed for fruit-tree manipulation.

[0085] In some embodiments, control and maneuvering are achieved through high-precision, resilient motion planning and position control, enabling the system to detect and bypass obstacles while accurately guiding the end effector to a sequence of planned cut points with a positional accuracy of 5 mm or better. Operational safety is supported by dynamically defined standoff distances from wires and trellis structures, along with the enforced use of propeller guards to ensure protection of the UAV during flight. Energy management and fleet coordination are facilitated through docking alignment using visual fiducials, combined with controlled task allocation across multiple UAVs.

[0086] In certain embodiments, an autonomous aerial canopy-management system includes a multirotor unmanned aerial vehicle (UAV) equipped with a manipulator that is mounted to the airframe and terminates in an end-effector for performing pruning or thinning tasks. According to an embodiment, a suite of sensors is integrated into the system, including at least one depth or LiDAR sensor, one or more color cameras, and an inertial sensor, which together provide comprehensive perception of the environment. The UAV further incorporates, in an embodiment, one or more processors with associated memory configured to execute instructions for canopy-management operations.

[0087] In operation of some embodiments, the processors (i.e., processing circuitries) fuse data from the various sensors to generate, in real time, a three-dimensional semantic model of the target plant. This model captures the detailed structure of the canopy, including branches, buds, spurs, fruit, and leaves, as well as nearby infrastructure elements such as wires, poles, and trellis members. Based on this model, the system computes a pruning or thinning plan that reflects both horticultural policies and operator- specified objectives.

[0088] In an embodiment, the UAV and its manipulator are then controlled to carry out the plan, with the end-effector guided to approach and actuate at designated target locations. During execution, the system continuously updates the semantic model and adapts trajectories to account for changes such as branch deflection, partial occlusions, or plant motion. To enhance stability during these operations, the manipulator incorporates a back-force isolation assembly that reduces reaction forces transmitted to the UAV, thereby preserving flight control accuracy and operational safety.

[0089] Certain embodiments includes a method of autonomously performing canopy management which begins with sensing a target plant using a UAV-mounted sensor suite that may include depth sensors, LiDAR, color cameras, and inertial sensors. Data from these sensors is fused to construct a real-time semantic three-dimensional model representing the structure of the plant canopy. This model captures details such as branches, buds, spurs, fruit, and leaves, as well as nearby infrastructure elements like wires, trellis members, and support poles.

[0090] Using this semantic model, the system computes candidate cuts or thinning actions, guided by species-specific horticultural policies and operator-defined objectives. For example, in apple trees trained to a tall-spindle system, the system may identify crossing or vertically oriented shoots for removal, maintain spacing along leader branches, and renew overstretched spurs. For peach trees during bloom thinning, the system may detect clusters of blossoms and plan selective removals using a compliant thinning tool, while applying exclusion masks to avoid fruitlets larger than a threshold size. In grapevines managed under a bilateral cordon system, the method may identify renewal spurs and internode lengths and designate precise locations for cuts to preserve vine productivity and balance.

[0091] In an embodiment, once candidate operations are identified, the method optimizes both the execution sequence and the coupled trajectories of the UAV and its manipulator. The UAV approaches each target location, during which reaction forces are mitigated either through a back-force isolation assembly integrated into the manipulator or by establishing a temporary attachment to the plant or trellis. At each target, the end-effector (e.g., a cutting blade, micro-saw, soft thinning tool, etc.) is actuated to carry out the planned pruning or thinning action.

[0092] Throughout execution, the semantic model is dynamically updated to incorporate new information, enabling the system to replan as needed. For example, where a branch deflects during cutting, where a portion of the canopy becomes occluded, or if additional structures such as hidden buds or wires become visible, the UAV is configured to adjust its plan in real time. This continuous loop of sensing, modeling, planning, and replanning ensures robust and adaptive canopy management across a variety of crops and training systems, delivering precision operations under dynamic field conditions.

[0093] In some embodiments, the autonomous aerial canopy-management system is configured to integrate a series of advanced components that enable precise, resilient, and safe operation in complex orchard and vineyard environments. At the core of the manipulator is a back-force isolation assembly incorporating a series-elastic element with an effective stiffness between 0.5 kN / m and 10 kN / m and a damping ratio between 0.1 and 0.7, designed to attenuate reaction forces transmitted to the UAV, for some embodiments. In an embodiment, a six-axis force / torque sensor located at the manipulator wrist provides additional feedback, allowing the processors to enforce wrench limits that preserve flight stability. To further stabilize operations at specific cutting locations, the system may employ a temporary attachment device, such as micro-spike clamps, micro-claws, vacuum cups, dry-adhesive pads, or trellis hooks, each equipped with a quick-release actuator. These devices are mounted on compliant supports that disengage under a breakaway force of 10-100 N, ensuring safety in dynamic conditions.

[0094] In an embodiment, the system is configured to support a versatile range of endeffectors tailored to canopy-management tasks. These include shears with blade openings of 2-30 mm and integrated diameter gauges to prevent cutting oversized branches, micro-saws equipped with spur-guard shrouds and chip-extraction paths,heated cutters, and soft grippers configured for blossom or fruit thinning. Selection among these tools is made automatically based on sensed parameters such as branch diameter, bud proximity, and spur density. Certain end-effectors may also include disinfection modules, such as heated cutting edges maintained between 60 °C and 120 °C or UV emitters, to reduce the risk of pathogen spread.

[0095] In an embodiment, perception and planning are performed at high speed, with the semantic model updated at a rate of at least 10 Hz and an end-to-end perception-to-plan latency under 150 ms. Wires and trellis members are identified through multi-view line or cylinder fitting, and a safety standoff of 5-15 cm is enforced during approach, actuation, and retreat. The semantic model itself is structured as a branch skeleton graph, annotated with branch order, diameter, insertion angle, internode length, and spur density. Candidate cuts are generated from this graph and scored by a multi-objective function balancing crop load, minimum spur density, proxies for light distribution, and collision risk.

[0096] Trajectory generation employs model predictive control with minimum-snap profiles while respecting constraints on velocity, acceleration, and jerk. UAV localization fuses LiDAR, inertial, and visual odometry, with loop closures constrained by trellis landmarks for long-term accuracy. Safety measures include propeller guards, geofencing, and tool interlocks that only enable end-effector actuation within a permitted cutting window. Each executed cut is verified against a planned target within a positional tolerance of 2-15 mm and is logged with metadata including pose, tool identity, force / torque profile, and imagery.

[0097] In some embodiments, the system is configured to adapt its actions to different horticultural policies drawn from a library of species- and training-system-specific rulesets. These include, for example, tall-spindle apples, bi-axis pears, open-vase peaches, central-leader cherries, hedgerow almonds, and bilateral cordon grapevines. A per-tree digital identity is maintained, storing plant-specific perception data, actuation locations, end-effector usage, and outcomes for long-term management. Phenological stage detection further enables the system to switch between pruning and thinning modes depending on whether blossoms or swelling buds are observed.

[0098] Operational robustness is enhanced through additional sensing and adaptation. A multispectral or near-infrared sensor estimates local vigor and biases cut selection towardimproved light distribution. Auxiliary illumination ensures flicker-free performance in low- light conditions. Trellis topology is detected and used to constrain feasible branch positions in the semantic model, reducing mapping errors and improving collision checking. The manipulator incorporates compliant or passive joints that provide tolerance for misalignment up to ±15°, while a downward-facing rangefinder maintains a consistent UAV standoff within rows. Wind conditions are estimated in real time, allowing the UAV to select approach orientations that keep reaction wrenches within safe limits.

[0099] The system is designed to operate continuously and at scale. The UAV autonomously returns to a docking station for battery exchange or charging once energy thresholds are reached, and resumes execution at the interrupted task upon redeployment. In multivehicle deployments, UAVs communicate over a peer-to-peer mesh network, sharing maps for collision avoidance and distributing tasks using auction-based or consensus algorithms. Together, these capabilities provide a resilient, adaptive, and scalable solution for precision canopy management across diverse crops and growing systems.

[0100] In some embodiments, the method of autonomously performing canopy management incorporates additional capabilities that enhance both horticultural precision and operational robustness. As part of the decision process, residual bud or spur density for each branch order is constrained to remain within a target range, ensuring balanced regrowth and consistent fruiting potential across the canopy. Following the execution of planned cuts or thins, the plant is rescanned to update the semantic model, with the system recording post-operation canopy characteristics such as estimated leaf-area index and canopy porosity. During pruning, the system applies exclusion masks around detected fruit measuring at least 20 mm to prevent accidental removal of valuable produce.

[0101] Optimization of the execution sequence is achieved using a traveling-salesman-type solver, where edge costs integrate time efficiency, collision risk, and expected horticultural benefit, resulting in an execution order that balances speed with biological outcomes. In situations where occlusions arise, the method enables the UAV to pause, retreat, and then re-approach the target along an alternative path whenever the probability of occlusion exceeds a defined threshold. Each plant is assigned a uniqueidentifier and linked to its geospatial coordinates, allowing for multi-season traceability and integration with digital orchard records.

[0102] The method further adapts to plant biomechanics by estimating branch stiffness through micro-deflection probing, and dynamically modulating approach speed and manipulator compliance in response. This ensures that delicate structures are engaged gently, while more rigid branches are addressed with the necessary force. Finally, the system supports shared autonomy, wherein control may be yielded to a human operator for intervention while the UAV continues to enforce active safety constraints such as collision avoidance, force limits, and geofencing. This combination of automated intelligence and supervised flexibility enables reliable, data-rich canopy management tailored to both horticultural standards and real-world operational challenges.

[0103] Certain embodiments include a computer-readable medium configured to store instructions that provide advanced perception, decision-making, and data management capabilities for autonomous canopy management. These instructions include the implementation of a semantic segmentation network trained to classify and label key plant features such as buds, spurs, fruit, leaves, branches, and surrounding infrastructure including wires. To represent plant geometry, the system performs three-dimensional skeletonization by fitting a minimum-spanning tree or Steiner tree over point-cloud data, with local radius estimation used to characterize branch thickness.

[0104] Horticultural expertise is embedded directly into the software through a parameterized policy table that maps species and training systems to weighted rules, allowing the system to adapt its decisions to apple tall-spindle, peach open-vase, grapevine bilateral cordon, or other forms of canopy architecture. Operational safety is ensured by anomaly detection routines that monitor temporal patterns of plant and UAV motion, halting endeffector actuation upon classification of unexpected movement.

[0105] For long-term record keeping and integration with external data platforms, the instructions further manage compression and synchronization of per-plant logs to a cloud service using differential updates. These logs may include perception data, planned and executed cut points, end-effector states, and outcomes, all tied to a digital identity for each plant. Beyond pruning, the same framework enables execution of a broader set ofcanopy-management tasks, including fruit thinning, shoot thinning, and leaf removal, expanding the system’s utility across multiple phenological stages and crop types.

[0106] The various embodiments disclosed herein can be implemented as hardware, firmware, software, or any combination thereof. Moreover, the software may be implemented as an application program tangibly embodied on a program storage unit or computer readable medium consisting of parts, or of certain devices and / or a combination of devices. The application program may be uploaded to, and executed by, a machine comprising any suitable architecture. Preferably, the machine is implemented on a computer platform having hardware such as one or more central processing units (“CPUs”), a memory, and input / output interfaces. The computer platform may also include an operating system and microinstruction code. The various processes and functions described herein may be either part of the microinstruction code or part of the application program, or any combination thereof, which may be executed by a CPU, whether or not such a computer or processor is explicitly shown. In addition, various other peripheral units may be connected to the computer platform such as an additional data storage unit and a printing unit. Furthermore, a non-transitory computer readable medium is any computer readable medium except for a transitory propagating signal.

[0107] All examples and conditional language recited herein are intended for pedagogical purposes to aid the reader in understanding the principles of the disclosed embodiment and the concepts contributed by the inventor to furthering the art, and are to be construed as being without limitation to such specifically recited examples and conditions. Moreover, all statements herein reciting principles, aspects, and embodiments of the disclosed embodiments, as well as specific examples thereof, are intended to encompass both structural and functional equivalents thereof. Additionally, it is intended that such equivalents include both currently known equivalents as well as equivalents developed in the future, i.e., any elements developed that perform the same function, regardless of structure.

[0108] It should be understood that any reference to an element herein using a designation such as “first,” “second,” and so forth does not generally limit the quantity or order of those elements. Rather, these designations are generally used herein as a convenient method of distinguishing between two or more elements or instances of an element. Thus, areference to first and second elements does not mean that only two elements may be employed there or that the first element must precede the second element in some manner. Also, unless stated otherwise, a set of elements comprises one or more elements.

[0109] As used herein, the phrase “at least one of” followed by a listing of items means that any of the listed items can be utilized individually, or any combination of two or more of the listed items can be utilized. For example, if a system is described as including “at least one of A, B, and C,” the system can include A alone; B alone; C alone; 2A; 2B; 2C; 3A; A and B in combination; B and C in combination; A and C in combination; A, B, and C in combination; 2A and C in combination; A, 3B, and 2C in combination; and the like.

Claims

CLAIMSWhat is claimed is:1 . A system for autonomously any of pruning, thinning, and shoot and leaf removal of fruit bearing plants, comprising: an unmanned aerial vehicle (UAV); a robotic arm attached to the UAV, the robotic arm configured to be moved in three orthogonal directions and to be rotated about three orthogonal axes; a cutting mechanism attached to an end of the robotic arm; the unmanned aerial vehicle, further comprising: one or more cameras sensitive to the visible spectrum, configured to view in real time the surroundings of the UAV and one or more plants; a processing circuitry; a memory, the memory containing instructions that, when executed by the processing circuitry, configure the system to: navigate the UAV to a target plant; capture image data of the target plant; generate, in real time, a three-dimensional semantic model of a target plant, the three-dimensional semantic model including features of the target plant and features of structures adjacent to the target plant based on the image data; compute, in real time, any of a pruning plan, a thinning plan, and a shoot and leaf removal plan, based on the three-dimensional semantic model, horticultural policies and operator-specified objectives for the target plant; and control, in real time, the UAV, the robotic arm, and the cutting mechanism to execute the any of a pruning plan, a thinning plan, and a shoot and leaf removal plan for the target plant.

2. The system of claim 1 , wherein the cutting mechanism is based on any of shears, a saw, and a heating element.

3. The system of claim 1 , wherein the system is powered by any of electricity from one or more batteries, a gaseous fossil fuel, and a liquid fossil fuel.

4. The system of claim 1 , wherein the memory contains further instructions that, when executed by the processing circuitry, configure the system to: isolate some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV.

5. The system of claim 4, wherein propulsion of the UAV isolates some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV.

6. The system of claim 4, wherein an assembly including a series-elastic element isolates some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV.

7. The system of claim 1 , wherein the system further comprises: a second robotic arm attached to the unmanned aerial vehicle, wherein the further instructions that, when executed by the processing circuitry, configure the system to: control, in real time, the second robotic arm to attach to the target plant.

8. The system of claim 4, wherein propulsion of the UAV isolates the UAV during any of pruning, thinning, and shoot and leaf removal of the target plant.

9. The system of claim 1 , wherein the system further comprises wireless communication circuitry, wherein the memory contains further instructions that, when executed by the processing circuitry, configure the system to: collect images of the target plant;digitally tag the target plant for identification; and wirelessly communicate, through the wireless communication circuitry, the tag and archived images of the target plant to a remote data center.

10. A non-transitory computer-readable medium storing a set of instructions for autonomously any of pruning, thinning, and shoot and leaf removal of fruit bearing plants, the set of instructions comprising: one or more instructions that, when executed by one or more processors of a device, cause the device to: navigate a UAV to a target plant; capture image data of the target plant; generate, in real time, a three-dimensional semantic model of a target plant, the three-dimensional semantic model including features of the target plant and features of structures adjacent to the target plant based on the image data; compute, in real time, any of a pruning plan, a thinning plan, and a shoot and leaf removal plan, based on the three-dimensional semantic model, horticultural policies and operator-specified objectives for the target plant; and control, in real time, the UAV, a robotic arm attached to the UAV, and a cutting mechanism attached to the robotic arm to execute the any of a pruning plan, a thinning plan, and a shoot and leaf removal plan for the target plant.

11. A method for autonomously any of pruning, thinning, and shoot and leaf removal of fruit bearing plants, comprising: navigate a UAV to a target plant; capture image data of the target plant; generate, in real time, a three-dimensional semantic model of a target plant, the three-dimensional semantic model including features of the target plant and features of structures adjacent to the target plant based on the image data; compute, in real time, any of a pruning plan, a thinning plan, and a shoot and leaf removal plan, based on the three-dimensional semantic model, horticultural policies and operator-specified objectives for the target plant; andcontrol, in real time, the UAV, a robotic arm attached to the UAV, and the cutting mechanism attached to the robotic arm to execute the any of a pruning plan, a thinning plan, and a shoot and leaf removal plan for the target plant.

12. The method of claim 11 , wherein the cutting mechanism is based on any of shears, a saw, and a heating element.

13. The method of claim 11 , wherein the UAV, the robotic arm, and the cutting mechanism are powered by any of electricity from one or more batteries, a gaseous fossil fuel, and a liquid fossil fuel.

14. The method of claim 11 , further comprising: isolating some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV.

15. The method of claim 14, further comprising: isolating some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV, by a propulsion of the UAV.

16. The method of claim 14, further comprising: isolating some or all of back force of the robotic arm and cutting mechanism during any of pruning, thinning, and shoot and leaf removal of the target plant, from the UAV, by an assembly including a series-elastic element.

17. The method of claim 11 , further comprising: controlling, in real time, a second robotic arm, connected to the UAV, to attach to the target plant.

18. The method of claim 14, further comprising:isolating the UAV during any of pruning, thinning, and shoot and leaf removal of the target plant by propulsion of the UAV.

19. The method of claim 11 , further comprising: collecting images of the target plant; digitally tagging the target plant for identification; and wirelessly communicating, through a wireless communication circuitry, the tag and archived images of the target plant to a remote data center.

Images