Electro-hydraulic servo building robot and migration learning adaptive control method thereof

Abstract

The application discloses an electro-hydraulic servo building robot and a migration learning adaptive control method thereof, and belongs to the technical field of intelligent control of building robots. The robot comprises a hydraulic system, a mechanical system and an electric control system. The hydraulic system adopts a double-pump parallel closed driving structure, is matched with an integrated valve block and an accumulator for pressure compensation and current stabilization. The mechanical system adopts an under-actuated inverted pendulum structure with rear wheels on the ground and front wheels in the air. The electric control system collects working condition information by relying on multiple sensors, and realizes working condition discrimination and safety protection through multiple sub-programs. The application introduces a DDPG migration learning algorithm, migrates network parameters and experience data obtained by training a light-load flat ground source task to a heavy-load complex target task, and quickly adapts to hydraulic nonlinear characteristics and complex working condition disturbances through a hierarchical optimization mode of freezing the bottom layer network and fine-tuning the high layer network. The application effectively improves the heavy-load carrying stability and intelligent operation level of the robot.

Description

Technical Field

[0001] This invention relates to the field of intelligent robot control technology, and in particular to an electro-hydraulic servo construction robot and its transfer learning adaptive control method. Background Technology

[0002] As construction moves towards high-efficiency operation and energy conservation, hydraulic drive systems, with their high power density, are widely used in heavy-duty construction robots. However, the inherent strong nonlinearity and parameter uncertainty of hydraulic systems constitute the core underlying technical challenges: on the one hand, hydraulic pumps suffer from oil leakage, resulting in a non-proportional relationship between output flow and input speed, and a strong nonlinear correlation between leakage and outlet pressure; on the other hand, the elastic modulus of the hydraulic fluid fluctuates dynamically due to system pressure and operating temperature, exhibiting significant uncertainty and being impossible to measure directly. These underlying physical factors greatly increase the difficulty of controller design.

[0003] Due to the combined effects of the aforementioned underlying defects, existing hydraulic construction robots have revealed numerous prominent problems in practical applications. Existing equipment generally suffers from energy redundancy, insufficient control agility, and a lack of adaptive adjustment capabilities. When facing unstructured terrain such as uneven construction surfaces, they are highly susceptible to safety risks such as jamming and overturning.

[0004] Even more critically, directly exploring and experimenting with intelligent control algorithms (such as reinforcement learning) on ​​such large hydraulic equipment faces extremely high trial-and-error costs. Especially under complex working conditions such as heavy loads and rugged terrain, blind exploration of algorithms in the early stages can easily lead to system malfunction due to sudden changes in hydraulic nonlinearity, causing equipment damage or even personnel safety accidents. This makes the rapid iteration and deployment of high-order intelligent algorithms in construction robots extremely difficult.

[0005] In summary, this invention develops an electro-hydraulic servo construction robot for unstructured environments. By adopting a dual-pump parallel hydraulic system layout and integrated structural design, it reduces the robot's size, minimizes energy redundancy, and improves control agility. Addressing the challenge of directly deploying and training algorithms under complex conditions, this invention further designs a transfer learning adaptive control method. This method cleverly transfers the prior knowledge of balance control and hydraulic dynamics characteristics acquired during pre-training in low-risk conditions (such as light loads and flat terrain) to high-risk and complex conditions (such as heavy loads and rugged terrain). This approach not only fundamentally avoids the safety hazards of large equipment exploring from scratch in harsh environments but also precisely overcomes the model adaptation difficulties caused by severe perturbations in hydraulic parameters, breaking through the technical bottlenecks posed by the strong nonlinearity and parameter uncertainty of hydraulic drive systems. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention discloses a highly integrated electro-hydraulic servo construction robot designed for unstructured environments and its transfer learning adaptive control method.

[0007] The specific solution adopted in this invention is as follows:

[0008] An electro-hydraulic servo construction robot for unstructured environments includes a hydraulic system, a mechanical system, and an electronic control system;

[0009] The hydraulic system comprises two independent closed-loop drive units and a shared accumulator. Each closed-loop drive unit includes a hydraulic motor, a large-displacement pump, a small-displacement pump, a servo motor, a solenoid directional valve, a check valve, and a relief valve. Each closed-loop drive unit is equipped with two servo motors and four solenoid directional valves. The large-displacement pump and the small-displacement pump are driven by independent servo motors. The hydraulic motor has two working ports. Each of the two working ports of the large-displacement pump is connected to the corresponding port of the hydraulic motor through a solenoid directional valve. Similarly, each of the two working ports of the small-displacement pump is connected to the corresponding port of the hydraulic motor through a solenoid directional valve. The large-displacement pump and the small-displacement pump form a parallel oil supply structure. The shared accumulator is connected across the two closed-loop drive units and works with the check valve to achieve system oil replenishment and pressure stabilization. The relief valve is used to limit the maximum working pressure of the corresponding closed-loop drive unit circuit to achieve overload protection.

[0010] The mechanical system is an underdriven inverted pendulum structure with the rear wheels on the ground and the front wheels suspended. The mechanical system includes a base, a drive wheel assembly, a large displacement pump assembly, a small displacement pump assembly, and a confluence valve block. The drive wheel assembly, the large displacement pump assembly, and the small displacement pump assembly are all fixedly installed on the base. The confluence valve block is fluidly connected to the large displacement pump assembly and the small displacement pump assembly, respectively, to realize oil circuit convergence and transfer.

[0011] The electronic control system includes an industrial computer, a motion controller, a servo driver, an expansion module, and a sensor group. The industrial computer is connected to the motion controller and the sensor group via Ethernet. The motion controller is connected to the servo driver and the expansion module via EtherCAT (Ethernet Automation Technology) bus. The expansion module is electrically connected to each solenoid directional valve. The sensor group is used to collect hydraulic pressure, machine body tilt angle, and wheel speed operation status signals in real time.

[0012] The electronic control system and hydraulic system are both integrated into the mechanical system. The electronic control system drives the hydraulic system to move by controlling the speed of the servo motor and the on / off state of the electromagnetic reversing valve based on the real-time feedback signal of the sensor group, thereby driving the mechanical system to achieve upright balanced walking and load and terrain adaptive driving.

[0013] Preferably, the confluence valve block is provided with three sets of external oil ports, one set of which is connected to the common accumulator, and the other two sets of external oil ports are respectively connected to the hydraulic motors corresponding to the two closed drive units through connecting pipes; the check valve and the relief valve are detachably connected to the confluence valve block and are configured to correspond to the circuit of the corresponding closed drive unit.

[0014] Preferably, the displacement of the large-displacement pump is not less than 4cc / r, and the displacement of the small-displacement pump is not greater than 2cc / r; the hydraulic motor is a cycloidal hydraulic motor with a rated flow rate of not less than 20L / min.

[0015] Preferably, the electromagnet of the electromagnetic directional valve is driven by DC; the on / off state of the electromagnetic directional valve is indirectly controlled by the motion controller through the expansion module.

[0016] Preferably, the sensor group includes a pressure sensor, an inertial measurement unit, and an angle encoder; the pressure sensor is disposed on the end face of the confluence valve block and is used to collect the hydraulic pressure signal of the hydraulic system; the inertial measurement unit is disposed on the middle of the front side of the wheel-type inverted pendulum structure and is used to detect the body tilt angle signal; the angle encoder is disposed on the drive wheel component of the wheel-type inverted pendulum structure and is used to detect the rear wheel speed signal.

[0017] Preferably, the large displacement pump component includes a first bracket, a first coupling, and a first transition valve block; the first coupling and the first transition valve block are both fixedly mounted on the first bracket, one end of the first coupling is drivenly connected to the corresponding servo motor, and the other end is drivenly connected to the large displacement pump.

[0018] The small displacement pump component includes a second bracket, a second coupling, and a second transition valve block; the second coupling and the second transition valve block are both fixedly mounted on the second bracket, one end of the second coupling is drivenly connected to the corresponding servo motor, and the other end is drivenly connected to the small displacement pump.

[0019] Both the first transition valve block and the second transition valve block are made of carbon structural steel.

[0020] Preferably, the drive wheel assembly includes two drive wheel bodies, a hub reducer, a transition bracket, a coupling, and a crossbeam;

[0021] The crossbeam is a transverse load-bearing component with a hollow inner cavity, and is arranged transversely between the two drive wheel bodies;

[0022] Two transition brackets are symmetrically installed at both ends of the crossbeam, with one end of each transition bracket extending into the hollow cavity of the crossbeam and the other end extending to the outside of the crossbeam.

[0023] Each of the transition brackets is fixedly installed with a hydraulic motor at one end of the crossbeam cavity. The output shaft of the hydraulic motor passes outward through the transition bracket and is connected to the input shaft of the corresponding side hub reducer via a coupling.

[0024] The hub reducer is fixedly mounted on the corresponding drive wheel body. The power output by the hydraulic motor is reduced and increased in torque through the coupling and hub reducer in sequence, and then drives the corresponding drive wheel body to rotate independently.

[0025] Preferably, the industrial control computer is equipped with a control program, which includes an initialization subroutine, a drive subroutine, a working condition adaptation subroutine, and a safety protection subroutine.

[0026] The initialization subroutine is used to load the dynamic link library file of the motion controller, establish the communication link between the industrial computer and the motion controller, and complete the initialization of the EtherCAT bus and sensor group.

[0027] The driving subroutine is used to collect sensor data from the sensor group in real time, convert the collected sensor data into international standard units of measurement, and determine whether an emergency stop condition is triggered. If an emergency stop condition is triggered, the safety protection subroutine is entered; if not, the working condition adaptation subroutine is entered.

[0028] The working condition adaptation subroutine runs after the driving subroutine. It first executes basic balance control to achieve under-actuated driving, and then determines the robot's terrain and load status based on sensor data converted to international standard units of measurement. In turn, it controls the on / off state of the electromagnetic reversing valve and the speed of the servo motor to achieve terrain and load adaptive driving.

[0029] The safety protection subroutine is used to monitor the operating status of the electro-hydraulic servo construction robot in real time and to implement protection functions such as emergency stop, data acquisition stoppage, and communication link disconnection under abnormal working conditions.

[0030] In addition, this invention also discloses the above-mentioned transfer learning adaptive control method for electro-hydraulic servo construction robots in unstructured environments, including the following steps:

[0031] S1: Task Modeling and Decomposition. The handling task of the electro-hydraulic servo construction robot is divided into source tasks and target tasks according to load mass, hydraulic pressure difference, and terrain complexity. Each task is modeled as a Markov decision process including state space, action space, and reward function. The source task is a flat ground condition with a load mass ≤10kg, hydraulic pressure difference ≤3.5MPa, body tilt angle ≤3°, and wheel speed fluctuation ≤5%. The target task is divided into three sub-tasks: heavy-load flat ground, heavy-load unstructured terrain, and full-condition heavy load. Specifically, the heavy-load flat ground sub-task is a flat ground condition with a load mass ≥30kg and hydraulic pressure difference ≥10MPa; the heavy-load unstructured terrain sub-task is based on the heavy-load flat ground sub-task, introducing a 3°~15° slope or a rugged terrain with wheel speed fluctuation >15%; and the full-condition heavy load sub-task is a heavy-load condition including stepped terrain.

[0032] S2: Source task policy pre-training. Under the source task conditions, collect interaction data during the robot's operation. Based on the DDPG (Deep Deterministic Policy Gradient) framework, train the Actor network and Critic network until convergence, and save the optimal network parameters and experience replay pool corresponding to the source task.

[0033] S3: Parameter and experience transfer, the parameters of the Actor network and Critic network after convergence of the source task are used as the initialization parameters of the target task, and the high-quality transition tuples in the experience replay pool of the source task are injected into the experience replay pool of the target task.

[0034] S4: Joint fine-tuning of target task. During the execution of the target task, the working condition adaptation subroutine of the electronic control system is used to determine the robot's load and terrain conditions in real time. Simultaneously, some of the weights of the lower-level networks of the Actor network and Critic network are frozen to retain the general control features obtained from the source task training. Then, the determined load and terrain conditions, as well as the hydraulic pressure difference, body tilt angle, and rear wheel speed signals collected by the sensors, are used as state inputs to perform targeted fine-tuning of the higher-level networks to adapt to the working condition requirements of the target task.

[0035] S5: Deployment and Application. The fine-tuned DDPG strategy is deployed to the industrial control computer. The industrial control computer generates corresponding control commands based on the real-time collected sensor data. By controlling the on / off state of the electromagnetic reversing valve and the speed of the servo motor, the hydraulic system and the mechanical system are driven to work together to achieve adaptive and stable handling of the robot under different load and terrain conditions.

[0036] Preferably, the state space includes fuselage tilt angle error, fuselage angular velocity, hydraulic system pressure difference, rear wheel speed, estimated load, and terrain type; the action space includes servo motor speed command and pump source switching command; the reward function includes tilt angle error penalty, angular velocity penalty, pressure difference change penalty, speed reward, and energy consumption penalty.

[0037] Compared with the prior art, the present invention has at least one of the following advantages or beneficial effects:

[0038] I. Hardware Level: This invention adopts a dual-pump parallel hydraulic drive structure, pump-controlled motor direct drive, and integrated layout of hydraulic components, enabling the system to have high power density (achieving high power output within a limited installation space) and flexible adjustment, effectively solving the technical problems of limited installation space, energy redundancy, and insufficient control agility of construction robots.

[0039] II. Software Level: This invention sets up adaptive control logic based on system pressure difference, tilt angle of the mechanical system in an upright state, wheel speed, and other sensor detection information to construct multi-dimensional judgment indicators of load, terrain, and speed, accurately identifying the robot's real-time working condition. By controlling the on / off state of the electromagnetic reversing valve and the speed of the servo motor, the wheel speed and output torque are adjusted in real time, which has the characteristics of flexible control. It solves the problems of the single speed adjustment mode and rigid control of the hydraulic pump in traditional construction robots, realizing flexible adaptive adjustment of speed and hydraulic flow under different terrains and loads, and the control response is flexible and reliable. With the help of safety protection subroutines, it realizes smooth transportation under heavy loads, rugged terrain, and stepped terrain in unstructured building terrain, and the stability and safety are significantly improved.

[0040] III. Algorithm and Intelligent Control Level: This invention introduces an inter-task policy transfer learning adaptive control method based on Deep Deterministic Policy Gradient (DDPG). Utilizing the Actor and Critic network parameters and experience replay pool data pre-trained by the robot on a lightly loaded, flat source task, it initializes parameters and reuses experience for the target task under heavy-load, unstructured terrain, achieving efficient transfer of prior knowledge for balanced control. This method, combined with the load-terrain joint decision logic of the adaptive working condition subroutine, enables the DDPG policy to quickly adapt to the hydraulic nonlinear response under heavy-load conditions through layered fine-tuning. This solves the technical problems of high zero-training cost, high risk of heavy-load exploration, and insufficient cross-working-condition generalization ability of traditional reinforcement learning algorithms on hydraulic construction robots. It significantly shortens the deployment cycle from debugging to stable operation, improving the robot's intelligence level and operational safety in complex construction scenarios. Attached Figure Description

[0041] The invention, its features, shape, and advantages will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. Like reference numerals denote like parts throughout the drawings. The drawings are not drawn to scale; the emphasis is on illustrating the gist of the invention.

[0042] Figure 1 This is an overall system block diagram of the electro-hydraulic servo construction robot in Embodiment 1 of the present invention;

[0043] Figure 2 This is a schematic diagram of the hydraulic closed-loop principle of the electro-hydraulic servo construction robot in Embodiment 1 of the present invention;

[0044] Figure 3 This is a front view of the mechanical system of the electro-hydraulic servo construction robot in Embodiment 1 of the present invention;

[0045] Figure 4 This is a side view of the mechanical system of the electro-hydraulic servo construction robot in Embodiment 1 of the present invention;

[0046] Figure 5 This is a schematic diagram of the drive wheel component of the electro-hydraulic servo construction robot in Embodiment 1 of the present invention;

[0047] Figure 6 This is a schematic diagram of the large-displacement pump component of the electro-hydraulic servo construction robot in Embodiment 1 of the present invention;

[0048] Figure 7 This is a schematic diagram of the small displacement pump component of the electro-hydraulic servo construction robot in Embodiment 1 of the present invention.

[0049] Figure 8 This is the overall flowchart of the control program for the electro-hydraulic servo construction robot in Embodiment 1 of the present invention.

[0050] Figure 9 This is a flowchart of the control program initialization subroutine for the electro-hydraulic servo construction robot in Embodiment 1 of the present invention.

[0051] Figure 10 This is a flowchart of the control program driving subroutine for the electro-hydraulic servo construction robot in Embodiment 1 of the present invention.

[0052] Figure 11 This is a flowchart of the adaptive control program subroutine for the electro-hydraulic servo construction robot in Embodiment 1 of the present invention.

[0053] Figure 12 This is a flowchart of the safety protection subroutine of the control program for the electro-hydraulic servo construction robot in Embodiment 1 of the present invention.

[0054] Figure 13This is a flowchart of the transfer learning adaptive control method for an electro-hydraulic servo construction robot in an unstructured environment, as described in Embodiment 2 of the present invention.

[0055] Figure 14 This is a flowchart of the transfer learning training process for the electro-hydraulic servo construction robot in Embodiment 2 of the present invention.

[0056] In the diagram: 1. Hydraulic system; 11. Hydraulic motor; 12. Large displacement pump; 13. Small displacement pump; 14. Servo motor; 15. Check valve; 16. Relief valve; 17. Accumulator; 18. Solenoid directional valve; 2. Mechanical system; 21. Drive wheel assembly; 211. Hub reducer; 212. Transition bracket; 213. Coupling; 214. Crossbeam; 22. Bracket; 23. Large displacement pump assembly; 231. First transition 232. Valve block; 233. First bracket; 233. First coupling; 24. Confluence valve block; 25. Base; 26. Small displacement pump component; 261. Second transition valve block; 262. Second bracket; 263. Second coupling; 3. Electrical control system; 31. Industrial computer; 321. Pressure sensor; 322. Inertial measurement unit; 323. Angle encoder; 33. Servo driver; 34. Motion controller; 35. Expansion module. Detailed Implementation

[0057] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but these are not intended to limit the scope of the invention.

[0058] This invention discloses an electro-hydraulic servo construction robot for unstructured environments and its transfer learning adaptive control method, addressing the technical shortcomings of existing construction robots, such as redundant energy loss during installation, strong hydraulic nonlinearity, poor control agility, and weak generalization ability across working conditions. This invention features innovative design across three layers: hardware structure, software working condition discrimination, and intelligent transfer learning algorithm. The hardware employs a dual-pump parallel closed-loop hydraulic drive, an integrated valve block layout, and a transition support transmission structure, achieving high power density, compact structure, and reliable transmission. The software utilizes multi-sensor fusion to achieve load-terrain joint discrimination and adaptive pump source switching. The algorithm introduces the DDPG transfer learning strategy to achieve efficient transfer of light-load prior knowledge to heavy-load complex working conditions, significantly improving the robot's upright balance and heavy-load handling capabilities in unstructured construction scenarios such as slopes, stairs, and rugged dirt roads.

[0059] Example 1:

[0060] Reference Figures 1-7As shown, this embodiment provides an electro-hydraulic servo construction robot for unstructured environments, including a hydraulic system 1, a mechanical system 2, and an electronic control system 3. The hydraulic system 1 is the power unit, the mechanical system 2 is the execution unit, and the electronic control system 3 is the control unit. This robot uses sensors to detect terrain parameters such as slopes (slope ≤ 25°), stairs (step height ≤ 100mm), asphalt roads, and dirt roads, dynamically adjusting the speed of large / small displacement pumps to correct the vertical attitude of the chassis. It can adapt to unstructured terrains such as slopes ≤ 25°, stairs ≤ 100mm in height, dirt roads, and gravel roads, with a rated heavy-duty carrying capacity ≥ 60kg. It can maintain a stable vertical position and walk stably in different terrains, adapting to complex work scenarios.

[0061] The hydraulic system 1 includes two independent closed-loop drive units and a shared accumulator 17. Each closed-loop drive unit corresponds independently to one side of the drive wheel, realizing independent power control of the left and right wheels. Each closed-loop drive unit includes a hydraulic motor 11, a hydraulic pump, two servo motors 14, four solenoid directional valves 18, a check valve 15, and a relief valve 16; among which, the hydraulic pump includes a large displacement pump 12 and a small displacement pump 13. In each closed-loop drive unit, the hydraulic motor 11 has two working ports. The large-displacement pump 12 is independently driven by a servo motor 14, and its two working ports are connected to the corresponding ports of the hydraulic motor 11 through a solenoid directional valve 18. The small-displacement pump 13 is independently driven by another servo motor 14, and its two working ports are also connected to the corresponding ports of the hydraulic motor 11 through a solenoid directional valve 18. The large-displacement pump 12 and the small-displacement pump 13 form a parallel oil supply structure. A shared accumulator 17 is connected across the two closed-loop drive units, and works with the check valve 15 of each closed-loop drive unit to achieve oil replenishment and pressure stabilization. The relief valve 16 of each unit is used to limit the maximum working pressure of the corresponding circuit (i.e., to adjust the set pressure of the hydraulic system 1) and suppress hydraulic nonlinear disturbances. The set pressure range of the relief valve 16 is 15MPa~20MPa. The hydraulic system 1 provides hydraulic power to the mechanical system 2.

[0062] Both the large-displacement and small-displacement pumps are fixed-displacement pumps, each driven by an independent servo motor; the large-displacement pump has a displacement ≥4cc / r, and the small-displacement pump has a displacement ≤2cc / r; the hydraulic motor is a cycloidal hydraulic motor with a rated flow rate ≥20L / min. The large and small-displacement pumps are arranged in parallel, enabling three power output modes: single large pump, single small pump, and dual pumps in parallel.

[0063] Hydraulic system 1 is mounted on mechanical system 2, which is a wheeled inverted pendulum structure with rear wheels on the ground and front wheels suspended. It achieves upright balance by relying on independent drive from the two rear wheels, and maintains upright balance by dynamically adjusting the speed of the rear wheels. The wheeled inverted pendulum structure includes a base 25, a drive wheel assembly 21, a large-displacement pump assembly 23, a small-displacement pump assembly 26, and a confluence valve block 24. The drive wheel assembly 21, the large-displacement pump assembly 23, and the small-displacement pump assembly 26 are all fixedly mounted on the base 25. The confluence valve block 24 has internal channels and is fluidly connected to the large-displacement pump assembly 23 and the small-displacement pump assembly 26, respectively. The drive wheel assembly 21 is equipped with a distributed load support mechanism, combined with the high-power output characteristics of the dual pumps in parallel, ensuring structural stability and sufficient power for heavy-duty transport in a vertical state. The base 25 adopts a high-strength load-bearing structure design, and the overall weight of mechanical system 2 is approximately 150 kg.

[0064] The external oil ports of the merging valve block 24 are divided into three groups. Two groups of external oil ports are located directly behind the merging valve block 24 and are connected to the hydraulic motors 11 corresponding to the two closed drive units through rigid pipelines. The other group of external oil ports is located directly in front of the merging valve block 24 and is also connected to the common accumulator 17 through rigid pipelines. The check valve 15 and the relief valve 16 are detachably connected to the merging valve block 24 and are set in accordance with the circuits of the corresponding closed drive units. The merging valve block 24 greatly reduces the external pipelines and improves the system integration and power density.

[0065] The electrical control system 3 is housed in an electrical box and includes an industrial computer 31, a pressure sensor 321, an inertial measurement unit 322, an angle encoder 323, four servo drives 33, a motion controller 34, and an expansion module 35. The industrial computer 31 communicates with the motion controller 34, pressure sensor 321, inertial measurement unit 322, and angle encoder 323 via Ethernet. The motion controller 34 interacts with the servo drives 33 and expansion module 35 via EtherCAT bus (real-time Ethernet). The expansion module 35 is connected to four electromagnetic directional valves 18 (the power and signal lines of all electromagnetic directional valves 18 are connected to the corresponding interfaces of the expansion module 35). The electromagnets in the electromagnetic directional valves 18 are driven by DC, and the opening and closing of the electromagnetic directional valves 18 is controlled by the motion controller 34. The expansion module 35 provides indirect control; the pressure sensor 321 is located on the end face of the confluence valve block 24 to collect the hydraulic pressure signal of the hydraulic system; the inertial measurement unit 322 is located in the middle of the front side of the wheel-type inverted pendulum structure to detect the body tilt angle signal; the angle encoder 323 is located on the drive wheel component of the wheel-type inverted pendulum structure to detect the rear wheel speed signal; the motion controller 34 receives the signals from the angle encoder 323, the inertial measurement unit 322, and the pressure sensor 321, and uploads the received signals to the industrial control computer 31 via Ethernet, and drives the servo motor 14 to move through the industrial control computer 31.

[0066] It should be noted that the specific connection method of each component of the hydraulic system 1 is as follows: the accumulator 17 is connected to two confluence valve blocks 24 through rigid pipes, and the relief valve 16 and the check valve 15 are installed in the confluence valve blocks 24 by threaded connection. In each closed drive unit, one servo motor 14 is connected to the large displacement pump 12 through the first coupling 233, and the other servo motor 14 is connected to the small displacement pump 13 through the second coupling 263. The large displacement pump 12 and the small displacement pump 13 are respectively connected to the confluence valve block 24 through the first transition valve block 231 and the second transition valve block 261. The large displacement pump 12 and the small displacement pump 13 jointly drive a hydraulic motor 11, which is controlled by two solenoid directional valves 18. Taking a single hydraulic motor 11 as an example:

[0067] Case 1: If only the small displacement pump 13 is working, the electromagnets of the two solenoid directional valves 18 corresponding to the large displacement pump 12 will be de-energized and disconnected, and the oil will only come out from the small displacement pump 13.

[0068] Case 2: Only the large displacement pump 12 is working, and the electromagnets of the two solenoid directional valves 18 corresponding to the small displacement pump 13 are energized and disconnected, so the oil only comes out from the large displacement pump 12.

[0069] Situation 3: Both the large displacement pump 12 and the small displacement pump 13 are working. At this time, neither of the solenoid directional valves 18 is energized, and oil can come out from both the large displacement pump 12 and the small displacement pump 13 at the same time.

[0070] Scenario 4: Neither the large displacement pump 12 nor the small displacement pump 13 is working, all solenoid directional valves 18 are energized, and neither the large displacement pump 12 nor the small displacement pump 13 produces oil at the same time.

[0071] The electromagnet is energized and de-energized by the industrial control computer calling the library functions provided by the motion controller manufacturer, sending the command to the motion controller, and then the motion controller outputs high / low levels through the digital output module in the EtherCAT bus drive expansion module.

[0072] The above operating steps ultimately enable the hydraulic pump to draw in and discharge oil.

[0073] It should also be noted that the confluence valve block 24 is connected to the hydraulic motor 11 through a rigid pipe fitting. The hydraulic motor 11 is connected to the external oil port of the confluence valve block 24 through a connecting pipe. The relief valve 16 is used to ensure that the working pressure of the hydraulic system 1 does not exceed the maximum working pressure. The accumulator 17 is used to provide oil source and a small charging pressure, and works with the check valve 15 to replenish oil into the hydraulic system 1, thereby realizing the circulation of the hydraulic system 1.

[0074] like Figure 2 The diagram shown is a detailed schematic of the hydraulic system 1 of this invention. The hydraulic system 1 employs a complex circuit configuration and includes a hydraulic motor 11, a large-displacement pump 12, a small-displacement pump 13, a servo motor 14, a check valve 15, a relief valve 16, an accumulator 17, and a solenoid directional valve 18. For a single hydraulic motor 11, the load demand is determined by the pressure sensor 321 in the hydraulic system. Under light load, the large-displacement pump is shut off, and only the small-displacement pump is driven, avoiding energy redundancy. Under heavy load, the two pumps are automatically switched to parallel operation to achieve precise matching between load and output, improving energy utilization efficiency. The parallel operation of the two hydraulic pumps can achieve a larger flow rate of oil output, maintaining the rated flow rate of the hydraulic motor at 20-25 L / min. Combined with the distributed load support logic of the mechanical system, the balance state is calibrated in real time through the pressure sensor 321, the inertial measurement unit 322, and the angle encoder 323, enabling stable vertical transport of loads ≥60 kg.

[0075] The large-displacement pump 12 is connected to the first transition valve block 231, and the small-displacement pump 13 is connected to the second transition valve block 261. Both the first transition valve block 231 and the second transition valve block 261 are made of carbon structural steel. The one-way valve 15 outputs oil from the accumulator 17 for replenishment.

[0076] In the closed loop of hydraulic system 1, the output flow of the large-displacement pump 12 and the small-displacement pump 13 is adjusted by changing the speed of the servo motor 14, thereby regulating the flow of the hydraulic motor 11 and ultimately adjusting the output speed of the hydraulic motor 11. The accumulator 17 in hydraulic system 1 works with the check valve 15 to replenish oil to hydraulic system 1. The servo motor 14 drives the hydraulic pump to rotate, and the oil output by the hydraulic pump then drives the hydraulic motor 11 to rotate. The hydraulic pump is a constant displacement pump, and the flow of hydraulic system 1 is adjusted by changing the speed of the servo motor 14.

[0077] The present invention enables the transmission and control of hydraulic energy through a hydraulic system 1. The hydraulic system 1, in conjunction with the electronic control system 3, drives the mechanical system 2 to perform corresponding actions. An electromagnetic directional valve 18 controls the opening and closing states of the large-displacement pump 12 and the small-displacement pump 13, thus forming oil circuits with different flow rates. These circuits drive the hydraulic motor 11, causing its rotational speed to vary. This satisfies the hydraulic transmission requirements of the hydraulic system 1 under various operating conditions, including precise speed tracking based on a single pump and high-speed operation based on multiple pumps in parallel.

[0078] The oil flow rate in the oil circuit can be categorized into four cases:

[0079] Case 1: When the small displacement pump 13 is working and the large displacement pump 12 is not working, the small displacement pump 13 works with the confluence valve block 24 to achieve oil suction and discharge. The oil from the small displacement pump 13 directly drives the hydraulic motor 11 to rotate, achieving precise tracking of the single pump speed (horizontal linear velocity 0.05~0.20m / s).

[0080] Scenario 2: When the large displacement pump 12 is working and the small displacement pump 13 is not working, the large displacement pump 12 works with the confluence valve block 24 to achieve oil suction and discharge. The oil from the large displacement pump 12 directly drives the hydraulic motor 11 to rotate, achieving precise tracking of the single pump speed. At the same speed, it can achieve precise tracking of a higher speed (horizontal linear velocity 0.1~0.40m / s).

[0081] Case 3: The large displacement pump 12 and the small displacement pump 13 are connected in parallel and work simultaneously. The oil is sucked and discharged by the servo motor 14 rotating in the same direction. The oil finally merges in the merging valve block 24, driving the hydraulic motor 11 to rotate (the maximum horizontal linear velocity can reach 1.75m / s).

[0082] Case 4: The large displacement pump 12 and the small displacement pump 13 are connected in parallel and neither of them is working. The hydraulic motor 11 stops rotating, which is equivalent to braking the entire device.

[0083] like Figure 3 and Figure 4The figures shown are the front view and side view of the mechanical system 2 of the present invention, respectively, specifically illustrating the structures such as the drive wheel component 21, bracket 22, large displacement pump component 23, confluence valve block 24, base 25, and small displacement pump component 26. The hydraulic fluid in the hydraulic system 1 is output to the hydraulic motor 11 through the confluence valve block 24, and the hydraulic motor 11 drives the drive wheel component 21, thereby rotating the hub reducer 211 in the drive wheel component 21. This structure, combined with the control program, can achieve balance control through the upright movement of the entire mechanism. Utilizing the pressure sensor 321, inertial measurement unit 322, and angle encoder 323 in the electronic control system 3, balance control of the under-driven system is achieved, thus realizing precise closed-loop control.

[0084] In mechanical system 2, the drive wheel component 21 is mounted on the bracket 22. The drive wheel component 21 and the bracket 22 together form a four-wheel chassis, such as... Figure 5 As shown, the drive wheel assembly 21 includes two drive wheel bodies, a hub reducer 211, a transition bracket 212, a coupling 213, and a crossbeam 214. The crossbeam 214 is a transverse load-bearing member with a hollow inner cavity, arranged transversely between the two drive wheel bodies (i.e., drive wheels are provided at both ends of the crossbeam 214). Two transition brackets 212 are symmetrically installed at both ends of the crossbeam 214. One end of each transition bracket 212 extends into the hollow inner cavity of the crossbeam 214, and the other end extends to the outside of the crossbeam 214 and connects to the tire of the corresponding drive wheel body. A hydraulic motor 11 of the hydraulic system is fixedly installed at one end of each transition bracket 212 that extends into the inner cavity. The output shaft of the hydraulic motor 11 passes outward through the transition bracket 212 and is connected to the corresponding side drive wheel body via the coupling 213. The input shaft of the hub reducer 211 is connected to the coupling 213, which is mounted on the transition bracket 212. The hub reducer 211 is fixedly mounted on the surface of the corresponding drive wheel body. The hydraulic motor 11 drives the hub reducer 211 to rotate via the coupling 213 and the transition shaft. The transition shaft is a self-machined shaft, which is connected to the hub reducer 211 and the coupling 213 with a flat key to achieve the final transmission of the power source. That is, the power output by the hydraulic motor 11 is reduced and increased in torque by the coupling 213 and the hub reducer 211 in sequence, and then drives the corresponding drive wheel body to rotate independently. Specifically, the hub reducer 211 is a planetary gear reducer, the tire is made of rubber, and the angle encoder 323 is an incremental encoder, and the output signal is a 5V differential signal (ABZ three-phase).

[0085] Figure 6This is a schematic diagram of the structure of the large displacement pump component 23. The large displacement pump component 23 includes a first bracket 232, a first transition valve block 231, and a first coupling 233. The first coupling 233 and the first transition valve block 231 are both fixedly mounted on the first bracket 232. One end of the first coupling 233 is connected to the corresponding servo motor 14, and the other end is connected to the large displacement pump 12 for transmission. The large displacement pump 12 is connected to the first transition valve block 231.

[0086] Figure 7 This is a schematic diagram of the structure of the small displacement pump component 26. The small displacement pump component 26 includes a second bracket 262, a second transition valve block 261, and a second coupling 263. The second coupling 263 and the second transition valve block 261 are both fixedly mounted on the second bracket 262. One end of the second coupling 263 is connected to the servo motor 14, and the other end is connected to the small displacement pump 13 for transmission. The small displacement pump 13 is connected to the second transition valve block 261.

[0087] In the electronic control system 3, the servo driver 33 and expansion module 35 connected to the servo motor 14 communicate with the motion controller 34 via real-time Ethernet. Simultaneously, the servo driver 33 sends control signals to the servo motor 14 to control its speed. The system pressure in the hydraulic system 1 can be acquired by the pressure sensor 321 in the electronic control system 3. In the mechanical system 2, the tilt angle signal of the four-wheel chassis in a vertical position is acquired by the inertial measurement unit 322 in the electronic control system 3, and the displacement signal of the four-wheel chassis in a vertical position is acquired by the angle encoder 323. The electromagnetic directional valve 18 directly controls the electromagnet in the electromagnetic directional valve 18 to be energized or de-energized through the digital output section of the expansion module 35, thereby realizing the parallel connection or disconnection of the large-displacement pump 12 and the small-displacement pump 13. The motion controller 34 communicates with the industrial computer 31 via an Ethernet port. The industrial computer 31 is responsible for program writing and secondary development of the motion controller, writing a balance control algorithm to ultimately achieve real-time balance of the mechanical system 2. The electromagnet in the electromagnetic reversing valve 18 is driven by DC; the opening and closing of the electromagnetic reversing valve 18 is indirectly controlled by the motion controller 34 through the expansion module 35; the pressure range of the overflow valve 16 is set to 15MPa~20MPa.

[0088] In the electronic control system 3, the servo drive 33 uses real-time Ethernet communication and is connected to the real-time Ethernet interface of the motion controller 34 via a network cable. Simultaneously, the motion controller 34 connects to the corresponding interface of the industrial computer 31 via an Ethernet interface to achieve network communication. The industrial computer 31 is responsible for program writing and secondary development of the motion controller 34, forming a control closed loop. The expansion module 35 is used to acquire the pressure signal from the pressure sensor 321 and feed it back to the control program.

[0089] The aforementioned electronic control system 3 and hydraulic system 1 are both integrated into mechanical system 2. Based on the feedback signals from the sensor group, electronic control system 3 drives hydraulic system 1 to move by controlling the speed of servo motor 14 and the on / off state of electromagnetic reversing valve 18, thereby driving mechanical system 2 to achieve upright balanced walking and load and terrain adaptive driving.

[0090] Continue to refer to, for example Figure 1 As shown, the specific working principle of the electro-hydraulic servo construction robot in this embodiment is as follows: the industrial control computer 31 in the electronic control system 3 programs the motion controller 34. The motion controller 34 directly collects the attitude and displacement data of the inertial measurement unit 322 and the angle encoder 323, and indirectly collects the pressure data of the pressure sensor 321 through the expansion module 35. This is used to determine the current working status of the mechanical system 2 and the hydraulic system 1, and sends control commands to the servo driver 33. Finally, by changing the speed of the servo motor 14, the large displacement pump 12 and the small displacement pump 13 are driven to rotate. The oil output by the large displacement pump 12 and the small displacement pump 13 then drives the hydraulic motor 11 to rotate in both directions, thus driving the mechanical system 2 to move. Among them, the pressure sensor 321 outputs an analog signal (0-10V), the angle encoder 323 is used to measure the displacement of the hydraulic wheel inverted pendulum, and the inertial measurement unit 322 (using an inertial measurement device that outputs TTL level signals, TTL level being transistor-transistor logic level) measures the angle of the mechanical system 2 when it is moving upright.

[0091] To further resolve the aforementioned technical issues, refer to Figures 8-12 The industrial control computer of the present invention is used to run a control program, which includes an initialization subroutine, a drive subroutine, a working condition adaptation subroutine, and a safety protection subroutine. The basic balance control of the electro-hydraulic servo construction robot is embedded in the working condition adaptation subroutine.

[0092] The specific implementation process of the initialization subroutine is as follows:

[0093] The first step is to load the library function file (dll) provided by the motion controller manufacturer and establish an Ethernet network communication link between the industrial computer 31 and the motion controller 34. If the network connection fails to be established, the program will be forcibly terminated and directly enter the safety protection subroutine.

[0094] The second step is for the motion controller 34 to perform the initialization configuration of the EtherCAT bus, which specifically involves detecting all devices on the EtherCAT bus and issuing corresponding address numbers and related configuration files.

[0095] The third step involves the subroutine initializing and calibrating the pressure sensor 321, inertial measurement unit 322, and angle encoder 323. For the pressure sensor 321, the industrial computer 31 reads its data via the motion controller 34 through the expansion module 35 (implemented through the motion controller's internal program), clears the historical pressure data from the previous calibration process (zeroing), and then re-receives the pressure value currently read by the pressure sensor. For the angle encoder mounted on the wheel, its current angle signal is read through the DB26 interface of the motion controller 34 and directly forwarded to the industrial computer 31 by the motion controller (implemented through the motion controller's internal program). Upon receiving the data, the industrial computer automatically clears the previous calibration data and re-receives the current encoder data as the initial pressure value. For the inertial measurement unit, its initialization is pre-programmed and burned by the manufacturer, so no additional initialization is required. After completing these three steps, the program enters the driver subroutine flow.

[0096] The specific implementation process of the driver subroutine is as follows:

[0097] Step 1: Read all sensor data, including data from the inertial measurement unit (IMU), pressure sensor, and angle encoder. IMU data includes triaxial angular velocity and triaxial acceleration, while the angle encoder measures the current wheel angle. The pressure sensor measures the pressure in the high-pressure and low-pressure chambers of the hydraulic system. The commands to read the pressure sensor and the angle encoder are executed by the industrial PC using library functions provided by the motion controller manufacturer. For the pressure sensor, the motion controller uses EtherCAT, via an extension module, to read the pressure sensor data and upload it to the industrial PC. For the angle encoder, the motion controller directly reads the angle signal through its DB26 interface and uploads it directly to the industrial PC. Reading data from the IMU is accomplished by the industrial PC using library functions provided by the IMU manufacturer.

[0098] Step 2: Convert to international standard units. The raw data from the pressure sensor is in kPa, and the raw data from the angle encoder is in degrees. For ease of calculation, the pressure sensor unit is converted to Pa, and the angle encoder unit is converted to rad. For the inertial measurement unit, since the raw data read from the manufacturer's library functions is already in international standard units, no additional conversion is needed.

[0099] Step 3: Data Output and Code Switching. Data output involves sending the sensor data values ​​read and converted in Step 2 to the code switching module within the operating condition adaptation subroutine and the driver subroutine. If the user manually terminates the program or presses the emergency stop button (using the external emergency stop switch), the system quickly switches to the safety protection subroutine; otherwise, it directly enters the operating condition adaptation subroutine.

[0100] The specific implementation process of the operating condition adaptation subroutine is as follows:

[0101] The first step is to achieve basic balance control, which is accomplished using an existing adaptive underactuated balance control method. In this step, the rear wheels of the mechanical system are on the ground, the front wheels are suspended, and upright balance control is achieved with a reference balance angle of 180 degrees. During this period, all solenoid directional valves are in the active state, and all hydraulic pumps are operational. Generally, the rotation direction of the rear wheels is consistent with the tilting direction of the mechanical system. When the mechanical system tilts clockwise, the rear wheels will rotate clockwise at the desired speed / torque according to the speed calculated by the balance control, and vice versa. During balance control, the rear wheels will maintain the tilt angle error of the mechanical system (the error between the current value measured by the inertial measurement unit and the balance angle) within ±5 degrees.

[0102] During balance control, the preset speed zones are as follows:

[0103] In the low-speed operating range (0.05~0.5m / s): the motor speed of the small displacement pump is finely adjusted first, while the large displacement pump is in the off state (all corresponding solenoid valves are energized, and the servo motor stops). The speed of the servo motor corresponding to the small displacement pump is controlled within the range of 50~300RPM. Based on the body attitude deviation (angle deviation ≤0.5°) fed back by the inertial measurement unit, the output torque of the hydraulic motor is adjusted by the oil output from the small displacement pump to achieve stable walking of the robot body in a vertical state.

[0104] Medium-speed operating range (0.5~1m / s): The large displacement pump is used for independent drive first. At this time, the small displacement pump is in the cut-off state (the corresponding solenoid valve is completely cut off and the corresponding servo motor stops rotating). At this time, the servo motor speed is adjusted to 300~600RPM to match the high-efficiency output range of the large displacement pump. While ensuring driving efficiency, the wheel set speed stability is monitored in real time by the angle encoder (speed fluctuation coefficient ≤3%) to avoid speed fluctuation in the medium-speed range.

[0105] High-speed operating range (1~2.75m / s): In this mode, the dual pumps operate in parallel. All electromagnetic reversing valves are activated, and all corresponding electromagnets are de-energized. The servo motor speed increases to 600~2000RPM. The power demand for high-speed driving is met by the combined output of the large-displacement pump and the small-displacement pump. At the same time, the terrain slope data and the body tilt deviation collected by the inertial measurement unit are used to calibrate the motor speed distribution ratio in real time to ensure the speed tracking accuracy (speed error ≤±0.05m / s) and operational stability under different operating conditions.

[0106] All of the above motor speed range settings are implemented in the industrial control computer, using library functions provided by the motion controller manufacturer to forcibly set the upper and lower speed limits.

[0107] The second step is load and terrain determination. The control program uses the inertial measurement unit and pressure sensor to determine the load and the current terrain:

[0108] 1. Terrain Determination: The robot's three-axis angular velocity and three-axis acceleration are collected in real time using an inertial measurement unit. The tilt angle, angular velocity, and attitude angle data are calculated using a Kalman filter algorithm (current technology). This data is then combined with real-time wheel rotation speed, cumulative displacement, and rotation speed fluctuation coefficient collected by the angle encoder. A comprehensive assessment of the terrain type is then performed (the rotation speed fluctuation coefficient is the percentage of the current speed calculation result divided by the previous speed calculation result): When the robot's tilt angle is ≤3° and the wheel rotation speed fluctuation coefficient is ≤5%, the terrain is determined to be flat ground or asphalt road; when the tilt angle is between 3° and 15° and the rotation speed fluctuation coefficient is ≤8°, the terrain is determined to be flat ground or asphalt road. When the tilt angle is less than or equal to 5° but the rotational speed fluctuation coefficient is greater than 15%, it is determined to be a normal slope; when the tilt angle is less than or equal to 5° but the rotational speed fluctuation coefficient is greater than 15%, it is determined to be a rugged dirt road; when the rear wheel displacement shows a step-like sudden change and the inertial measurement unit detects high-frequency attitude oscillation (high-frequency attitude oscillation refers to the absolute value of the difference between the pitch angle measured by the inertial measurement unit and the reference balance angle (180 degrees) in the electronic control system continuously hovering between 0 and 5 degrees, and high frequency refers to the signal frequency of this difference being greater than or equal to 1 Hz), and the hydraulic system pressure value also shows a significant periodic zigzag change, then the robot's current position is determined to be a stepped terrain, ensuring the accuracy and reliability of terrain recognition.

[0109] 2. Load Determination: The hydraulic system pressure signal is collected in real time by a pressure sensor, and the load condition is determined based on the pressure difference range. When the pressure difference is ≤3.5MPa, it is determined to be a light load condition of ≤10kg; when the pressure difference is in the range of 3.5~10MPa, it is determined to be a medium to heavy load condition of 10~30kg; when the pressure difference is ≥10MPa, it is determined to be a heavy load condition of >30kg. This provides a quantitative basis for matching the power output of large and small displacement pumps.

[0110] The third step, based on the judgment results above and the preset speed zones, is to perform adaptive switching of the pump unit's operating mode and precise control of the electromagnetic directional valve:

[0111] 1. Flat / normal slope operation under no-load conditions: No additional operation is required at this time. By default, all servo motors rotate normally under basic balance control, all solenoid directional valves are open, and all electromagnets are de-energized. If the user has additional requirements for the driving speed under balance control, the solenoid directional valve directly connected to the large displacement pump or small displacement pump can be manually disconnected and the corresponding servo motor can be stopped from rotating, or the corresponding large and small displacement pump speeds can be manually adjusted directly.

[0112] 2. Load Adaptive Process:

[0113] On flat roads and ordinary slopes, when an additional load is placed on the mechanical structure of the construction robot, the industrial control computer first sends commands using the library functions provided by the motion controller manufacturer to open all solenoid valves (de-energize the electromagnets). At this time, all hydraulic pumps are connected to the hydraulic system. After detecting a change in load (the pressure change range of the hydraulic system is greater than 3MPa per unit time), the basic balance control algorithm automatically calculates the total motor speed required to maintain balance control using existing adaptive laws. This speed is then evenly distributed to the servo motor groups corresponding to the two wheels (large displacement pump + small displacement pump as one group) to achieve basic balance control after loading. Afterwards, the program further adjusts the speed and pump power on / off based on the ground conditions and load type, as follows.

[0114] 2.1 Light load and flat road surface / normal slope working condition: After completing the basic balance control of load, terrain recognition and load application, it will switch to the balanced medium speed zone first to maintain the basic driving function on flat road surface under load.

[0115] 2.2 Medium and heavy load and flat road surface / ordinary slope working conditions: After completing the basic balance control of load, terrain recognition and load application, the robot is forcibly switched to the low speed zone after balance, maintaining the ground travel under heavy load on flat road surface under load, ensuring basic balance while avoiding the robot overturning due to excessive load.

[0116] In the above state, if the user needs to increase the driving speed, they can manually switch the on / off state of the solenoid directional valve and control the motor speed to increase the speed. At this time, the control program will receive the relevant program adjustment instructions sent by the user on the industrial control computer and execute them using library functions.

[0117] When the system is in complex terrain conditions, regardless of whether it is unloaded or loaded, the load adaptive operation logic is as follows:

[0118] 2.3 Complex terrain conditions under no-load conditions (rugged dirt roads, steps, etc.): The system automatically switches to balance control in the low-speed zone. That is, firstly, the large-displacement pump is shut off in advance, and then balance control is performed.

[0119] 2.4 Complex Terrain Conditions with Load (Rugged dirt roads, steps, etc.): In this case, the system will prioritize judging the terrain where the robot is located. When in complex terrain conditions and the system detects a sudden change in hydraulic system pressure exceeding 3MPa per unit time, the system recognizes that an additional load is being added. When the load is to be added to the robot, the robot will maintain balance control in a low-speed range. After confirming that the load has been added, the system will continue to travel in a low-speed range. During this time, manual speed adjustment by the user is prohibited: regardless of whether the solenoid valve is manually adjusted or the motor speed is adjusted, the system will forcibly prevent the execution of such adjustment functions. When the control program receives any speed adjustment command sent by the user through the industrial control computer, the program will use the library functions provided by the motion controller manufacturer to forcibly prevent the execution of the user-sent speed adjustment command.

[0120] The specific implementation process of the security protection subroutine is as follows:

[0121] Step 1: Forcefully stop the motor rotation. The industrial computer 31 uses the library functions provided by the motion controller 34 manufacturer to send a command to force the servo motor to stop rotating. This command is sent to each servo driver via EtherCAT through the motion controller 34. After receiving the command, the servo driver adjusts the voltage of the servo motor, ultimately causing the motor to stop.

[0122] Step 2: Forcefully stop receiving all sensor data. The industrial computer 31 uses the library functions provided by the motion controller 34 and the inertial measurement unit manufacturer to send a command to forcefully stop receiving all sensor data. At this time, all sensor readings inside the industrial computer will be 0.

[0123] Step 3: Disconnect the network connection between the industrial computer and the motion controller. The industrial computer 31 uses the library functions provided by the motion controller 34 manufacturer to disconnect from the motion controller and end all operations.

[0124] It should be noted that in this embodiment, the data reading of the motor drive, solenoid valve on / off, encoder, and pressure sensor is uniformly implemented by library functions provided by the motion controller manufacturer; the data reading function of the inertial measurement unit is implemented by the underlying driver code provided by its manufacturer, and this embodiment only calls the above functions. The initialization configuration of the EtherCAT bus in the initialization subroutine is the relevant code provided by the motion controller manufacturer. The servo motor speed control command is implemented by the industrial control computer calling the relevant library function (provided by the motion controller manufacturer) to send relevant commands to the motion controller. The motion controller then sends control commands to the servo driver through its built-in program (provided by the manufacturer). Finally, the servo driver controls the voltage of the servo motor to achieve forward / reverse rotation or stop. The solenoid valve on / off command is implemented by the industrial control computer calling the relevant library function (provided by the motion controller manufacturer) to send relevant commands to the motion controller. The motion controller then sends control commands to the expansion module through its built-in program (provided by the manufacturer). Finally, the digital output module of the expansion module controls the electromagnet of the solenoid directional valve to be energized / de-energized, thus realizing the on / off function of the solenoid valve. The pressure sensor data is transmitted from the industrial control computer to the motion controller by calling relevant library functions (provided by the motion controller manufacturer). After the motion controller sends relevant instructions, it reads the corresponding points of the analog input module in the expansion module through its built-in program (provided by the manufacturer). After reading the corresponding pressure sensor data (this function is provided by the motion controller manufacturer), the reading results are finally returned to the industrial control computer.

[0125] Load assessment involves calculating the pressure difference between the inlet and outlet ports of the hydraulic system and determining whether the pressure difference is within a specified range, which mainly includes light load, medium load, and heavy load. Terrain assessment involves evaluating the wheel speed calculated by the inertial measurement unit and encoder. Terrain includes four types: flat land, ordinary slope, rugged dirt road, and stepped terrain. The overall implementation of adaptive working conditions is based on the combined results of load and terrain assessments, and adjustments are made accordingly. Wheel speed is obtained by differentiating the angle value measured by the encoder and converting it to linear velocity. Basic balance control uses an underactuated balance control method based on adaptive control, which is existing technology.

[0126] Example 2:

[0127] like Figure 13 and 14 As shown, this invention addresses the problem that the control algorithm of the aforementioned electro-hydraulic servo construction robot requires parameter retuning or even model retraining when facing sudden load changes or terrain shifts. It provides a transfer learning adaptive control method for electro-hydraulic servo construction robots in unstructured environments, based on the method described in Embodiment 1. This method can run in conjunction with the aforementioned working condition adaptation subroutine or can be directly deployed after offline training. The main steps are as follows:

[0128] S1: Task modeling and decomposition. The handling task of the electro-hydraulic servo construction robot is divided into source tasks T according to the load mass, hydraulic pressure difference, and terrain complexity. s With target task T T Each task is modeled as containing a state space. Action space and reward function Markov decision process; where the source task T s For flat terrain conditions with a load capacity ≤10kg, hydraulic pressure differential ≤3.5MPa, fuselage tilt angle ≤3°, and wheel speed fluctuation ≤5%; target task T T The system is divided into three sub-tasks: heavy-load flat ground, heavy-load unstructured terrain, and full-condition heavy load. Specifically: the heavy-load flat ground sub-task is a flat ground condition with a load mass ≥30kg and a hydraulic pressure difference ≥10MPa; the heavy-load unstructured terrain sub-task introduces a 3°–15° slope or a rugged terrain with wheel speed fluctuations >15% on top of the heavy-load flat ground sub-task; and the full-condition heavy load sub-task includes heavy-load conditions with stepped terrain. This state space... Includes fuselage tilt error, fuselage angular velocity, hydraulic system differential pressure, rear wheel speed, estimated load, and terrain type; operating space. Includes servo motor speed commands and pump source switching commands; reward function. It includes tilt angle error penalty, angular velocity penalty, differential pressure change penalty, velocity bonus and energy consumption penalty.

[0129] Specifically, Markov Decision Process (MDP) modeling is used. For the balance and handling control of electro-hydraulic servo construction robots, an MDP model is established, mainly consisting of:

[0130] Define the state vector. Based on the sensor configuration of this invention, the state vector... Defined as:

[0131]

[0132] in The fuselage pitch angle error measured by the inertial measurement unit 322; It is the pitch angular velocity; The pressure difference between the high and low pressure chambers is measured by pressure sensor 321 (characterizing the current load torque). The linear velocity of the wheel assembly is calculated by the differential of the angle encoder 323.

[0133] The load mass is estimated based on decoupling logic. Considering that the dynamic pressure difference of the hydraulic system strongly couples the inertial torque caused by mechanical acceleration with the nonlinear frictional force of hydraulic components, a steady-state feature extraction mechanism is introduced to ensure the accuracy of the mass estimation: the system only samples the average pressure difference within a certain time window when the inertial measurement unit 322 detects that the fuselage is in static upright equilibrium (i.e., the tilt angle error is continuously within ±1° and the angular velocity is close to 0), or is in a quasi-steady-state condition of uniform straight-line travel on flat ground (wheel speed fluctuation coefficient ≤2%). Under this quasi-steady-state condition, the influence of the system's accelerating inertial force and dynamic frictional resistance is minimized, and the measured pressure difference is mainly used to overcome the gravitational torque generated by the effective load. The purely physical load mass calibrated and estimated based on the steady-state pressure difference is then calculated. By inputting the network as an invariant feature state quantity, the distribution shift caused by dynamic disturbances to network observations is effectively avoided.

[0134] This represents the terrain category feature value (e.g., 0: flat land, 1: slope, 2: rugged).

[0135] Define the action vector. Action vector For mixed outputs of continuous and discrete variables, this invention maps them uniformly to a continuous control quantity:

[0136]

[0137] in The normalized target speed command for the servo motor (-1 to 1 corresponds to reverse rotation to full forward speed); For pump source mode coefficient (when The "only small displacement pumps operate" option is specified in the corresponding operating condition adaptation subroutine. (Corresponding to "dual pumps operating in parallel"). The industrial control computer calls library functions based on this output to control the servo driver 33 and the solenoid directional valve 18.

[0138] In this invention, the state space is the set of all possible values ​​for fuselage tilt angle error, angular velocity, hydraulic differential pressure, wheel speed, estimated load, and terrain type; the state vector is a set of specific state data collected by sensors at any given time. The action space is the entire executable range of servo motor speed and pump source switching commands; the action vector is a set of specific control commands output by the controller in real time. The DDPG network outputs the corresponding action vector based on the state vector at the current moment, achieving adaptive control.

[0139] In terms of specific deployment and control handover logic, this invention adopts a residual compensation architecture (directly adding the traditional adaptive calculation results and the DDPG results) to achieve a smooth fusion of the traditional adaptive law and the DDPG strategy. Under normal operating conditions with flat and light loads, the underlying layer mainly relies on the existing adaptive underactuated balance control law embedded in the operating condition adaptation subroutine to output the basic motor speed command; when facing large load nonlinear changes or complex terrain, the target speed command output by the DDPG network... Instead of directly taking over or overriding the underlying commands, the DDPG agent denormalizes them and uses them as feedforward compensation, which is then superimposed on the basic speed command generated by the underlying adaptive control law. Through this hybrid control architecture, the DDPG agent focuses on addressing the dynamic residuals and disturbances caused by hydraulic nonlinearity and complex terrain, thus ensuring that the hydraulic system possesses both the fundamental absolute stability of traditional control and the cross-condition adaptive agility based on learning algorithms. The industrial control computer, based on the final superimposed output, calls library functions to control the servo driver 33 and the solenoid directional valve 18.

[0140] Define a reward function. To guide the robot to maintain balance and track the target speed with minimal energy consumption, the following design is proposed:

[0141]

[0142] in:

[0143] Tilt error penalty. When the tilt angle of the robot body deviates from the equilibrium position, a negative reward is generated, and the penalty increases with the greater the deviation (square relationship), thereby incentivizing the robot to maintain an upright posture. is the tilt angle penalty weighting coefficient, which is a positive real number.

[0144] This is an angular velocity penalty term. It generates a negative reward when the robot body experiences rapid shaking or violent swaying, used to suppress attitude oscillations and make the robot's movement more stable. ω is the angular velocity penalty weighting coefficient, which is a positive real number.

[0145] Pressure differential sudden change penalty. A negative reward is generated when there is a large sudden change in the pressure difference between the high and low pressure chambers of the hydraulic system. This is used to prevent sudden changes in output torque caused by hydraulic shock, and plays a role in anti-vibration and protecting hydraulic components. is the differential pressure penalty weighting coefficient, which is a positive real number.

[0146] This is a speed bonus. When the fuselage tilt angle error is less than a preset safety threshold... (That is, when the robot is roughly upright) apply the linear velocity of the current wheelset. A proportional positive reward is used to incentivize the robot to walk at the speed required to complete the task while maintaining a stable upright posture. This is an indicator function; it takes the value 1 if the condition within the parentheses is true, and 0 otherwise. This is the speed reward weighting coefficient, which is a positive real number.

[0147] This is an energy consumption penalty item. A penalty is applied to the absolute value of the motor speed command; the higher the speed, the greater the penalty, in order to suppress excessive control output and save energy consumption of the hydraulic system while ensuring balance. is the energy consumption penalty weighting coefficient, which is a positive real number.

[0148] In summary, this reward function aims to minimize tilt angle error (balance index), suppress pressure differential abrupt changes (anti-shake), reward forward speed when stable, and penalize excessive motor speed commands to conserve energy.

[0149] S2: Source Task Strategy Pre-training. Under the source task conditions, the interaction data during the robot's operation is collected in real time through the hydraulic system, mechanical system, and electronic control system of the electro-hydraulic servo construction robot. Based on the DDPG framework, the Actor network is trained. With Critic Network Perform offline or online training until convergence, and save the optimal network parameters corresponding to the source task. , and experience replay pool .

[0150] Specifically: Actor Network The input is a state vector. (6-dimensional), the output is an action vector. (2D). The network structure adopts a three-layer fully connected layer design: the first hidden layer contains 256 neurons, the second hidden layer contains 128 neurons, both using the ReLU activation function; the output layer contains 2 neurons, using the Tanh activation function. Constrained in the interval [-1, 1], The constraint is in the interval [0, 1].

[0151] Critic Network The input is a state vector. With action vectors The concatenated vector (8 dimensions in total) outputs a single Q-value (scalar). The network structure also adopts a three-layer fully connected layer design, with 256 and 128 neurons in the hidden layers, using the ReLU activation function, and the output layer has a linear output.

[0152] Pre-training was performed on the source task (defined as a load weight ≤ 10kg and flat terrain). During training, the DDPG algorithm updated the Critic network parameters by minimizing the Temporal Difference Error (TD Error) and updated the Actor network parameters using the deterministic policy gradient theorem. The gradient direction of the Actor network update is as follows:

[0153]

[0154] in:

[0155] The objective function J of the Actor network is related to its parameters θ. μ The gradient represents the direction and magnitude of the Actor network parameter updates. The goal of reinforcement learning is to maximize the cumulative reward; therefore, the parameters are updated along this gradient direction.

[0156] N: Expectation operator, which means calculating the average on a mini-batch of data randomly sampled from the experience replay pool, instead of using all the data.

[0157] The state-action value function output by the Critic network represents the estimated long-term cumulative reward that can be obtained according to the current policy after taking action a in state s.

[0158] θ is the partial derivative of the Q-value output by the Critic network with respect to action 'a', taking its value at a = μ(s). It indicates to the Actor which direction a fine-tuning of the action in the current state would most quickly improve the Critic's score. Q This is the set of parameters for the Critic network (i.e., the weights and biases of neurons in each layer).

[0159] Actor policy network, input state s, output deterministic action a. θ μ These are the parameters of the Actor network (i.e., the weights and biases of neurons in each layer).

[0160] The action output by the Actor network is related to its own parameter θ μ The gradient represents how the parameters of the Actor network change to adjust the output action in a specified direction.

[0161] The meaning of this formula is: the gradient of the Q-value given by the Critic network with respect to the action (i.e., telling the Actor which direction to adjust the action to obtain higher value), multiplied by the gradient of the Actor network's own parameters with respect to the output, and the result is the update direction of the Actor's parameters. The expectation E represents the average value calculated on a batch of sampled data.

[0162] Once the robot's tilt angle error in the source task stabilizes within ±2° and the cumulative reward converges, save the code files related to the Actor network weights and the Critic network weights. Then, recycle the source task experience pool, denoted as... It contains all the state transition tuples accumulated during the training process.

[0163] S3: Parameter and experience transfer, using the converged Actor and Critic network parameters from the source task as the initialization parameters for the target task. , ; and replay the source task experience pool High-quality transition tuples within Inject into the experience replay pool of the target task The specific steps are as follows:

[0164] The current state refers to the state vector observed by the robot at time t, which includes the body pitch angle error, pitch angular velocity, pressure difference between the high and low pressure chambers of the hydraulic system, wheel set linear velocity, estimated load mass, and terrain category feature value.

[0165] The robot is in state s t Below, the action vector output by the Actor network and executed after exploring noise perturbations contains the normalized servo motor target speed command ω. cmd And the pump source mode coefficient β. The industrial control computer uses this to control the servo motor speed and the on / off state of the solenoid directional valve.

[0166] The reward function at time t. The robot is performing action a. t Then, the environment provides an immediate reward value based on a preset reward function. This value measures the reward received in state s. t Take action a t The immediate effectiveness or poorness of the effect.

[0167] The state at the next moment. During the execution of action a. t Afterwards, the new state vector observed by the robot at time t+1 has a structure similar to s. t They are exactly the same. It records the actual changes in the environment after the action is executed, which is used by the Critic network to calculate the temporal difference target value.

[0168] When the target task is triggered, the following transfer learning process is executed:

[0169] Step 1: Parameter Inheritance. Create a new Actor network and a Critic network for the target task, and directly load the convergence weights saved in the source task as initialization parameters, i.e., the initial parameters of the target task. , This ensures that the target task possesses prior knowledge of the source task's balance control from the outset, avoiding blind exploration from scratch caused by random initialization.

[0170] Step Two: Bottom-level Freezing and High-level Fine-tuning. Because the bottom-level dynamic characteristics (such as the basic trend of gravity torque compensation and the linear speed regulation response of the hydraulic pump) of the electro-hydraulic servo system are highly similar under different load conditions, while the high-level control strategies (such as the pump displacement switching logic under heavy load and the selection of speed ranges under different terrains) need to be differentiated according to specific working conditions. Therefore, this invention adopts a layered migration strategy:

[0171]

[0172] This formula means that the parameters of the first two layers of the source task's Actor network (i.e., the first and second hidden layers, responsible for extracting general dynamic features) are frozen. During the fine-tuning training of the target task, these two layer parameters remain unchanged and do not participate in gradient updates; only the network weights of the third layer (i.e., the output decision layer) are trained and updated. This layered transfer learning method retains the general feature extraction capabilities already learned by the source task while allowing the target task to quickly adapt to new working conditions at the decision level.

[0173] Experience replay pool from the source mission In the process, the transition tuples (i.e., pitch angle errors) of the successfully transported trajectory are selected. (Those state transition data that are close to 0 and have high reward values) are injected into the experience replay pool of the target task. In this process, the data is mixed with new data collected from the target task itself. This source task data provides the Critic network with a priori standards for reference states and actions, which can accelerate the convergence of the Q-value function (i.e., the value function).

[0174] S4: Joint fine-tuning of target task. During the execution of the target task, the working condition adaptation subroutine running through the electronic control system first determines the robot's load and terrain status in real time; simultaneously, some of the weights of the lower-level networks of the Actor network and Critic network are frozen to retain the general control features obtained from the source task training; then, the load status and terrain status determined in real time by the working condition adaptation subroutine, as well as the hydraulic pressure difference, body tilt angle, and rear wheel speed signals collected by the sensors, are used as status inputs to perform targeted fine-tuning of the higher-level networks to adapt to the working condition requirements of the target task.

[0175] Specifically, this involves the safety exploration of adaptive logic under operating conditions. During the fine-tuning exploration phase of transfer learning, the DDPG algorithm typically adds a certain intensity of random exploration noise to actions to explore better strategies. However, under heavy-load conditions, random action commands may cause mismatched torque output from the hydraulic motor, resulting in significant vehicle tilting or even overturning. To ensure safety during the training process, this invention introduces a safety shielding mechanism. In each control cycle, the judgment conditions are as follows:

[0176] If pressure difference >10MPa (i.e., currently under heavy load conditions), and tilt angle If the absolute value is greater than 5° (i.e., the vehicle body tilt exceeds the preset safety limit), then a forced intervention action will be output (such as only using a small displacement pump and low speed to correct the attitude).

[0177] Once the tilt angle error of the vehicle body returns to within 5°, control is returned to the DDPG strategy. This safety shielding mechanism ensures the physical safety of transfer learning during training on a real robot, preventing equipment damage caused by exploration noise.

[0178] To prevent destructive pressure pulses (hydraulic shocks) in the hydraulic system caused by sudden changes in servo motor speed due to hard switching of control during safety intervention and withdrawal, this invention introduces a smooth transition mechanism between DDPG strategy commands and conservative safety action commands. Specifically, a dynamic switching weight coefficient is set at the algorithm deployment level. ( The actual motor action commands issued to the underlying actuators are as follows: ,in The safety action command is calculated for the low-speed zone balance control logic in the operating condition adaptation subroutine. When the safety shielding mechanism is triggered, the weight... The weights linearly decrease from 1 to 0 within a set transition period (e.g., 0.5 seconds) to achieve smooth weight reduction; similarly, when the attitude recovers to a safe threshold (tilt error ≤ 5°) and control is ready to be handed over, the weights... Then, following the first-order inertial filtering curve, it smoothly transitions back to 1 from 0. This motion continuity constraint design fundamentally eliminates the risks of mechanical jitter and hydraulic shock caused by discrete state machine switching, significantly improving the safety of heavy-duty construction robot deployment.

[0179] S5: Deployment and Application. The fine-tuned DDPG strategy is deployed to the industrial control computer. The industrial control computer generates corresponding control commands based on the real-time collected sensor data. By controlling the on / off state of the electromagnetic reversing valve and the speed of the servo motor, the hydraulic system and the mechanical system are driven to work together to achieve adaptive and stable handling of the robot under different load and terrain conditions.

[0180] Those skilled in the art should understand that variations can be implemented by combining existing technology with the above embodiments. Such variations do not affect the essence of the present invention and will not be elaborated upon here.

[0181] The preferred embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and the devices and structures not described in detail should be understood as being implemented in a conventional manner in the art. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the present invention. This does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the present invention's technical solutions still fall within the protection scope of the present invention.

Claims

1. An electro-hydraulic servo building robot for unstructured environment, characterized by, This includes hydraulic systems, mechanical systems, and electrical control systems; The hydraulic system comprises two independent closed-loop drive units and a shared accumulator. Each closed-loop drive unit includes a hydraulic motor, a large-displacement pump, a small-displacement pump, a servo motor, a solenoid directional valve, a check valve, and a relief valve. Each closed-loop drive unit is equipped with two servo motors and four solenoid directional valves. The large-displacement pump and the small-displacement pump are driven by independent servo motors. The hydraulic motor has two working ports. Each of the two working ports of the large-displacement pump is connected to the corresponding port of the hydraulic motor through a solenoid directional valve. Similarly, each of the two working ports of the small-displacement pump is connected to the corresponding port of the hydraulic motor through a solenoid directional valve. The large-displacement pump and the small-displacement pump form a parallel oil supply structure. The shared accumulator is connected across the two closed-loop drive units and works with the check valve to achieve system oil replenishment and pressure stabilization. The relief valve is used to limit the maximum working pressure of the corresponding closed-loop drive unit circuit to achieve overload protection. The mechanical system is an underdriven inverted pendulum structure with the rear wheels on the ground and the front wheels suspended. The mechanical system includes a base, a drive wheel assembly, a large displacement pump assembly, a small displacement pump assembly, and a confluence valve block. The drive wheel assembly, the large displacement pump assembly, and the small displacement pump assembly are all fixedly installed on the base. The confluence valve block is fluidly connected to the large displacement pump assembly and the small displacement pump assembly, respectively, to realize oil circuit convergence and transfer. The electronic control system includes an industrial computer, a motion controller, a servo driver, an expansion module, and a sensor group. The industrial computer is connected to the motion controller and the sensor group via Ethernet. The motion controller is connected to the servo driver and the expansion module via EtherCAT bus. The expansion module is electrically connected to each solenoid directional valve. The sensor group is used to collect hydraulic pressure, machine body tilt angle, and wheel speed operation status signals in real time. The electronic control system and hydraulic system are both integrated into the mechanical system. The electronic control system drives the hydraulic system to move by controlling the speed of the servo motor and the on / off state of the electromagnetic reversing valve based on the real-time feedback signal of the sensor group, thereby driving the mechanical system to achieve upright balanced walking and load and terrain adaptive driving.

2. The electro-hydraulic servo construction robot for unstructured environments according to claim 1, characterized in that, The merging valve block is provided with three sets of external oil ports. One set of external oil ports is connected to the common accumulator, and the other two sets of external oil ports are respectively connected to the hydraulic motors corresponding to the two closed drive units through connecting pipes. The check valve and the relief valve are detachably connected to the merging valve block and are set in accordance with the circuit of the corresponding closed drive unit.

3. The electro-hydraulic servo construction robot for unstructured environments according to claim 1, characterized in that, The displacement of the large-displacement pump is not less than 4cc / r, and the displacement of the small-displacement pump is not greater than 2cc / r; the hydraulic motor is a cycloidal hydraulic motor with a rated flow rate of not less than 20L / min.

4. The electro-hydraulic servo construction robot for unstructured environments according to claim 1, characterized in that, The electromagnet of the electromagnetic reversing valve is driven by DC; the on / off state of the electromagnetic reversing valve is indirectly controlled by the motion controller through the expansion module.

5. The electro-hydraulic servo construction robot for unstructured environments according to claim 1, characterized in that, The sensor group includes a pressure sensor, an inertial measurement unit, and an angle encoder; the pressure sensor is located on the end face of the confluence valve block and is used to collect the hydraulic pressure signal of the hydraulic system; the inertial measurement unit is located in the middle of the front side of the wheel-type inverted pendulum structure and is used to detect the tilt angle signal of the body; the angle encoder is located on the drive wheel component of the wheel-type inverted pendulum structure and is used to detect the rear wheel speed signal.

6. The electro-hydraulic servo construction robot for unstructured environments according to claim 1, characterized in that, The large displacement pump component includes a first bracket, a first coupling, and a first transition valve block; the first coupling and the first transition valve block are both fixedly mounted on the first bracket, one end of the first coupling is connected to the corresponding servo motor, and the other end is connected to the large displacement pump. The small displacement pump component includes a second bracket, a second coupling, and a second transition valve block; the second coupling and the second transition valve block are both fixedly mounted on the second bracket, one end of the second coupling is drivenly connected to the corresponding servo motor, and the other end is drivenly connected to the small displacement pump. Both the first transition valve block and the second transition valve block are made of carbon structural steel.

7. The electro-hydraulic servo construction robot for unstructured environments according to claim 1, characterized in that, The drive wheel assembly includes two drive wheel bodies, a hub reducer, a transition bracket, a coupling, and a crossbeam; The crossbeam is a transverse load-bearing component with a hollow inner cavity, and is arranged transversely between the two drive wheel bodies; Two transition brackets are symmetrically installed at both ends of the crossbeam, with one end of each transition bracket extending into the hollow cavity of the crossbeam and the other end extending to the outside of the crossbeam. Each of the transition brackets is fixedly installed with a hydraulic motor at one end of the crossbeam cavity. The output shaft of the hydraulic motor passes outward through the transition bracket and is connected to the input shaft of the corresponding side hub reducer via a coupling. The hub reducer is fixedly mounted on the corresponding drive wheel body. The power output by the hydraulic motor is reduced and increased in torque through the coupling and hub reducer in sequence, and then drives the corresponding drive wheel body to rotate independently.

8. The electro-hydraulic servo construction robot for unstructured environments according to claim 1, characterized in that, The industrial control computer is equipped with a control program, which includes an initialization subroutine, a driver subroutine, a working condition adaptation subroutine, and a safety protection subroutine. The initialization subroutine is used to load the dynamic link library file of the motion controller, establish the communication link between the industrial computer and the motion controller, and complete the initialization of the EtherCAT bus and sensor group. The driving subroutine is used to collect sensor data from the sensor group in real time, convert the collected sensor data into international standard units of measurement, and determine whether an emergency stop condition is triggered. If an emergency stop condition is triggered, the safety protection subroutine is entered; if not, the working condition adaptation subroutine is entered. The working condition adaptation subroutine runs after the driving subroutine. It first executes basic balance control to achieve under-actuated driving, and then determines the robot's terrain and load status based on sensor data converted to international standard units of measurement. In turn, it controls the on / off state of the electromagnetic reversing valve and the speed of the servo motor to achieve terrain and load adaptive driving. The safety protection subroutine is used to monitor the operating status of the electro-hydraulic servo construction robot in real time and to implement protection functions such as emergency stop, data acquisition stoppage, and communication link disconnection under abnormal working conditions.

9. A transfer learning adaptive control method for an electro-hydraulic servo construction robot for unstructured environments based on any one of claims 1-8, characterized in that, Includes the following steps: S1: Task Modeling and Decomposition. The handling task of the electro-hydraulic servo construction robot is divided into source tasks and target tasks according to load mass, hydraulic pressure difference, and terrain complexity. Each task is modeled as a Markov decision process including state space, action space, and reward function. The source task is a flat ground condition with a load mass ≤10kg, hydraulic pressure difference ≤3.5MPa, body tilt angle ≤3°, and wheel speed fluctuation ≤5%. The target task is divided into three sub-tasks: heavy-load flat ground, heavy-load unstructured terrain, and full-condition heavy load. Specifically, the heavy-load flat ground sub-task is a flat ground condition with a load mass ≥30kg and hydraulic pressure difference ≥10MPa; the heavy-load unstructured terrain sub-task is based on the heavy-load flat ground sub-task, introducing a 3°~15° slope or a rugged terrain with wheel speed fluctuation >15%; and the full-condition heavy load sub-task is a heavy-load condition including stepped terrain. S2: Source task strategy pre-training. Under the source task conditions, collect the interaction data during the robot's operation. Based on the DDPG framework, train the Actor network and Critic network until convergence, and save the optimal network parameters and experience replay pool corresponding to the source task. S3: Parameter and experience transfer, the parameters of the Actor network and Critic network after convergence of the source task are used as the initialization parameters of the target task, and the high-quality transition tuples in the experience replay pool of the source task are injected into the experience replay pool of the target task. S4: Joint fine-tuning of target task. During the execution of the target task, the working condition adaptation subroutine of the electronic control system is used to determine the robot's load and terrain conditions in real time. Simultaneously, some of the weights of the lower-level networks of the Actor network and Critic network are frozen to retain the general control features obtained from the source task training. Then, the determined load and terrain conditions, as well as the hydraulic pressure difference, body tilt angle, and rear wheel speed signals collected by the sensors, are used as state inputs to perform targeted fine-tuning of the higher-level networks to adapt to the working condition requirements of the target task. S5: Deployment and Application. The fine-tuned DDPG strategy is deployed to the industrial control computer. The industrial control computer generates corresponding control commands based on the real-time collected sensor data. By controlling the on / off state of the electromagnetic reversing valve and the speed of the servo motor, the hydraulic system and the mechanical system are driven to work together to achieve adaptive and stable handling of the robot under different load and terrain conditions.

10. The transfer learning adaptive control method for an electro-hydraulic servo construction robot for unstructured environments according to claim 9, characterized in that, The state space includes fuselage tilt angle error, fuselage angular velocity, hydraulic system pressure difference, rear wheel speed, estimated load, and terrain type; the action space includes servo motor speed command and pump source switching command; the reward function includes tilt angle error penalty, angular velocity penalty, pressure difference change penalty, speed reward, and energy consumption penalty.

Images