Underground space exploration multi-modal bionic robot and its deformation control system and method

Abstract

The application discloses an underground space exploration multi-modal bionic robot and a deformation control system and method thereof, and belongs to the technical field of underground space exploration robots.The multi-modal bionic robot comprises a foldable telescopic main trunk, at least four bionic limbs and a hybrid foot propulsion unit; the deformation control system of the multi-modal bionic robot comprises an environment sensing module, a mode selection module, a control module and a feedback adjustment module, wherein the mode selection module is internally provided with a deep dynamic weighting network, can autonomously decide a flight mode, a walking mode or a climbing mode according to terrain features and output a mode instruction; and the deformation control method of the multi-modal bionic robot sequentially executes the steps of environment sensing, mode decision, structure reconstruction, mode control and closed-loop feedback.The application realizes multi-modal seamless switching and precise control in a complex underground environment through the reconfigurable limb-wing fusion design and the intelligent decision closed loop, and improves the environmental adaptability and survival ability of the robot.

Description

Technical Field

[0001] This invention belongs to the field of underground space exploration robot technology, specifically relating to a multimodal bionic robot for underground space exploration and its deformation control system and method, which is applicable to the exploration of underground or enclosed spaces with multidimensional and unstructured characteristics, such as mine roadways, urban underground smart integrated pipe corridors, and post-disaster ruins. Background Technology

[0002] In recent years, with the increasing demand for the construction of mine tunnels and urban underground smart integrated pipe corridors, the task of underground unstructured space exploration has been growing daily, and multimodal bionic robots have become the most promising breakthrough method. Most existing underground exploration mobile platforms rely on a single motion mode: ground robots cannot fly across ravines or climb vertical cliffs; aircraft cannot operate stably in narrow spaces or under strong airflow interference, and have short endurance; climbing robots are usually slow and have limited obstacle-crossing capabilities. They are ill-suited to underground exploration spaces with complex topologies, insufficient lighting, and numerous unknown obstacles.

[0003] To address the limitations of single-mode locomotion, existing research has attempted to create bimodal robots by simply combining ground movement with flight capabilities, or by superimposing walking with climbing functions. However, these solutions suffer from the following drawbacks: First, the mode-switching mechanisms are complex, and the independent arrangement of multiple power systems results in excessive size and weight, making them prone to mechanical interference or jamming in confined underground spaces. Second, the transitions between different modes lack autonomous decision-making capabilities, relying on manual remote control or preset rules, making it difficult to dynamically adjust according to real-time terrain. Third, the adhesion method between the feet and the rock wall is limited, leading to poor adhesion reliability in slippery or dusty environments, and making slippage and detachment during climbing easy.

[0004] Furthermore, existing deformable robots are mostly designed for single tasks, with complex deformation mechanisms that are prone to getting stuck in compact underground spaces, and lack adaptive climbing and walking capabilities for typical underground media such as pipes. Therefore, there is an urgent need for an underground space exploration robot that can integrate multimodal motion capabilities such as flight, walking, and climbing, and possess autonomous modal decision-making and deformation control capabilities. This invention draws inspiration from the ability of flying squirrels, flying lizards, bats, and other creatures to move flexibly in three-dimensional space, as well as the adhesion mechanism of insect feet, and proposes a reconfigurable limb-wing integrated multimodal biomimetic robot and its deformation control system and method. It is suitable for exploring underground or enclosed spaces with multidimensional and unstructured characteristics, such as urban underground smart integrated pipe corridors and post-disaster ruins, effectively improving the reliability, safety, and environmental adaptability of underground space exploration robots. Summary of the Invention

[0005] This invention addresses the aforementioned problems and overcomes the shortcomings of existing technologies by providing a multimodal biomimetic robot for underground space exploration, along with its deformation control system and method. This effectively improves the reliability, safety, and environmental adaptability of the underground space exploration robot.

[0006] To achieve the above objectives, the present invention adopts the following technical solution.

[0007] In a first aspect, the present invention provides a multimodal biomimetic robot for underground space exploration, comprising: A foldable and retractable main body; At least four bionic limbs connected to the main torso; Hybrid foot propulsion unit installed at the end of each bionic limb; And the control center integrated on the main torso; The bionic limb adopts a modular design, with each limb equipped with multiple active joints. Each active joint is driven by a servo motor, achieving 360° rotation. ° Rotate; The hybrid foot propulsion unit includes at least a vector-deflectable ducted fan propeller, an adaptive adhesive footpad surrounding the ducted fan inlet, and an ankle joint connecting the bionic limb to the hybrid foot propulsion unit.

[0008] As a preferred embodiment of the present invention, the main body adopts a telescopic X-shaped frame or a variable geometric truss structure, and is equipped with an electric telescopic drive mechanism for quickly switching between the extended state and the retracted state; in the extended state, the lateral span of the main body is expanded to improve aerodynamic stability in flight mode, and in the retracted state, the overall volume of the main body is reduced to improve compactness when walking or climbing.

[0009] As another preferred embodiment of the present invention, the adaptive adhesive footpad is a composite structure of a biomimetic microneedle array and an electrically controlled negative pressure suction cup; the ducted fan propeller is controlled to reverse in climbing or walking mode to generate auxiliary adsorption force at the negative pressure suction cup; the electrically controlled negative pressure suction cup is equipped with an adsorption pressure monitoring element and an air replenishment and pressurization mechanism to maintain adsorption stability.

[0010] Secondly, the present invention provides a deformation control system for a multimodal bionic robot for underground space exploration. This deformation control system is applied to the aforementioned multimodal bionic robot for underground space exploration and includes: The environmental perception module, integrated on the main body, is used to collect terrain feature data of underground space and robot body state data. The mode selection module is connected to the environment perception module and has a built-in deep dynamic trade-off network. It is used to autonomously decide whether to adopt flight mode, walking mode or climbing mode based on the terrain feature data and output modal commands. The control module is connected to the mode selection module. The control module runs on the control center and adopts a hierarchical control architecture. It generates corresponding control strategies according to the modal instructions and drives the robot's various execution components, including the active joints of the bionic limbs, the ducted fan thrusters, and the adaptive adhesive footpads. The feedback adjustment module is connected to the environmental perception module, the control module, the active joint of the bionic limb, the ducted fan propeller, and the adaptive adhesive footpad, respectively. It is used to collect the operating parameters of each actuator in real time, compare them with the command parameters output by the control module, calculate the deviation value, and output the adjustment command to correct the control parameters when the deviation value exceeds the preset threshold.

[0011] As another preferred embodiment of the present invention, the deep dynamic tradeoff network includes: The multi-scale perceptual encoder is a hybrid structure of Swin Transformer and 3D graph convolution, where Swin Transformer is used to extract macro-terrain features and 3D graph convolution is used to extract local micro-features, outputting a high-dimensional terrain feature tensor. A multi-objective trade-off decision maker includes a cost module and a differentiable optimization module. The cost module comprises an efficiency cost unit, a stability cost unit, and an energy consumption cost unit, which respectively predict the time cost, stability risk value, and energy consumption per unit distance required to adopt flight, walking, and climbing modes under the current terrain. The differentiable optimization module sums the cost values ​​by weight and selects the mode with the minimum total cost as the main output mode. The weights of each cost unit are dynamically generated according to the task priority. The emergency mode gating controller is a long short-term memory network used to monitor the state deviation under the current execution mode in real time. When there is a drastic fluctuation or failure, it outputs an emergency switching flag and a backup mode suggestion, and forces a switch to the optimal backup mode.

[0012] As another preferred embodiment of the present invention, the environmental perception module includes a lidar, an infrared sensor, a vision camera, an inertial measurement unit, a gas sensor, a temperature sensor, a humidity sensor, and a data storage unit. The lidar is used to detect the outline of the underground space, the distance to obstacles, and the terrain undulations. The infrared sensor is used to identify terrain features and obstacles in dim environments, adapting to underground light-free scenarios. The vision camera is used to collect underground environmental image information and, in conjunction with image recognition algorithms, to identify detailed features of rock wall textures and cracks. The inertial measurement unit is used to monitor the robot's own posture, movement speed, and acceleration. The gas sensor is used to detect the concentration of toxic and harmful gases in the underground space, and when the concentration exceeds a preset threshold, it automatically issues an alarm signal and records the data. The temperature and humidity sensors are used to monitor the temperature and humidity parameters of the underground environment in real time. The data storage unit is used to store environmental perception data and robot operating parameters, supporting real-time data transmission to the ground control terminal to realize remote monitoring and data backtracking of the underground environment. The environmental perception module, based on multi-sensor fusion data, uses filtering algorithms to remove noise and evaluates the terrain features ahead, obstacle distribution, and its own movement status in real time.

[0013] As another preferred embodiment of the present invention, the control module is configured to adopt differentiated control strategies for different motion modes: in flight mode, a flight controller based on dynamic control allocation algorithm is used, combined with a dual closed-loop control structure and a disturbance observer; in walking or climbing mode, a controller based on whole-body dynamics is used, combined with a hierarchical quadratic programming algorithm, taking the adsorption force of the adaptive adhesive footpad as an active control variable, and dynamically adjusting the negative pressure value of the electronically controlled negative pressure suction cup and the reverse rotation speed of the ducted fan.

[0014] As another preferred embodiment of the present invention, the feedback adjustment module also has a fault diagnosis function, which monitors the operating status of the active joints, ducted fan propeller and adaptive adhesive foot pad of the bionic limb in real time. When any of the above-mentioned actuators is detected to be faulty, it immediately feeds back to the control center and triggers the corresponding fault tolerance mechanism and alarm signal.

[0015] As another preferred embodiment of the present invention, the control module also has a fault-tolerant control mechanism: when a certain adaptive adhesive footpad slips or becomes unbalanced, it simultaneously adjusts the adsorption pressure of the remaining adaptive adhesive footpads, starts the ducted fan of the corresponding bionic limb that has slipped to rotate in the forward direction to provide compensating thrust, and adjusts the rotation angle of the active joint of the corresponding bionic limb to correct the robot's center of gravity position; when a single ducted fan or adaptive adhesive footpad fails, the control strategy is automatically reconstructed, and the functions of the failed execution components are compensated by the remaining execution components.

[0016] Thirdly, the present invention provides a deformation control method for a multimodal bionic robot for underground space exploration, which is implemented using the aforementioned deformation control system of the multimodal bionic robot for underground space exploration, and includes the following steps: Step S1, Environmental Perception: Collect terrain feature data and robot body state data of underground space through multi-sensor fusion, and remove noise through filtering; Step S2, Mode Decision: The deep dynamic trade-off network autonomously decides whether to adopt flight mode, walking mode or climbing mode based on terrain feature data, and outputs modal commands. Step S3, Structural Reconstruction: Control the main trunk extension and retraction switching, bionic limb posture adjustment, and hybrid foot propulsion unit working mode switching according to the modal instructions, so that the robot is reconstructed into a configuration that matches the current mode; Step S4, Modal Control: Based on the modal commands, the corresponding control strategy is used to drive the robot to perform flight, walking, or climbing movements; Step S5, Closed-loop feedback: Real-time acquisition of the active joint angle of the bionic limb, the rotation speed of the ducted fan propeller, and the adsorption pressure of the adaptive adhesive footpad, and comparison with the command parameters. When the deviation exceeds the standard, the control parameters are automatically corrected.

[0017] Beneficial effects of this invention: 1. This invention provides a multimodal bionic robot for underground space exploration, along with its deformation control system and method. It integrates bionic technology, multi-sensor fusion technology, automatic control technology, mechanical manufacturing technology, and artificial intelligence technology. Through a foldable and extendable main body and modular bionic limbs, coupled with a hybrid foot propulsion unit, it achieves seamless dynamic switching between flight, walking, and climbing modes, significantly expanding the coverage of underground space exploration. This invention employs an adaptive adhesive footpad with a composite structure of a bionic microneedle array and an electrically controlled negative pressure suction cup, combined with the auxiliary adsorption force generated by the reverse rotation of a ducted fan, significantly improving the robot's adhesion reliability on dry and slippery rock surfaces.

[0018] 2. This invention employs multi-sensor fusion perception and deep dynamic trade-off network decision-making. This network integrates a multi-scale perception encoder combining SwingTransformer and 3D graph convolution, a multi-objective trade-off decision-maker including a differentiable optimization module, and an emergency mode gating system. It possesses self-learning and adaptive capabilities, allowing it to autonomously adapt to unknown underground terrain without human intervention and enabling forced emergency mode switching, significantly improving the robot's survivability. This invention, through a hierarchical control architecture and differentiated control strategies, coupled with fault diagnosis, fault-tolerant control, and emergency mode switching mechanisms, enhances the robot's operational stability, fault tolerance, and survivability, ensuring the continuous and efficient advancement of underground exploration missions. Attached Figure Description

[0019] Figure 1 This is one of the structural schematic diagrams of a multimodal biomimetic robot for underground space exploration according to the present invention (with the main body extended).

[0020] Figure 2 This is the second structural schematic diagram of a multimodal biomimetic robot for underground space exploration according to the present invention (in the state of main body contraction).

[0021] Figure 3 This is a schematic block diagram of the deformation control system of a multimodal bionic robot for underground space exploration according to the present invention.

[0022] Figure 4 This is a schematic diagram of the depth dynamic trade-off network of the deformation control system of a multimodal bionic robot for underground space exploration according to the present invention.

[0023] Figure 5 This is a flowchart illustrating the deformation control method for a multimodal biomimetic robot for underground space exploration according to the present invention.

[0024] The diagram is labeled as follows: 1 is the environmental perception module, 2 is the mode selection module, 3 is the control module, 4 is the feedback adjustment module, 5 is the multi-scale perception encoder, 6 is the multi-objective trade-off decision maker, 7 is the emergency modal gating controller; 11 is the main trunk, 12 is the bionic limb, 13 is the hybrid foot propulsion unit, 14 is the active joint; 1301 is the ducted fan propulsion unit, 1302 is the adaptive adhesive footpad, and 1303 is the ankle joint. Detailed Implementation

[0025] To make the technical problems solved, the technical solutions, and the beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0026] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a multimodal bionic robot for underground space exploration, comprising a foldable and retractable main body 11, four bionic limbs 12 connected to the main body, a hybrid foot propulsion unit 13 installed at the end of each bionic limb 12, and a control center integrated on the main body 11.

[0027] The main body 11 adopts a retractable X-shaped frame made of high-strength, lightweight alloy, balancing structural strength with the overall lightweight requirements of the robot. An electrically operated telescopic drive mechanism is installed inside the main body 11. This mechanism can employ an electric push rod or a lead screw and nut mechanism, enabling rapid switching between extended and retracted states. In the extended state, the lateral span of the main body 11 expands to improve aerodynamic stability in flight mode and counteract the effects of underground airflow disturbances. In the retracted state, the overall volume of the main body shrinks to improve compactness during walking or climbing, facilitating passage through narrow underground passages.

[0028] The bionic limb 12 adopts a modular design, with each bionic limb 12 equipped with multiple active joints 14. Each active joint 14 is driven by a servo motor, enabling 360° rotation. ° It can rotate flexibly, balancing mobility and structural rigidity.

[0029] The hybrid foot propulsion unit 13 includes: a vector-deflectable ducted fan thruster 1301 (deflection angle between 0-90°). ° The device comprises an adaptive adhesive footpad 1302 surrounding the ducted fan inlet, and an ankle joint 1303 connecting the bionic limb 12 and the hybrid foot propulsion unit 13. The ankle joint 1303 is a two-degree-of-freedom rotational joint that can adjust the posture of the ducted fan propeller 1301 and the adaptive adhesive footpad 1302, allowing the footpad to closely adhere to rock surfaces with different inclination angles, while adapting to the force requirements of flight, walking, and climbing modes. The adaptive adhesive footpad 1302 is a composite structure of a bionic microneedle array and an electrically controlled negative pressure suction cup: the bionic microneedles can penetrate tiny gaps in the rock wall to form a mechanical engagement; the electrically controlled negative pressure suction cup provides adhesion through vacuum adsorption. The ducted fan propeller 1301 can be reversed in climbing or walking mode, generating auxiliary adsorption force at the electrically controlled negative pressure suction cup to supplement the adsorption strength of the electrically controlled negative pressure suction cup and prevent the adaptive adhesive footpad from slipping. The electronically controlled negative pressure suction cup has a built-in pressure sensor to monitor the adsorption pressure of the adaptive adhesive footpad in real time; when the adsorption pressure is detected to be lower than the preset threshold, the air replenishment and pressurization mechanism is automatically activated to ensure adsorption stability.

[0030] The control center is electrically connected to the main torso 11, the bionic limbs 12, and the hybrid foot propulsion unit 13. The control center can be an STM32 or ARM series high-performance microcontroller, or other types of low-power smart chips. The control center is used to coordinate and control the extension and folding of the main torso 11, the rotation angle of the active joints 14 of the bionic limbs 12, and the working mode of the hybrid foot propulsion unit 13, so as to achieve seamless dynamic switching between the three modes of flight, walking, and climbing. During the mode switching process, the control center adjusts the motion parameters of the electric extension and retraction drive mechanism of the main torso 11, the active joints 14 of the bionic limbs 12, and the hybrid foot propulsion unit 13 in real time to avoid motion interference and ensure that the switching process is smooth and reliable. Among them, the flight mode is used to cross deep underground pits and move quickly over a large area, the walking mode is used for leveling passages for detection, and the climbing mode is used for detecting complex terrains such as rock walls and steep walls.

[0031] The specific configuration of the flight mode: the main torso 11 is fully extended, and the four bionic limbs 12 are symmetrically extended, with an extension angle of 135 degrees. ° (range 120-150) ° The ducted fan rotates in the forward direction to provide lift and thrust, and the adaptive adhesive footpads 1302 keep the robot in a retracted state, resulting in optimal robot aerodynamic stability.

[0032] The specific configuration of the walking modality: the main torso 11 is fully retracted; the bionic limb 12 is adjusted to a walking posture, and the unfolding angle of the bionic limb 12 is 50 degrees. ° (range 45-60) ° The ducted fan stops rotating; the adaptive adhesive foot pad 1302 unfolds, and walking is achieved by relying on the friction between the adaptive adhesive foot pad 1302 and the ground.

[0033] The specific configuration of the climbing mode: the main body 11 is moderately contracted; the bionic limbs 12 dynamically adjust the unfolding angle according to the rock wall inclination angle; the duct fan reverses to generate an auxiliary adsorption force towards the rock wall; the adaptive adhesive foot pads 1302 are closely attached to the rock wall, microneedles are inserted into the gaps, and the electrically controlled negative pressure suction cups adsorb; the active joints 14 of the bionic limbs 12 rotate in coordination to move the entire body upward, realizing the climbing movement.

[0034] like Figure 3 and Figure 4 As shown in the figure, the present invention provides a deformation control system for a multimodal bionic robot for underground space exploration. The deformation control system is applied to the aforementioned multimodal bionic robot for underground space exploration and includes an environment perception module 1, a mode selection module 2, a control module 3, and a feedback adjustment module 4. The modules are interconnected.

[0035] The environmental perception module 1 is integrated into the main body 11 and is used to collect terrain feature data of the underground space and robot body status data. Specifically, the environmental perception module 1 includes a lidar, an infrared sensor, a vision camera, an inertial measurement unit, a gas sensor, a temperature sensor, a humidity sensor, and a data storage unit. The lidar is used to detect the outline of the underground space, the distance to obstacles, and the terrain undulations. The infrared sensor is used to identify terrain features and obstacles in dim environments, adapting to underground light-free scenarios. The vision camera is used to collect underground environmental image information and, in conjunction with image recognition algorithms, to identify detailed features of rock wall textures and cracks. The inertial measurement unit is used to monitor the robot. The robot's posture, speed, and acceleration are monitored. A gas sensor detects the concentration of toxic and harmful gases in the underground space; when the concentration exceeds a preset threshold, it automatically issues an alarm and records the data. Temperature and humidity sensors monitor the temperature and humidity parameters of the underground environment in real time. A data storage unit stores environmental perception data and robot operating parameters, supporting real-time data transmission to the ground control terminal for remote monitoring and data backtracking of the underground environment. The environmental perception module 1, based on multi-sensor fusion data, uses filtering algorithms to remove noise and assesses the terrain features, obstacle distribution, and its own motion status in real time, providing accurate data support for the mode selection module 2 and the feedback adjustment module 4.

[0036] The mode selection module 2 is connected to the environment perception module 1. The mode selection module 2 incorporates a deep dynamic trade-off network and is used to autonomously decide whether to adopt a flight mode, walking mode, or climbing mode based on the terrain feature data, and output modal commands. The deep dynamic trade-off network includes a multi-scale perception encoder 5, a multi-objective trade-off decision maker 6, and an emergency modal gating device 7. The multi-scale perception encoder 5 is a hybrid structure of Swin Transformer and 3D graph convolution. The Swin Transformer is used to extract macroscopic terrain features, such as gully width, rock wall inclination, and undulation; the 3D graph convolution is used to extract local microscopic features, such as rock wall roughness, crack distribution, and adaptive adhesive footpad contact points. The input is the terrain feature data from the environment perception module 1, and the output is a high-dimensional terrain feature tensor. The multi-objective trade-off decision maker 6 includes a cost module and a differentiable optimization module. The cost module contains an efficiency cost unit, a stability cost unit, and an energy consumption cost unit, which respectively predict the time cost, stability risk value, and energy consumption per unit distance required to adopt the flight, walking, and climbing modes under the current terrain. The differentiable optimization module sums the cost values ​​by weight and selects the mode with the minimum total cost as the main output mode. The weight of each cost unit is dynamically generated according to the task priority.

[0037] Specifically, the differentiable optimization module performs a weighted summation of each cost value, and the total cost calculation formula is as follows: Cost total = W eff C eff + W stab C stab + W ene C ene ;in, Cost total For total cost, C eff For efficiency and cost, C stab To stabilize costs, C ene For energy consumption costs, W eff , W stab , W ene The corresponding weights are used; these weights are dynamically generated based on task priority. The differentiable optimization module performs a weighted summation of each cost value and then selects the total cost. Cost total The minimum mode serves as the main output mode of the multi-objective trade-off decision maker. The emergency mode gating device 7 is a long short-term memory network (LSTM) used to monitor the state deviation under the current execution mode in real time. Its inputs are terrain feature tensors and robot body state data. When there are drastic fluctuations (such as sudden slippage while walking or failure of grip while climbing) or malfunctions, it outputs an emergency switching flag and a backup mode suggestion, forcibly switching to the optimal backup mode, which greatly improves the robot's survivability.

[0038] The control module 3 is connected to the mode selection module 2. The control module 3 runs on the control center. The control module 3 adopts a hierarchical control architecture. The upper decision layer generates the corresponding control strategy according to the modal instructions output by the mode selection module. The lower execution layer drives the robot's execution components through the control center. The execution components include the active joint 14 of the bionic limb 12, the ducted fan thruster 1301, and the adaptive adhesive foot pad 1302. The control module 3 is configured to adopt differentiated control strategies for different motion modes. (1) In flight mode: a flight controller based on dynamic control allocation algorithm is adopted. The attitude controller is designed in combination with a double closed-loop control structure, and a disturbance observer is added to improve the robot's robustness to external disturbances such as underground airflow disturbance and terrain occlusion. The dynamic control allocation algorithm adopts a weighted generalized inverse method to reasonably allocate the lift, thrust, and control torque required for flight to each ducted fan to avoid the ducted fan from saturating. (2) Walking or climbing mode: A controller based on whole-body dynamics is adopted, combined with a hierarchical quadratic programming algorithm, to fully consider the influence of the mass distribution of the bionic limb 12 on the robot's balance, and to optimize the distribution of the driving force of the active joint 14 of the bionic limb 12 and the adsorption force of the adaptive adhesive foot pad 1302 in real time. The adsorption force of the adaptive adhesive foot pad 1302 is used as an active control variable to dynamically adjust the negative pressure value of the electronically controlled negative pressure suction cup and the reverse rotation speed of the ducted fan to ensure the stability of the robot's center of gravity.

[0039] The feedback adjustment module 4 is connected to the environment perception module 1, the control module 3, the active joint 14 of the bionic limb 12, the ducted fan thruster 1301, and the adaptive adhesive footpad 1302. The feedback adjustment module 4 collects the operating parameters of each actuator in real time through the control center (operating parameters include: the angle of the active joint 14 of the bionic limb 12, the rotation speed of the ducted fan thruster 1301, the adsorption pressure of the adaptive adhesive footpad 1302, and the robot posture), compares them with the command parameters output by the control module 3, calculates the deviation value, and if the deviation exceeds a preset threshold (e.g., joint angle deviation > 2), the error is corrected. ° If the adsorption pressure deviation is greater than 10%, an adjustment command is output to correct the control parameters. The feedback adjustment module 4 also has a fault diagnosis function, which monitors the operating status of the active joint 14, ducted fan propeller 1301 and adaptive adhesive foot pad 1302 of the bionic limb 12 in real time. When any of the above-mentioned actuators is detected to be faulty or the adaptive adhesive foot pad 1302 is found to be slipping or unbalanced, it immediately feeds back to the control center and triggers the corresponding fault tolerance mechanism and alarm signal.

[0040] In addition, the control module 3 also has a fault-tolerant control mechanism. After receiving the fault-tolerant signal triggered by the feedback adjustment module 4, the control module 3 executes the following fault-tolerant control strategy: When a certain adaptive adhesive foot pad 1302 slips or becomes unbalanced, it simultaneously executes: ① adjusting the adsorption pressure of the remaining adaptive adhesive foot pads 1302, ② starting the ducted fan of the corresponding bionic limb 12 that has slipped to rotate in the forward direction to provide compensating thrust, ③ adjusting the rotation angle of the active joint 14 of the corresponding bionic limb 12 to correct the robot's center of gravity position; when a single ducted fan thruster 1301 or adaptive adhesive foot pad 1302 fails, the control strategy is automatically reconstructed, and the functions of the failed execution components are compensated by the remaining execution components. For example, if a ducted fan fails in flight mode, the thrust distribution of the remaining three fans is adjusted and the failed bionic limb 12 is retracted.

[0041] Specifically, the training of the deep dynamic trade-off network employs a hybrid strategy of multi-stage reinforcement learning and inverse reinforcement learning. First, it learns basic decision rules from expert trajectories through imitation learning (inverse reinforcement learning). Then, it further optimizes the network using reinforcement learning (such as the PPO algorithm) in a simulation environment. The reward function comprehensively considers detection efficiency, stability, and energy consumption. After training, the network parameters are fixed in the control center.

[0042] like Figure 5 As shown in the figure, an embodiment of the present invention provides a deformation control method for a multimodal bionic robot for underground space exploration, which is implemented using the deformation control system of the aforementioned multimodal bionic robot for underground space exploration, and includes the following steps: Step S1, Environmental Perception: Collect terrain feature data and robot body state data of underground space through multi-sensor fusion, and remove noise through filtering; Step S2, Mode Decision: The deep dynamic trade-off network autonomously decides whether to adopt flight mode, walking mode or climbing mode based on terrain feature data, and outputs modal commands. Step S3, Structural Reconstruction: According to the modal commands, control the extension and retraction of the main body 11, the posture adjustment of the bionic limbs 12, and the switching of the working mode of the hybrid foot propulsion unit 13, so that the robot is reconstructed into a configuration that matches the current mode; in the flight mode, the main body 11 is fully extended and the bionic limbs 12 are in a symmetrical extended state; in the walking mode, the main body 11 is retracted and the bionic limbs 12 are in a walking posture; in the climbing mode, the main body 11 is retracted and the bionic limbs 12 adjust the extension angle according to the rock wall inclination angle.

[0043] Step S4, Modal Control: Based on the modal instructions, the corresponding control strategy is used to drive the robot to perform flight, walking, or climbing movements; in flight mode, a dynamic control allocation algorithm is used to allocate lift and torque; in walking / climbing mode, a whole-body dynamics controller is used to optimize the allocation of joint torque and footpad suction force.

[0044] Step S5, Closed-loop feedback: Real-time acquisition of the active joint 14 angle of the bionic limb 12, the rotational speed of the ducted fan thruster 1301, and the adsorption pressure of the adaptive adhesive footpad 1302 are compared with the command parameters. If the deviation exceeds the standard, the control parameters are automatically corrected. When any actuator malfunction is detected, the fault tolerance mechanism is triggered.

[0045] Through the aforementioned deformation control system and methods, the multimodal autonomous switching and precise control of the multimodal bionic robot for underground space exploration are achieved, maintaining stable detection capabilities in scenarios such as obstruction and complex terrain.

[0046] It is understood that the above specific description of the present invention is only for illustrating the present invention and is not limited to the technical solutions described in the embodiments of the present invention. Those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention to achieve the same technical effect; as long as the use needs are met, they are all within the protection scope of the present invention.

Claims

1. A multimodal biomimetic robot for underground space exploration, characterized in that: include: A foldable and retractable main body; At least four bionic limbs connected to the main torso; Hybrid foot propulsion unit installed at the end of each bionic limb; And the control center integrated on the main torso; The bionic limb adopts a modular design, with each bionic limb having multiple active joints, each of which is driven by a servo motor. The hybrid foot propulsion unit includes at least a vector-deflectable ducted fan propeller, an adaptive adhesive footpad surrounding the ducted fan inlet, and an ankle joint connecting the bionic limb to the hybrid foot propulsion unit.

2. The multimodal biomimetic robot for underground space exploration according to claim 1, characterized in that: The main body adopts a telescopic X-shaped frame or a variable geometry truss structure, and is equipped with an electric telescopic drive mechanism for rapid switching between the extended and retracted states. In the extended state, the lateral span of the main body is expanded to improve aerodynamic stability in flight mode, and in the retracted state, the overall volume of the main body is reduced to improve compactness when walking or climbing.

3. The multimodal biomimetic robot for underground space exploration according to claim 1, characterized in that: The adaptive adhesive footpad is a composite structure of a biomimetic microneedle array and an electrically controlled negative pressure suction cup; the ducted fan propeller is controlled to reverse in climbing or walking mode, generating an auxiliary adsorption force at the negative pressure suction cup; The electrically controlled negative pressure suction cup is equipped with an adsorption pressure monitoring element and a gas replenishment and pressurization mechanism to maintain adsorption stability.

4. A deformation control system for a multimodal bionic robot for underground space exploration, characterized in that: The deformation control system is applied to the multimodal bionic robot for underground space exploration as described in any one of claims 1 to 3, comprising: The environmental perception module, integrated on the main body, is used to collect terrain feature data of underground space and robot body state data. The mode selection module is connected to the environment perception module and has a built-in deep dynamic trade-off network. It is used to autonomously decide whether to adopt flight mode, walking mode or climbing mode based on the terrain feature data and output modal commands. The control module is connected to the mode selection module. The control module runs on the control center and adopts a hierarchical control architecture. It generates corresponding control strategies according to the modal instructions and drives the robot's various execution components, including the active joints of the bionic limbs, the ducted fan thrusters, and the adaptive adhesive footpads. The feedback adjustment module is connected to the environmental perception module, the control module, the active joint of the bionic limb, the ducted fan propeller, and the adaptive adhesive footpad, respectively. It is used to collect the operating parameters of each actuator in real time, compare them with the command parameters output by the control module, calculate the deviation value, and output the adjustment command to correct the control parameters when the deviation value exceeds the preset threshold.

5. The deformation control system of a multimodal bionic robot for underground space exploration according to claim 4, characterized in that: The deep dynamic tradeoff network includes: The multi-scale perceptual encoder is a hybrid structure of Swin Transformer and 3D graph convolution, where Swin Transformer is used to extract macro-terrain features and 3D graph convolution is used to extract local micro-features, outputting a high-dimensional terrain feature tensor. A multi-objective trade-off decision maker includes a cost module and a differentiable optimization module. The cost module comprises an efficiency cost unit, a stability cost unit, and an energy consumption cost unit, which respectively predict the time cost, stability risk value, and energy consumption per unit distance required to adopt flight, walking, and climbing modes under the current terrain. The differentiable optimization module sums the cost values ​​by weight and selects the mode with the minimum total cost as the main output mode. The weights of each cost unit are dynamically generated according to the task priority. The emergency mode gating controller is a long short-term memory network used to monitor the state deviation under the current execution mode in real time. When there is a drastic fluctuation or failure, it outputs an emergency switching flag and a backup mode suggestion, and forces a switch to the optimal backup mode.

6. The deformation control system of a multimodal bionic robot for underground space exploration according to claim 4, characterized in that: The environmental perception module includes a lidar, an infrared sensor, a visual camera, an inertial measurement unit, a gas sensor, a temperature sensor, a humidity sensor, and a data storage unit. The lidar is used to detect the outline of underground space, the distance to obstacles, and the terrain undulations. The infrared sensor is used to identify terrain features and obstacles in dim environments, adapting to underground light-free scenarios. The visual camera is used to collect underground environmental image information and, in conjunction with image recognition algorithms, to identify detailed features of rock wall textures and cracks. The inertial measurement unit is used to monitor the robot's own posture, speed, and acceleration; the gas sensor is used to detect the concentration of toxic and harmful gases in the underground space, and automatically issues an alarm signal and records the data when the concentration exceeds a preset threshold; the temperature sensor and humidity sensor are used to monitor the temperature and humidity parameters of the underground environment in real time; the data storage unit is used to store environmental perception data and robot operating parameters, and supports real-time data transmission to the ground control terminal to realize remote monitoring and data backtracking of the underground environment; the environmental perception module, based on multi-sensor fusion data, uses filtering algorithms to remove noise and evaluates the terrain features ahead, obstacle distribution, and its own motion status in real time.

7. The deformation control system of a multimodal bionic robot for underground space exploration according to claim 4, characterized in that: The control module is configured to employ differentiated control strategies for different motion modes: in flight mode, a flight controller based on a dynamic control allocation algorithm is used, combined with a dual closed-loop control structure and a disturbance observer; in walking or climbing mode, a controller based on whole-body dynamics is used, combined with a hierarchical quadratic programming algorithm, using the adsorption force of the adaptive adhesive footpad as an active control variable to dynamically adjust the negative pressure value of the electronically controlled negative pressure suction cup and the reverse rotation speed of the ducted fan.

8. The deformation control system of a multimodal bionic robot for underground space exploration according to claim 4, characterized in that: The feedback adjustment module also has a fault diagnosis function, which monitors the operating status of the active joints, ducted fan thrusters and adaptive adhesive foot pads of the bionic limb in real time. When any of the above-mentioned actuators is detected to be faulty, it immediately feeds back to the control center and triggers the corresponding fault tolerance mechanism and alarm signal.

9. The deformation control system of a multimodal bionic robot for underground space exploration according to claim 4, characterized in that: The control module also has a fault-tolerant control mechanism: when an adaptive adhesive footpad slips or becomes unbalanced, it simultaneously adjusts the adsorption pressure of the remaining adaptive adhesive footpads, activates the ducted fan of the corresponding bionic limb to rotate forward to provide compensating thrust, and adjusts the rotation angle of the active joint of the corresponding bionic limb to correct the robot's center of gravity position; when a single ducted fan or adaptive adhesive footpad fails, the control strategy is automatically reconstructed, and the functions of the failed execution components are compensated by the remaining execution components.

10. A deformation control method for a multimodal biomimetic robot for underground space exploration, characterized in that: The deformation control system of the multimodal bionic robot for underground space exploration as described in any one of claims 4 to 9 is used, comprising the following steps: Step 1, Environmental Perception: Collect terrain feature data and robot body state data of underground space through multi-sensor fusion, and remove noise through filtering; Step 2, Mode Decision: Using a deep dynamic tradeoff network, the system autonomously decides whether to adopt flight mode, walking mode, or climbing mode based on terrain feature data, and outputs modal commands. Step 3: Structural Reconstruction: Based on the modal instructions, control the main torso extension and retraction switching, the bionic limb posture adjustment, and the hybrid foot propulsion unit working mode switching to reconstruct the robot into a configuration that matches the current mode; Step 4: Modal control: Based on the modal commands, the corresponding control strategy is used to drive the robot to perform flight, walking, or climbing movements; Step 5, Closed-loop feedback: Real-time acquisition of the active joint angles of the bionic limb, the rotation speed of the ducted fan propeller, and the adsorption pressure of the adaptive adhesive footpad, and comparison with the command parameters. When the deviation exceeds the standard, the control parameters are automatically corrected.

Images