An autonomous navigation method and system for flying robots in microgravity environments

Abstract

The application provides a flying robot autonomous navigation method and system for a microgravity environment, and belongs to the field of robot autonomous navigation. In order to solve the problem that the autonomous navigation of the flying robot in the existing space station cabin seriously depends on preset identification or manual remote control and the obstacle avoidance algorithm is insufficient, the application realizes accurate environment perception and positioning through multi-sensor fusion and anti-interference positioning algorithm, and can effectively deal with the interference of shielding or floating objects. A multi-hypothesis pose estimation algorithm is used to improve the integrity and accuracy of the environment map modeling; through the topological guided Hybrid A* path planning algorithm combined with the multi-stage PID trajectory tracking control, the robot can more agilely and efficiently respond to the motion instructions issued by astronauts during flight; the robot has a hierarchical dynamic obstacle avoidance function, through different obstacle avoidance strategies, early avoidance, active deceleration and reverse obstacle avoidance are realized, so that the system can flexibly cope with diversified obstacles in the microgravity environment.

Description

Technical Field

[0001] This invention relates to the field of autonomous navigation technology for robots, and more specifically, to an autonomous navigation method and system for flying robots in microgravity environments. Background Technology

[0002] The space station serves as a strategic platform for long-term human habitation in space, conducting deep space exploration technology verification, and carrying out cutting-edge scientific experiments. With the increasing number of space station modules and the growing variety of experimental equipment, the tasks undertaken by astronauts are becoming increasingly diversified and sophisticated. These tasks include the precise assembly and debugging of equipment inside and outside the modules, rapid emergency response in high-risk areas, and complex operations such as collaborative operation and maintenance management of multiple modules. The demands on operational precision, response speed, and collaborative efficiency far exceed human physiological limits. Manual operation has a low tolerance for error and high risk, especially in microgravity environments where even minor operational deviations or delays can trigger a chain of safety hazards. Against this backdrop, space station intravehicular robots with autonomous operating capabilities are gradually becoming core equipment for ensuring the efficient operation of the space station.

[0003] Early fixed robotic arms used rigid connection mechanisms to perform tasks such as transporting equipment within the cabin, but their operating range was limited by mechanical structure constraints and could not cover dynamic target areas. While mobile humanoid robots possessed anthropomorphic operational capabilities, their energy supply efficiency and structural complexity made them unsuitable for long-term on-orbit maintenance. In recent years, microgravity-based flying robots, due to their three-dimensional omnidirectional maneuverability and modular design advantages, have become an important development direction for autonomous in-cabin operation equipment. These robots do not rely on fixed tracks or ground contact, can freely hover within the cabin, and perform cross-cabin collaborative tasks. Their lightweight configuration and flexible deployment capabilities are significantly superior to traditional robotic arms and humanoid robots. Some systems have already achieved basic suspension control and simple path tracking functions, initially verifying the feasibility of on-orbit application.

[0004] However, existing flight robot navigation systems still suffer from fundamental flaws: on the one hand, positioning technology heavily relies on pre-set QR code markers or manual remote control, making it prone to positioning failures in unmarked areas or complex dynamic scenarios; on the other hand, traditional obstacle avoidance algorithms struggle to adapt to the highly dynamic environment inside the cabin, resulting in autonomous navigation capabilities far from practical application. This strong dependence on human intervention and fixed markers severely limits the application potential of flight robots in complex situations such as large-scale expansion on space stations and concurrent execution of multiple tasks, necessitating an urgent improvement in the autonomous navigation capabilities of flight robots. Summary of the Invention

[0005] The technical problem to be solved by this invention is:

[0006] To address the problem that the autonomous navigation of existing space station cabin flying robots relies heavily on preset markers or manual remote control, and that obstacle avoidance algorithms are insufficient.

[0007] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0008] This invention provides an autonomous navigation method for a flying robot in a microgravity environment, comprising the following steps:

[0009] S100 performs real-time modeling of 3D environment maps, including ORB-SLAM-based feature point extraction, unique descriptor matching, pose estimation, construction of local and global maps, minimization of local map stitching errors using graph optimization algorithms, and 3D localization and cabin map construction through loop closure detection. It calculates the precise pose of the flying robot using calibrated QR codes, optimizes positioning accuracy using particle filtering via the AMCL algorithm, and provides auxiliary recovery when ORB-SLAM fails. It generates a 3D surface model of the environment based on the TSDF algorithm and truncates the TSDF to an effective range to obtain an updated cabin map.

[0010] S200, motion planning and control under topological constraints: Design a topology-guided Hybrid A* path planning algorithm, which integrates the topology of grid maps and curve smoothing optimization techniques to achieve rapid path generation in dynamic environments; Through simulation and physical data modeling of kinematic and dynamic models, design a multi-level PID control method to follow the position, velocity, angle and angular velocity of the desired target, enabling the robot to respond agilely and efficiently to the motion commands issued by the astronauts;

[0011] S300 features a distance-based dynamic obstacle avoidance design, including prevention of instantaneous drift in positioning data, dynamic obstacle avoidance, emergency response, and emergency control, to improve the robustness and safety of flying robots.

[0012] Further, in step S100, including,

[0013] S110. Feature extraction is performed using the ORB algorithm, with the following formula:

[0014]

[0015] Where I represents the input image, x i Let d represent the coordinates of the i-th keypoint in the image. i The descriptor for the i-th keypoint is given, and N represents the number of keypoints.

[0016] Each feature point in the image has a uniquely matching descriptor. By matching these descriptors, the positions of identical feature points across frames are determined, thereby estimating their changing positions in the image and providing fundamental support for calculating the trajectory of the flying robot; the formula is:

[0017]

[0018] Where, d m and d n Let these represent the descriptors in image m and image n, respectively; Match(d m ,d n ) indicates that for d m and d n Hamming distance between them;

[0019] After the initial feature matching is completed, the robot's pose in the microgravity environment is estimated using the following formula:

[0020]

[0021] Where X represents the set of coordinates of all map points, X j Let x represent the coordinates of map point j, and let x represent the set of coordinates of all feature points in the image. j Let J represent the coordinates of the feature point in the image corresponding to map point j, T represent the transformation matrix, π represent the projection function, and J represent the total number of map points.

[0022] A local map is constructed based on the pose of the flying robot over a period of time. An incremental algorithm is then used to seamlessly stitch the local map together to form a global map. The formulas for constructing the local and global maps are as follows:

[0023]

[0024] Where K represents all keyframes involved in map construction, and M represents all map points in keyframe K;

[0025] A graph optimization algorithm is used to minimize the error of local map stitching. The objective function F(L, X) is expressed as:

[0026]

[0027] Where L represents the set of all camera poses, ρ represents the robust kernel function, and e op This represents the pose error between keyframe o and keyframe p. This represents the camera pose corresponding to keyframe o. Ωop represents the map point coordinates corresponding to keyframe p; Ωop represents the covariance matrix of the pose errors of keyframe o and keyframe p.

[0028] The optimization problem can be expressed as:

[0029]

[0030] When the flying robot revisits the previously visited area, ORB-SLAM identifies loop closures by comparing feature descriptors between the current view and existing keyframes in the database. Once a loop closure is identified, the state estimation problem for the in-cabin robot is transformed into a nonlinear least squares problem. The goal is to jointly optimize the camera pose and the position of the 3D points by minimizing the reprojection error of all observed 3D points, expressed as:

[0031]

[0032] in, The pose error between the feature points and map points of o Feature points of o keyframes, K p This is the camera intrinsic parameter matrix, where P is the 3D coordinate of the map point. It is a rotation matrix. It is a translation matrix.

[0033] Furthermore, in step S100, the method further includes:

[0034] S120. Pre-calibrate the absolute position of the AprilTag QR code in the world coordinate system, capture the AprilTag QR code image through the vision sensor on the flying robot, and calculate the position, deflection angle and physical size of the QR code in the image, thereby calculating the precise pose of the flying robot relative to the QR code in real time.

[0035] S130. The AprilTag QR code pose is used as an independent observation source and input into the state estimator along with the visual odometry, and then fused by weighted covariance matrix.

[0036] Initial particles are generated based on AMCL and assigned the same weights. The particle pose is estimated using a motion model.

[0037] Randomly generate particles with the same weights and calculate their new poses using a motion model:

[0038]

[0039] in, Let be the conditional probability distribution of the current coordinates at time t; Indicates the coordinates of the previous time step. The probability, Indicates the coordinates of the previous time step. The current input is Current coordinates are The probability of; Let be the coordinates of the q-th particle at time t;

[0040] The particle distribution is optimized by resampling, and the final estimated position x is calculated.t :

[0041]

[0042] Where Q is the total number of particles, It is the weight of the q-th particle;

[0043] Particle weights are updated using the latest observational data acquired by the sensors.

[0044]

[0045] in, For sensor observation data, Indicates coordinates The probability of;

[0046] S140, 3D mapping is based on the TSDF algorithm. By iteratively updating the TSDF volume mesh and fusing new measurement data, a 3D surface model of the environment is generated, including:

[0047]

[0048] Where S is the surface of the object, and dist(p,S) is the shortest distance from point p to surface S;

[0049] To avoid the effects of noise and measurement errors, TSDF is truncated to a limited range:

[0050]

[0051] in, The truncation threshold is a predefined constant;

[0052] For each voxel v, its TSDF value is updated using the following formula:

[0053]

[0054] Among them, T r It is the weight of the r-th measurement. It is the r-th voxel measured. r The corresponding TSDF value, where R represents the total number of measurements.

[0055] Furthermore, in step S300, to prevent instantaneous drift of positioning data, the gyroscope, ORB-SLAM algorithm, and QR code recognition technology are integrated. Under normal circumstances, the gyroscope provides high-frame-rate angular velocity data, ORB-SLAM generates accurate angle information based on feature point matching, and QR code recognition technology provides high-precision world coordinate position and attitude under visible conditions. When ORB-SLAM fails due to occlusion or environmental interference, the current velocity and attitude are locked first, and the device is slowly rotated to help ORB-SLAM recapture feature points and quickly resume normal operation.

[0056] Furthermore, in step S300, for dynamic obstacle avoidance and emergency response, a fusion of binocular camera and TOF sensor is used. The binocular camera collects environmental information in the forward and backward directions, and the TOF sensor covers point cloud data in the left and right directions, which are combined to generate an omnidirectional dynamic obstacle map. During normal operation, the flying robot can dynamically adjust its trajectory to avoid obstacles and plan a collision-free safe path. The TOF distance data around the fuselage is detected in real time. When the distance between the flying robot and the obstacle is between 15cm and 50cm, the relative speed between the flying robot and the obstacle is controlled to be less than the safety threshold. When the distance between the flying robot and the obstacle is within 15cm, the flying robot is forced to generate an obstacle avoidance trajectory in the opposite direction.

[0057] Furthermore, in step S300, for the emergency control function, the flying robot is equipped with a remote controller for manually activating the emergency stop or braking function; when in the emergency stop function, all fans of the flying robot are immediately shut off, and the movement is quickly terminated; when in the braking function, the speed and angular velocity of the flying robot are reduced to zero to avoid collision with surrounding equipment or instruments.

[0058] This invention discloses an autonomous navigation system for a flying robot in a microgravity environment, comprising a sensor module, a computing module, a control module, and a communication module.

[0059] The sensor module includes a binocular camera, a TOF sensor, a gyroscope, and an AprilTag QR code, which are used to generate environmental depth information, provide real-time angular velocity data, and realize pose initialization and position calibration, respectively.

[0060] The computing module uses NVIDIA Orin NX high-performance computing units to run core algorithms for environmental perception, path planning, and dynamic obstacle avoidance.

[0061] The control module is based on the FreeRTOS operating system for task scheduling and management, and supports sensor data processing, PID control and device attitude adjustment.

[0062] The communication module has data transmission and command reception functions, enabling remote command input and task feedback with the astronaut's intelligent interactive devices.

[0063] Compared with the prior art, the beneficial effects of the present invention are:

[0064] ① Through multi-sensor fusion and anti-interference localization algorithms, accurate environmental perception and localization are achieved, effectively addressing interference from occlusions or floating objects. A multi-hypothesis pose estimation algorithm is employed to improve the completeness and accuracy of environmental map modeling, thereby significantly enhancing astronaut navigation efficiency in complex microgravity environments.

[0065] ② The system utilizes a topology-guided Hybrid A* path planning algorithm to provide spatially optimal path planning, ensuring stable and efficient mission execution. Combined with multi-level PID trajectory tracking control, the robot can respond more agilely and efficiently to motion commands issued by the astronauts during flight.

[0066] ③ The system has a graded dynamic obstacle avoidance function. Through different obstacle avoidance strategies at different distances, it can avoid obstacles in advance, actively decelerate, and reverse obstacle avoidance. The graded obstacle avoidance function not only improves the safety and navigation flexibility in confined spaces, but also reduces the operational burden on astronauts, enabling the system to flexibly cope with diverse obstacles in microgravity environments. Attached Figure Description

[0067] Figure 1 This is a flowchart of an autonomous navigation method for a flying robot in a microgravity environment, as described in an embodiment of the present invention.

[0068] Figure 2 This is a navigation flowchart of the flying robot inside the space station cabin in an embodiment of the present invention;

[0069] Figure 3 This is a diagram illustrating the mapping effect of the flying robot inside the space station in an embodiment of the present invention.

[0070] Figure 4 This is a diagram illustrating the effect of AMCL (Ampere-Clearing-Landed Circular-Assisted) ORB-SLAM localization in an embodiment of the present invention.

[0071] Figure 5 This is a diagram illustrating the effect of motion planning and control of the space station cabin flying robot under topological constraints in an embodiment of the present invention. Detailed Implementation

[0072] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0073] Specific Implementation Plan 1: Combining Figures 1 to 5As shown, this invention provides an autonomous navigation method for a flying robot in a microgravity environment, comprising the following steps:

[0074] The S100 system enables real-time environmental perception and high-precision mapping. It utilizes an embedded Ubuntu operating system to run core algorithms responsible for mapping, planning, obstacle avoidance, and control functions. Based on ORB-SLAM, feature point extraction, matching, and loop closure detection are used to complete 3D localization and in-cabin map construction. Combined with multi-camera and TOF point cloud data fusion, it achieves all-around environmental perception. The AMCL algorithm optimizes positioning accuracy through particle filtering, providing auxiliary recovery when ORB-SLAM fails. Inter-frame motion estimation is performed using an extended Kalman filter (EKF) and visual-IMU fusion technology. The system optimizes pose output by analyzing the covariance matrix, reducing drift errors, and uses an incremental algorithm to stitch local maps together to generate a global map, ensuring global consistency and dynamic updates.

[0075] Specifically, including,

[0076] S110 and ORB-SLAM, as 3D localization and mapping algorithms based on ORB feature points, achieve fast and accurate tracking and mapping by optimizing map initialization, keyframe selection, and loop closure detection processes. Furthermore, they integrate angular velocity data from the robot's gyroscope and QR code positioning information to ensure high accuracy and high frame rate performance in the solution process. The robot's D435i dual cameras are used to collect front and rear field-of-view information, and independent ORB-SLAM nodes are run to generate real-time environmental perception maps in the front and rear directions, significantly enhancing the completeness of the mapping.

[0077] Feature extraction is performed using the ORB algorithm, and the formula is as follows:

[0078]

[0079] Where I represents the input image, x i Let d represent the coordinates of the i-th keypoint in the image. i The descriptor for the i-th keypoint is given, and N represents the number of keypoints.

[0080] Each feature point in an image has a uniquely matching descriptor. By matching these descriptors, the positions of identical feature points across frames are determined, thereby estimating their changing positions in the image and providing fundamental support for calculating the robot's motion trajectory. The formula is:

[0081]

[0082] Where, d m and d n Let these represent the descriptors in image m and image n, respectively; Match(d m ,dn ) indicates that for d m and d n Hamming distance between them;

[0083] After the initial feature matching is completed, the robot's pose in the microgravity environment is estimated using the following formula:

[0084]

[0085] Where X represents the set of coordinates of all map points, X j Let x represent the coordinates of map point j, and let x represent the set of coordinates of all feature points in the image. j Let J represent the coordinates of the feature point in the image corresponding to map point j, T represent the transformation matrix, π represent the projection function, and J represent the total number of map points.

[0086] A local map is constructed based on the pose of the flying robot over a period of time. An incremental algorithm is then used to seamlessly stitch the local map together to form a global map. The formulas for constructing the local and global maps are as follows:

[0087]

[0088] Where K represents all keyframes involved in map construction, and M represents all map points in keyframe K;

[0089] Using a graph optimization algorithm, the error of local map stitching is minimized. The objective function F(L, X) can be expressed as:

[0090]

[0091] Where L represents the set of all camera poses, ρ represents the robust kernel function, and e op This represents the pose error between keyframe o and keyframe p. Indicates keyframe o Ωop represents the map point coordinates corresponding to keyframe p; Ωop represents the covariance matrix of the pose errors of keyframe o and keyframe p.

[0092] Therefore, the optimization problem can be expressed as:

[0093]

[0094] Complex structures in microgravity environments can easily lead to misjudging the position of flying robots. Loop closure detection can effectively avoid the accumulation of such errors. When the flying robot revisits a previously visited area, ORB-SLAM3 identifies loop closures by comparing the feature descriptors between the current view and existing keyframes in the database. Once a loop closure is identified, the state estimation problem of the in-cabin flying robot is transformed into a nonlinear least squares problem, commonly known as BA (Bundle Adjustment) optimization. The optimization objective of BA is to jointly optimize the camera pose and the position of 3D points by minimizing the reprojection error of all observed 3D points. The specific optimization problem can be expressed as:

[0095]

[0096] in, The pose error between the feature points and map points of o K is the feature point of the o-th keyframe. p This is the camera intrinsic parameter matrix, where P is the 3D coordinate of the map point. It is a rotation matrix. It is a translation matrix;

[0097] During this optimization process, the algorithm adjusts the camera pose by minimizing the reprojection error. The location P of the three-dimensional points is used to improve the accuracy and consistency of the map and eliminate errors caused by position drift. This is especially important for robots that operate for long periods of time, because long-term drift can cause the robot to mistakenly believe that it is in a different location, thus affecting the accuracy of task execution.

[0098] S120. Pre-calibrate the absolute position of the AprilTag QR code in the world coordinate system, capture the AprilTag QR code image through the vision sensor on the flying robot, and calculate the position, deflection angle and physical size of the QR code in the image, thereby calculating the precise pose of the flying robot relative to the QR code in real time.

[0099] S130. The AprilTag QR code pose is used as an independent observation source and input into the state estimator along with the visual odometry, and then fused by weighted covariance matrix.

[0100] The state estimator is based on the AMCL algorithm. AMCL (Adaptive Monte Carlo Localization) is a particle filter-based localization method that generates initial particles and assigns them the same weights, and estimates the particle pose by combining the motion model. The visual odometry observation model is used to adjust the weight distribution of the particles, and the resampling step further improves the localization accuracy by optimizing the particle distribution. Finally, the optimal position is determined by calculating the mean and variance of the weights.

[0101] When used in conjunction with the ORB-SLAM algorithm, AMCL can effectively assist the system in recovering pose and enhance the stability and robustness of localization when ORB-SLAM information is lost. The main steps are as follows:

[0102] Randomly generate particles with the same weights and calculate their new poses using a motion model:

[0103]

[0104] in, Let be the conditional probability distribution of the current coordinates at time t; Indicates the coordinates of the previous time step. The probability, Indicates the coordinates of the previous time step. The current input is Current coordinates are The probability of; Let be the coordinates of the q-th particle at time t;

[0105] The particle distribution is optimized by resampling, and the final estimated position x is calculated. t :

[0106]

[0107] Where Q is the total number of particles, It is the weight of the q-th particle;

[0108] Particle weights are updated using the latest observational data acquired by the sensors.

[0109]

[0110] Among them, z t This represents the sensor's observation data. Indicates coordinates The probability of;

[0111] S140. 3D mapping is based on the TSDF (Truncate Signed Distance Function) algorithm. By iteratively updating the TSDF volumetric mesh and fusing new measurement data, a 3D surface model of the environment is generated for downstream tasks such as robot navigation, path planning, and obstacle avoidance. The main steps are:

[0112]

[0113] Where S is the surface of the object, and dist(p,S) is the shortest distance from point p to surface S;

[0114] To avoid the effects of noise and measurement errors, TSDF is truncated to a limited range:

[0115]

[0116] in, The truncation threshold is a predefined constant;

[0117] For each voxel v, its TSDF value is updated using the following formula:

[0118]

[0119] Among them, T r It is the weight of the r-th measurement. It is the r-th voxel measured. r The corresponding TSDF value, where R represents the total number of measurements;

[0120] S200, motion planning and control under topological constraints: In order to avoid the flight trajectory of the flying robot from affecting the normal work of astronauts, the control and trajectory planning algorithms of the flying robot need to be optimized in view of the collaborative mission requirements of astronauts and the characteristics of microgravity.

[0121] The Hybrid A* path planning algorithm based on topology guidance is adopted, and the cost function formula is designed as follows:

[0122]

[0123] The heuristic cost function h(n) is used to estimate the distance from node n to the target node goal:

[0124]

[0125] Where, d Euclidean Represents Euclidean distance;

[0126] The actual cost function g(n) is a weighted combination of time cost, path length cost, and energy consumption cost:

[0127]

[0128] Where α, β and γ are weighting coefficients corresponding to path length, time and energy consumption; It is a node and The distance between them From node Move to The time required From node Move to The required energy;

[0129] By integrating the topology of grid maps and curve smoothing optimization techniques, the algorithm enables rapid path generation in dynamic environments, ensuring that the flying robot finds a spatially optimal solution. The algorithm can plan a collision-free and spatially optimal navigation trajectory, and when environmental changes are detected (such as the addition of obstacles or adjustment of the target position), it can quickly complete path replanning, ensuring mission continuity and the safety of equipment operation.

[0130] By using simulation and physical data modeling platforms to study kinematics and dynamics, a multi-level PID control method is designed to follow the desired target position, velocity, angle, and angular velocity, enabling the robot to respond agilely and efficiently to motion commands issued by astronauts. This includes establishing a six-degree-of-freedom kinematic model of the flight robot and constructing a dynamic model based on the thrust characteristics of the propulsion fan, which is used to analyze the force and response characteristics of the onboard equipment under various operating modes, so as to dynamically adjust the control strategy. Combining kinematic and dynamic analysis, precise positioning of the target is achieved by adjusting the thruster output, while angular velocity adjustment is used to ensure the stability and accuracy of the flight attitude, thereby improving navigation accuracy and system response speed.

[0131] S300, a distance-based dynamic obstacle avoidance design, including prevention of instantaneous drift in positioning data, dynamic obstacle avoidance, emergency response, and emergency control; specifically including,

[0132] S310. To prevent instantaneous drift of positioning data, a gyroscope, ORB-SLAM algorithm, and QR code recognition technology are integrated to achieve accurate positioning and attitude perception. Under normal circumstances, the gyroscope provides high-frame-rate angular velocity data, the ORB-SLAM algorithm generates accurate angle information based on feature point matching, and the QR code recognition technology provides high-precision world coordinate position and attitude under visible conditions. The three methods complement each other, enabling the robot to obtain real-time and stable pose information. When the ORB-SLAM algorithm fails due to occlusion or environmental interference, the system will prioritize locking the current velocity and attitude, and at the same time, slowly rotate the device to help ORB-SLAM recapture feature points and quickly resume normal operation. This mechanism effectively prevents instantaneous drift of positioning data and improves the robustness of the system.

[0133] For dynamic obstacle avoidance and emergency response, the S320 employs a D435i binocular camera and TOF sensor fusion scheme to address the dense floating objects inside the cabin. Two D435i binocular cameras collect environmental information in the forward and backward directions, while the TOF sensors cover point cloud data in the left and right directions. This combination generates an omnidirectional dynamic obstacle map, providing a basis for path planning. During normal operation, the robot can dynamically adjust its trajectory to avoid obstacles and plan a collision-free safe path. Real-time detection of TOF distance data around the robot's body is maintained. When the robot is 15cm to 50cm away from an obstacle, the relative speed between the robot and the obstacle is controlled to be below a safe threshold. When the robot is within 15cm of an obstacle, a reverse obstacle avoidance trajectory is forcibly generated. The robot is equipped with 10 TOF sensors with a 65-degree field of view, enabling omnidirectional obstacle perception and rapid response, effectively ensuring equipment safety.

[0134] Emergency control function: The robot is equipped with a dedicated remote control that can be manually activated to start the emergency stop or braking function; the emergency stop function will immediately shut down all the robot's fans and quickly stop the movement; the braking function will reduce the speed and angular velocity of the equipment to zero to avoid collisions with surrounding equipment or instruments; this function can be used flexibly by the operator according to the actual situation to improve operational safety.

[0135] Specific Implementation Plan Two: Combining Figure 2 As shown, this invention provides an autonomous navigation system for a flying robot in a microgravity environment, comprising an integrated sensor module, a computing module, a control module, and a communication module.

[0136] The sensor module includes a binocular camera, a TOF sensor, a gyroscope, and an AprilTag QR code. The binocular camera is used to generate environmental depth information, the TOF sensor is used to measure distance, the gyroscope is used to provide real-time angular velocity data, as well as to realize pose initialization and position calibration, and the AprilTag QR code recognition is used to provide high-precision world coordinate position and attitude under visible conditions, thereby ensuring the high-precision positioning and stability of the navigation system.

[0137] The computing module uses NVIDIA Orin NX high-performance computing units to run core algorithms for environmental perception, path planning, and dynamic obstacle avoidance, enabling it to process complex task data in real time and efficiently.

[0138] The control module uses an STM32F407 series microcontroller and is based on the FreeRTOS operating system for task scheduling and management. It supports sensor data processing, PID control and device posture adjustment, thereby ensuring the robot's robustness in different environments.

[0139] The communication module has data transmission and command reception functions, enabling remote command input and task feedback with the astronaut's intelligent interactive devices, thus improving the system's response speed and task execution efficiency;

[0140] By integrating the above modules, this invention provides a robot autonomous navigation system with environmental perception, dynamic obstacle avoidance and high-precision positioning capabilities. It can achieve stable operation and efficient task execution in complex microgravity environments, and has significant technical advantages and broad application prospects.

[0141] The other combinations and connections in this implementation scheme are the same as in Specific Implementation Scheme 1.

[0142] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. An autonomous navigation method for a flying robot in a microgravity environment, characterized in that, Includes the following steps: S100 performs real-time modeling of 3D environment maps, including ORB-SLAM-based feature point extraction, unique descriptor matching, pose estimation, construction of local and global maps, minimization of local map stitching errors using graph optimization algorithms, and 3D localization and cabin map construction through loop closure detection. It calculates the precise pose of the flying robot using calibrated QR codes, optimizes positioning accuracy using particle filtering via the AMCL algorithm, and provides auxiliary recovery when ORB-SLAM fails. It generates a 3D surface model of the environment based on the TSDF algorithm and truncates the TSDF to an effective range to obtain an updated cabin map. include, S110. Feature extraction is performed using the ORB algorithm, with the following formula: Where I represents the input image, x i Let d represent the coordinates of the i-th keypoint in the image. i The descriptor for the i-th keypoint is given, and N represents the number of keypoints. Each feature point in the image has a uniquely matching descriptor. By matching these descriptors, the positions of identical feature points across frames are determined, thereby estimating their changing positions in the image and providing fundamental support for calculating the trajectory of the flying robot; the formula is: Where, d m and d n Let these represent the descriptors in image m and image n, respectively; Match(d m ,d n ) indicates that for d m and d n Hamming distance between them; After the initial feature matching is completed, the robot's pose in the microgravity environment is estimated using the following formula: Where X represents the set of coordinates of all map points, X j Let x represent the coordinates of map point j, and let x represent the set of coordinates of all feature points in the image. j Let J represent the coordinates of the feature point in the image corresponding to map point j, T represent the transformation matrix, π represent the projection function, and J represent the total number of map points. A local map is constructed based on the pose of the flying robot over a period of time. An incremental algorithm is then used to seamlessly stitch the local map together to form a global map. The formulas for constructing the local and global maps are as follows: Where K represents all keyframes involved in map construction, and M represents all map points in keyframe K; A graph optimization algorithm is used to minimize the error of local map stitching. The objective function F(L, X) is expressed as: Where L represents the set of all camera poses, ρ represents the robust kernel function, and e op This represents the pose error between keyframe o and keyframe p. This represents the camera pose corresponding to keyframe o. Ωop represents the map point coordinates corresponding to keyframe p; Ωop represents the covariance matrix of the pose errors of keyframe o and keyframe p. The optimization problem can be expressed as: When the flying robot revisits the previously visited area, ORB-SLAM identifies loop closures by comparing feature descriptors between the current view and existing keyframes in the database. Once a loop closure is identified, the state estimation problem for the in-cabin robot is transformed into a nonlinear least squares problem. The goal is to jointly optimize the camera pose and the position of the 3D points by minimizing the reprojection error of all observed 3D points, expressed as: in, This represents the pose error between the feature points and map points in keyframe o. K is the feature point of the o-th keyframe. p This is the camera intrinsic parameter matrix, where P is the 3D coordinate of the map point. It is a rotation matrix. It is a translation matrix; S200, motion planning and control under topological constraints: Design a topology-guided Hybrid A* path planning algorithm, which integrates the topology of grid maps and curve smoothing optimization techniques to achieve rapid path generation in dynamic environments; Through simulation and physical data modeling of kinematic and dynamic models, design a multi-level PID control method to follow the position, velocity, angle and angular velocity of the desired target, enabling the robot to respond agilely and efficiently to the motion commands issued by the astronauts; S300, a distance-based dynamic obstacle avoidance design, including prevention of instantaneous drift of positioning data, dynamic obstacle avoidance and emergency response and emergency control, is used to improve the robustness and safety of flying robots. in, To prevent instantaneous drift of positioning data, a gyroscope, ORB-SLAM algorithm, and QR code recognition technology are integrated. Under normal circumstances, the gyroscope provides high-frame-rate angular velocity data, ORB-SLAM generates accurate angle information based on feature point matching, and QR code recognition technology provides high-precision world coordinate position and attitude under visible conditions. When ORB-SLAM fails due to occlusion or environmental interference, the current velocity and attitude are locked first, and the device is slowly rotated to help ORB-SLAM recapture feature points and quickly resume normal operation. For dynamic obstacle avoidance and emergency response, a fusion of binocular cameras and TOF sensors is used. The binocular cameras collect environmental information in the forward and backward directions, while the TOF sensors cover point cloud data in the left and right directions. The combination generates an omnidirectional dynamic obstacle map. During normal operation, the flying robot can dynamically adjust its trajectory to avoid obstacles and plan a collision-free safe path. The TOF distance data around the fuselage is detected in real time. When the distance between the flying robot and the obstacle is between 15cm and 50cm, the relative speed between the flying robot and the obstacle is controlled to be less than a safe threshold. When the distance between the flying robot and the obstacle is less than 15cm, the flying robot is forced to generate an obstacle avoidance trajectory in the opposite direction.

2. The autonomous navigation method for a flying robot in a microgravity environment according to claim 1, characterized in that: In step S100, it also includes, S120. Pre-calibrate the absolute position of the AprilTag QR code in the world coordinate system, capture the AprilTag QR code image through the vision sensor on the flying robot, and calculate the position, deflection angle and physical size of the QR code in the image, thereby calculating the precise pose of the flying robot relative to the QR code in real time. S130. The AprilTag QR code pose is used as an independent observation source and input into the state estimator along with the visual odometry, and then fused by weighted covariance matrix. Initial particles are generated based on AMCL and assigned the same weights, and the particle pose is estimated by combining the motion model. include, Randomly generate particles with the same weights and calculate their new poses using a motion model: in, Let be the conditional probability distribution of the current coordinates at time t; Indicates the coordinates of the previous time step. The probability, Indicates the coordinates of the previous time step. The current input is Current coordinates are The probability of; Let be the coordinates of the q-th particle at time t; The particle distribution is optimized by resampling, and the final estimated position x is calculated. t : Where Q is the total number of particles, It is the weight of the q-th particle; Particle weights are updated using the latest observational data acquired by the sensors. in, For sensor observation data, Indicates coordinates The following observations were obtained The probability of; S140, 3D mapping is based on the TSDF algorithm. By iteratively updating the TSDF volume mesh and fusing new measurement data, a 3D surface model of the environment is generated, including: Where S is the surface of the object, and dist(p,S) is the shortest distance from point p to surface S; To avoid the effects of noise and measurement errors, TSDF is truncated to a limited range: in, The truncation threshold is a predefined constant; For each voxel v, its TSDF value is updated using the following formula: Among them, T r It is the weight of the r-th measurement. It is the r-th voxel measured. r The corresponding TSDF value, where R represents the total number of measurements.

3. The autonomous navigation method for a flying robot in a microgravity environment according to claim 2, characterized in that: In step S300, for the emergency control function, the flying robot is equipped with a remote controller for manually activating the emergency stop or braking function; when the emergency stop function is activated, all the fans of the flying robot are immediately shut off, and the movement is quickly terminated; when the braking function is activated, the speed and angular velocity of the flying robot are reduced to zero to avoid collision with surrounding equipment or instruments.

Images