Controller and system for obstacle avoidance in a constrained environment

Abstract

The application relates to the technical field of mobile obstacle avoidance, and discloses a controller and system for restricted environment obstacle avoidance with accurate environment information acquisition and rapid reaction, which comprises at least one radar sensing module (110) for detecting the dynamic distance between a controlled object and a front obstacle; at least one attitude sensing module (120) for acquiring the attitude parameters of the obstacle in a three-dimensional space in real time; at least one visual sensing module (130) for detecting the shape parameters of the obstacle in the three-dimensional space; a main control module (140) for outputting an adjustment signal according to the dynamic distance, the attitude parameters and the shape parameters; and an execution module (150) for correcting the motion path of the controlled object according to the adjustment signal to avoid the obstacle.

Description

Technical Field

[0001] This invention relates to the field of mobile obstacle avoidance technology, and more specifically, to a controller and system for obstacle avoidance in confined environments. Background Technology

[0002] With the widespread application of embedded mobile devices such as drones, autonomous mobile robots, and IoT smart terminals, navigation and obstacle avoidance technology has become a core component in ensuring their safe operation and autonomous task capability. These devices typically have limited computing power, scarce storage resources, and constrained energy supply, placing stringent requirements on the lightweight, low-power, and high real-time performance of their sensing systems.

[0003] Currently, navigation and obstacle avoidance solutions for such devices generally employ airborne or vehicle-mounted sensors (such as ultrasonic sensors, infrared sensors, monocular or binocular vision cameras, millimeter-wave radar, and low-beam lidar) to collect and report environmental information in real time along the travel path. The data collected by the sensors is then processed by an embedded processing unit for obstacle detection and path planning, thereby achieving obstacle avoidance control. However, existing technologies have significant limitations in terms of perception capabilities.

[0004] Due to limitations in device size, power consumption, and cost, the effective sensing range of the aforementioned sensors is typically short. In applications involving high-speed movement or requiring advance path planning, the limited sensing range prevents the device from acquiring environmental information over long distances, making it difficult to identify obstacles in the far field or predict potential risk areas in a timely manner. This, in turn, limits the foresight of obstacle avoidance strategies and the optimality of path planning. Existing perception systems are mostly based on static or quasi-static environmental assumptions, and lack timely and effective detection and response capabilities for sudden dynamic events (such as falling objects, temporary construction obstacles, etc.) in the travel route. Due to the dual constraints of sensor sensing range and system processing delay, the equipment often only starts obstacle avoidance response after the event enters the line of sight range, resulting in insufficient reaction time and a decrease in obstacle avoidance success rate. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address the shortcomings of the existing technology, which is that the limited sensing distance makes it impossible for the device to obtain environmental information in a long distance range, making it difficult to identify obstacles in the far field or predict potential risk areas in a timely manner, and due to the dual constraints of sensor sensing range and system processing delay, the device often only starts obstacle avoidance response after the event enters the line of sight. The present invention provides a controller and system for obstacle avoidance in limited environments that can obtain environmental information accurately and react quickly.

[0006] The technical solution adopted by this invention to solve its technical problem is: to construct a controller for obstacle avoidance in confined environments, comprising: At least one radar sensing module is used to detect the dynamic distance between the controlled object and the obstacle in front; At least one attitude sensing module is used to acquire the attitude parameters of the obstacle in three-dimensional space in real time; At least one visual sensing module is used to detect the shape parameters of the obstacle in three-dimensional space; A main control module, whose input terminals are respectively connected to the output terminals of the radar sensing module, the attitude sensing module and the vision sensing module, outputs adjustment signals according to the dynamic distance, the attitude parameters and the shape parameters; An execution module, whose input is connected to the output of the main control module, corrects the movement path of the controlled object according to the adjustment signal to avoid obstacles.

[0007] In some implementations, the main control module acquires the current speed of the controlled object with a planning period of 100ms to 200ms, samples multiple feasible trajectories in the current speed space using a dynamic window method, and evaluates the cost based on the distance between each feasible trajectory and the obstacle, the motion constraints of the controlled object, and the predicted dynamic obstacle motion trend, and selects the optimal trajectory to generate a replanning path.

[0008] In some implementations, the main control module includes at least: A data processing unit is used to perform spatiotemporal alignment and feature fusion of the dynamic distance, the attitude parameters and the morphological parameters to generate obstacle fusion feature data; A decision-making unit, whose input is connected to the data processing unit, generates the adjustment signal based on the obstacle fusion feature data.

[0009] In some implementations, the main control module further includes a storage unit. The storage unit contains a pre-stored obstacle avoidance strategy mapping table corresponding to different obstacle types; The decision-making unit identifies the obstacle type based on the obstacle fusion feature data, matches the corresponding obstacle avoidance strategy from the obstacle avoidance strategy mapping table, and generates the adjustment signal.

[0010] In some embodiments, the execution module includes a motion control unit and a drive unit. The motion control unit is used to analyze the adjustment signal and generate speed control commands and steering control commands; The input terminal of the drive unit is connected to the output terminal of the motion control unit, and it drives the motion mechanism of the controlled object according to the speed control command and the steering control command.

[0011] In some embodiments, the visual sensing module includes at least one depth camera disposed on the vehicle body of the controlled object to acquire depth images of the near-peripheral area of ​​the vehicle body in order to identify nearby obstacles.

[0012] In some implementations, the attitude sensing module includes an inertial measurement unit for acquiring the attitude angle of the obstacle in real time, and the main control module determines the interference risk between the controlled object and the surrounding environment based on the attitude angle.

[0013] Secondly, a control system for obstacle avoidance in a confined environment detects the dynamic distance between the controlled object and the obstacle in front through at least one radar sensing module, acquires the attitude parameters of the obstacle in three-dimensional space in real time through at least one attitude sensing module, and detects the morphological parameters of the obstacle in three-dimensional space through at least one visual sensing module. The dynamic distance, the attitude parameters, and the shape parameters are input into a main control module, and the dynamic distance, attitude parameters, and shape parameters are fused to generate comprehensive obstacle feature information. The main control module generates adjustment signals based on the comprehensive feature information of the obstacle and the preset obstacle avoidance strategy; The adjustment signal is output to an execution module, which then corrects the movement path of the controlled object based on the adjustment signal to avoid the obstacle.

[0014] The obstacle avoidance controller for confined environments described in this invention includes at least one radar sensing module for detecting the dynamic distance between the controlled object and obstacles in front; at least one attitude sensing module for acquiring the attitude parameters of the obstacle in three-dimensional space in real time; and at least one vision sensing module for detecting the shape parameters of the obstacle in three-dimensional space. The main control module outputs adjustment signals based on the dynamic distance, attitude parameters, and shape parameters. The execution module corrects the movement path of the controlled object based on the adjustment signals to avoid obstacles. Compared with existing technologies, the limited sensing capability of a single sensor is overcome through the coordinated operation of radar, attitude, and vision sensing modules. Radar is responsible for distance detection, vision acquires shape details, and attitude sensing provides spatial orientation information of obstacles. The fusion of these three sensors enables the controller to simultaneously obtain distance information of far-field obstacles and fine shape information of near-field obstacles, solving the technical defect that "the effective sensing distance is usually short." The introduction of attitude and morphology parameters enables the controller to identify the movement trend and spatial attitude of obstacles, significantly improving its ability to detect sudden dynamic events (such as falling objects and temporary construction obstacles), and providing richer decision-making basis for early obstacle avoidance. Attached Figure Description

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a circuit schematic diagram of an embodiment of the obstacle avoidance controller provided by the present invention. Detailed Implementation

[0016] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0017] like Figure 1 As shown, in the first embodiment of the obstacle avoidance controller for confined environments of the present invention, the obstacle avoidance controller 100 for confined environments includes at least one radar sensing module 110, at least one attitude sensing module 120, at least one vision sensing module 130, a main control module 140, and an execution module 150. Among them, the radar sensing module 110 can be selected as a low-cost low-beam lidar or millimeter-wave radar, which is installed in front of or around the controlled object (such as: drone / robot). Its main function is to detect the dynamic distance between the controlled object and the obstacles in front in real time. This module has the advantages of long detection distance (e.g., up to 10-20 meters) and insensitivity to light, which can effectively make up for the shortcomings of visual sensors in long-distance or light-changing environments.

[0018] For example, when moving at high speed in a corridor, the radar sensing module 110 can detect a stationary box 20 meters away in advance and send the distance data (such as "19.8 meters") to the main control module 140 in real time. The attitude sensing module 120 preferably includes an inertial measurement unit (IMU), which can be mounted on the obstacle (in a collaborative scenario) or more commonly, infer the attitude of the obstacle by sensing the attitude changes of the controlled object itself and combining them with external perception. This module is used to acquire the attitude parameters of the obstacle in three-dimensional space in real time, such as pitch angle, roll angle and yaw angle. These parameters are crucial for judging the stability and potential movement trend of the obstacle. For example, when the vision module detects a cylindrical object in front, but the attitude sensing module 120 detects that the object's attitude angle is changing rapidly (e.g., the yaw angle is rotating at a speed of 30 degrees / second), the main control module 140 can determine that the object may be a rolling dynamic obstacle (such as a falling cylinder) rather than a static stone pillar, thereby triggering a higher priority obstacle avoidance response and controlling the drone / robot to decelerate or avoid it laterally. The vision sensing module 130 includes at least one depth camera (RGB-D camera or 3D camera), typically positioned at the front or around the controlled object's body, to acquire depth images of the near-peripheral area of ​​the drone / robot to identify fine morphological parameters of nearby obstacles, such as outline, volume, and category (e.g., distinguishing between "wall," "pedestrian," or "chair"). The depth camera provides centimeter-level near-field perception accuracy.

[0019] For example, when a drone / robot approaches a half-open door, the radar module may not detect it well due to the door panel being too thin or the angle being too wide. However, the visual sensing module 130 can clearly capture the edge of the door panel, the opening and closing angle, and the depth of the space behind the door, forming accurate 3D point cloud data to provide decision-making basis for the main control module 140. The main control module 140 is the core of the controller. It is usually composed of an embedded processor (such as an ARM Cortex-M series or a low-power FPGA) and its peripheral circuits. Its input terminals are connected to the output terminals of the three sensing modules mentioned above, and are used to receive dynamic distance, attitude parameters and morphological parameters, and generate adjustment signals based on this information. The execution module 150 actually corrects the motion path of the controlled object based on the adjustment signal; Specifically, at least one radar sensing module 110 is used to detect the dynamic distance (e.g., far / near) between the controlled object and the obstacle in front during the movement of the controlled object (e.g., flying or driving close to the ground), and then feeds back the acquired dynamic distance to the main control module 140. Furthermore, at least one attitude sensing module 120 is used to acquire the attitude parameters of the obstacle in three-dimensional space in real time (such as pitch angle, roll angle or yaw angle), and then feed the acquired attitude parameters back to the main control module 140. Furthermore, at least one visual sensing module 130 is used to detect the shape parameters (such as contour, volume, and category) of the obstacle in three-dimensional space, and then feeds back the acquired shape parameters to the main control module 140. The main control module 140 has its input terminal connected to the output terminal of the radar sensing module 110, and is used to receive feedback on the dynamic distance. The input terminal of the main control module 140 is connected to the output terminal of the attitude sensing module 120 to receive feedback attitude parameters, and The input terminal of the main control module 140 is connected to the output terminal of the vision sensing module 130, which receives feedback morphological parameters and outputs adjustment signals based on dynamic distance, attitude parameters and morphological parameters. An execution module 150 has its input terminal connected to the output terminal of the main control module 140. It receives adjustment signals and corrects the movement path of the controlled object to avoid obstacles based on the input adjustment signals.

[0020] This technical solution overcomes the limited sensing capabilities of a single sensor by utilizing the coordinated operation of three sensing modules: radar, attitude sensing, and vision. Radar is responsible for distance detection, vision acquires morphological details, and attitude sensing provides spatial orientation information of obstacles. The fusion of these three sensors enables the controller to simultaneously obtain distance information of far-field obstacles and fine morphological information of near-field obstacles, thus solving the technical defect that "the effective sensing distance is usually short." The introduction of attitude and morphology parameters enables the controller to identify the movement trend and spatial attitude of obstacles, significantly improving its ability to detect sudden dynamic events (such as falling objects and temporary construction obstacles), and providing richer decision-making basis for early obstacle avoidance.

[0021] In some implementations, the main control module 140 obtains the current speed of the controlled object with a planning period of 100ms to 200ms, uses the dynamic window method to sample multiple feasible trajectories in the current speed space, and performs cost evaluation based on the distance between each feasible trajectory and the obstacle, the motion constraints of the controlled object and the predicted dynamic obstacle motion trend, and selects the optimal trajectory to generate a replanning path.

[0022] For example, a quadcopter drone is flying at a speed of 2.5 m / s along a preset route in a confined environment (such as an indoor warehouse or a gap in the woods), and the planning cycle of the main control module 140 is set to 150 ms. At this time, the radar sensing module 110 detects a sudden dynamic obstacle about 5m ahead—another drone flying in a cross path. At the same time, the attitude sensing module 120 detects the attitude parameters (pitch angle, roll angle, yaw angle) of the obstacle drone, indicating that it is performing a right turn maneuver. The visual sensing module 130 identifies it as a "dynamic aircraft". Within the current planning cycle, the main control module 140 acquires the current speed of the UAV (2.5 m / s, horizontal straight-line). Using a dynamic window method, it samples and generates multiple feasible trajectories within the three-dimensional velocity space (horizontal linear velocity 0–4.0 m / s, vertical linear velocity -1.0–1.5 m / s, yaw rate -1.0–1.0 rad / s), including: Track A: Continue straight and decelerate to 1.5 m / s, maintaining the current altitude; Trajectory B: Horizontally to the right, maintaining a linear velocity of 2.0 m / s and a yaw rate of 0.6 rad / s; Track C: Emergency climb, vertical linear velocity increases to 1.2 m / s, horizontal velocity remains unchanged; Trajectory D: Slightly adjust to the left and descend 0.5m simultaneously, with the horizontal speed decreasing to 1.8 m / s.

[0023] The main control module 140 performs cost evaluation for each trajectory: Based on the dynamic distance change rate measured by the radar module, trajectory A is predicted to have a minimum distance of only 0.3m from the obstacle drone after 1.2 seconds, posing a serious risk of collision. Trajectory B will further reduce the distance due to the obstacle drone turning right, resulting in insufficient safety margin. The vertical ascent of trajectory C will cause the two to be offset at the altitude level, with a predicted minimum distance of 1.2m. Trajectory D moves to the left and descends, with a predicted minimum distance of 0.8m, which is within the safety boundary.

[0024] Motion constraint cost: The vertical acceleration of trajectory C (1.2 m / s²) and the combined motion (horizontal + vertical) of trajectory D are both within the flight control and motor response capabilities of the UAV (maximum vertical acceleration ≤ 2.0 m / s², maximum yaw rate ≤ 1.5 rad / s); neither trajectory A nor trajectory B exceeds the motion constraints.

[0025] The dynamic obstacle movement trend cost: The attitude sensing module 120 detected that the yaw angle change rate of the drone was 40° / s (turning right). Combined with its current speed and direction, it can predict that it will move to the right and forward within the next second. Trajectory B chooses to detour to the right, which is in line with the predicted movement direction of the obstacle, making it difficult to effectively increase the distance; Trajectory C chooses to climb upwards, which is orthogonal to the plane of movement of the obstacle, and has the highest efficiency in avoiding collisions; Trajectory D descends to the left, which can partially avoid collisions, but is limited by the descent space (there is a shelf below), and its safety is slightly lower than that of Trajectory C. Considering the above three costs, the main control module 140 ultimately selects trajectory C (emergency climb) as the optimal replanning path and generates the corresponding adjustment signal. Upon receiving this signal, the execution module 150 increases the lift of the four rotors within 150 ms, causing the UAV to climb 1.2m with a vertical acceleration of 1.2 m / s² while maintaining its original horizontal heading. After safely passing the intersecting obstacle UAVs, it returns to its original altitude and continues flying.

[0026] In some embodiments, the main control module 140 includes at least a microcontroller U101, wherein the microcontroller U101 is configured with A data processing unit (not shown) is used to perform spatiotemporal alignment and feature fusion of dynamic distance, attitude parameters and morphological parameters to generate obstacle fusion feature data; A decision-making unit (not shown) has its input connected to the data processing unit, and generates adjustment signals based on obstacle fusion feature data.

[0027] In some implementations, the main control module 140 also includes a storage unit (not shown). The storage unit contains a pre-stored obstacle avoidance strategy mapping table corresponding to different obstacle types; The decision-making unit identifies obstacle types based on obstacle fusion feature data, matches the corresponding obstacle avoidance strategy from the obstacle avoidance strategy mapping table, and generates adjustment signals.

[0028] For example, a drone is performing an inspection mission in an orchard environment, flying at an altitude of 3 meters and a speed of 3 m / s. At this time, a complex obstacle scene appears about 8 meters ahead—a branch of a fruit tree is being blown by the wind, and a translucent plastic film is hanging from the branch; The working process of the data processing unit: The data processing unit simultaneously receives raw data from three types of sensing modules: The radar sensing module 110 detected an effective reflective surface 8m ahead, with a dynamic range of 7.8m. However, due to the semi-transparent nature of the plastic film, some radar waves penetrated, resulting in unstable echo signals and a range fluctuation of ±0.2m. The depth camera of the visual sensing module 130 captured a fine point cloud of the tree branch (morphological parameters: the branch diameter is about 3cm, the horizontal extension is about 1.5m, the plastic film size is about 0.6m×0.5m, and it sways with the wind). The attitude sensing module 120 calculates the swing period of the branch end to be about 2 seconds and the amplitude to be about 0.3m by visual-inertial fusion of multiple frames of images (attitude parameter: the motion trend relative to the UAV).

[0029] The data processing unit performs the following fusion operations: Time alignment unifies the radar ranging data (sampling frequency 20 Hz), visual depth image (30 Hz), and attitude estimation results (50 Hz) to the same timestamp. Interpolation algorithms are used to compensate for the time offset of each sensor, ensuring that the data in each fusion cycle comes from the same physical time (e.g., t=0 ms, t=50 ms, t=100 ms, etc.). Spatial alignment transforms radar polar coordinate data and camera pixel coordinate data into a world coordinate system with the UAV's center of mass as the origin. Through pre-calibrated extrinsic parameter matrices (rotation matrix R and translation vector T), accurate spatial correspondence is achieved for the same obstacle detected by different sensors.

[0030] For feature fusion, the data processing unit employs a weighted fusion strategy for the complex scenario of tree branches and plastic film: For the rigid structure of tree branches, the morphological parameters of the visual module have a high confidence level (weight set to 0.8), while radar distance data is used as an auxiliary (weight 0.2). For semi-transparent plastic film, the depth information confidence of the vision module is low (weight 0.3), but combined with the multi-frame pose change trend, it can be inferred that it belongs to the "low density, penetrable" obstacle type. The final obstacle fusion feature data includes: Obstacle A (tree branch), type "rigid dynamic obstacle", bounding box [2.1 m, 1.5 m, 0.3 m], motion tendency "periodic oscillation", minimum safe distance 0.5 m; Obstacle B (plastic film), type "semi-transparent flexible obstacle", bounding box [0.6 m, 0.5 m, 0.05 m], confidence level 0.65, suggested strategy "can be crossed by slowing down, no need for large detours".

[0031] The working process of the decision-making unit: After receiving the aforementioned fused feature data, the decision-making unit makes a comprehensive judgment based on the current status of the drone (speed 3 m / s, remaining battery 72%, task priority "inspection"). For the tree branch (obstacle A), the decision unit determines it as an obstacle that "needs to be avoided" and generates an adjustment signal with the action of "shifting 0.8m to the left while keeping the height unchanged" to avoid the swing range of the tree branch (the predicted swing range is the current position ±0.3m, plus a safety margin of 0.5m). For the plastic film (obstacle B), the decision unit determines it as a "low-risk passable" obstacle and generates the action in the adjustment signal as "maintain heading, reduce speed to 1.5 m / s, and slightly reduce blade speed to reduce wind disturbance" to avoid consuming extra energy due to large detours; The decision-making unit merges the two sub-decisions into a single adjustment signal and outputs it to the execution module 150. The execution module 150 then simultaneously adjusts the rotational speed of the six rotors. Within 150 ms, the UAV completes a 0.8m leftward shift and deceleration to 1.5 m / s, safely passing over the side of the plastic film while avoiding swaying tree branches. It then resumes its original speed to continue the inspection mission.

[0032] Through the above process, the data processing unit achieves effective fusion of multi-source heterogeneous sensors, transforming the raw data into structured, semantically meaningful obstacle feature information. The decision-making unit then makes differentiated and task-adaptive obstacle avoidance decisions based on the fused high-quality information, significantly improving the UAV's ability to handle diverse obstacles in complex natural environments.

[0033] In some embodiments, the execution module 150 includes a motion control unit U102 and drive units M1-M4, wherein the drive units M1-M4 are motors or electric motors. Among them, the motion control unit U102 is used to analyze the adjustment signal and generate speed control command and steering control command; The input terminals of drive units M1-M4 are connected to the output terminal of motion control unit U102, which can drive the motion mechanism of the controlled object according to speed control commands and steering control commands.

[0034] For example, a quadcopter drone is performing a reconnaissance mission in a narrow gap between buildings. Its current flight speed is 2 m / s, and its heading angle is 0° (due north). The main control module 140 generates an adjustment signal based on an obstacle detected ahead (a protruding corner of a wall). This adjustment signal contains the following semantic information: "Yaw 30° to the left, reduce horizontal speed to 1.2 m / s, and climb 0.5 m, completing the maneuver within 1 second." The working process of the motion control unit: After receiving the above adjustment signal, the motion control unit performs the following parsing and conversion operations: Signal parsing: breaking down abstract semantic instructions into specific motion control objectives. Yaw angle target value: 30° (currently 0°, needs to be changed by +30°); Target horizontal speed: 1.2 m / s (currently 2.0 m / s, needs to be reduced by 0.8 m / s); Altitude target: Current altitude + 0.5m; Machining time budget: 1.0 second.

[0035] Motion decomposition: Decomposes the complex motion into control commands in three independent channels: Horizontal channel: Deceleration command, deceleration rate is 0.8 m / s²; Yaw channel: Left turn instruction, yaw rate of 30° / s; Vertical path: Climb command, vertical speed is 0.5 m / s; Command generation: Convert the above control objectives into speed control and steering control commands executable at the motor level. Speed ​​control command: Desired resultant velocity vector (northward velocity 1.0 m / s, eastward velocity 0.6 m / s, vertical velocity 0.5 m / s), the corresponding horizontal velocity of this vector is 1.2 m / s (√(1.0²+0.6²)), and the direction is approximately 31° northward; Steering control command: Yaw angle velocity +30° / s (positive value for counterclockwise / left turn), yaw acceleration limit 60° / s²; Additional constraints: Maximum roll angle limited to 15°, maximum pitch angle limited to 12°, to ensure smooth maneuvering.

[0036] The motion control unit packages the above instructions into high-frequency control frames (e.g., one frame every 10ms) and sends them to the drive unit via a communication interface (such as UART or CAN bus).

[0037] Command decomposition: The received speed control and steering control commands are decomposed into individual rotational speed commands for each of the four rotors. A classic UAV control allocation matrix is ​​used. In the hovering state, the base speed of the four motors is approximately 5000 RPM (revolutions per minute). Execute the deceleration command: Increase the speed difference between the two front motors to make the drone pitch backward, thereby generating a horizontal deceleration force; Execute yaw command: Increase the speed difference between the clockwise rotating motor group (M1, M3) and the counterclockwise rotating motor group (M2, M4) to generate yaw torque; Execute the climbing command: synchronously increase the speed of the four motors to increase the total lift.

[0038] The final calculated target speeds of the four motors are as follows: M1 (front right, counter-clockwise): 6200 RPM M2 (front left, clockwise): 5800 RPM M3 (Rear Right, Counter-clockwise): 5400 RPM M4 (rear left, clockwise): 5600 RPM Closed-loop control: The electronic speed controller inside the drive unit employs closed-loop control (such as a PID algorithm), monitoring the actual speed of each motor in real time (detected via Hall sensors or back EMF) and comparing it with the target speed. It dynamically adjusts the PWM (Pulse Width Modulation) duty cycle to ensure the motor speed accurately follows the command. For example, if the actual speed of M1 is only 6100 RPM, the electronic speed controller will increase the PWM duty cycle from 65% to 67% to compensate for load fluctuations.

[0039] Motion execution: Driven by the drive unit, the four motors move simultaneously according to the calculated speed difference. t=0 ms: The drone maintains its original flight state; t=50 ms: The UAV begins to generate a yaw moment to the left and a pitch moment to the rear, while the total lift increases; At t=300 ms: the yaw angle changes by approximately 9°, the horizontal speed decreases to 1.7 m / s, and the altitude increases by 0.15 m; At t=600 ms: the yaw angle changes by approximately 18°, the horizontal speed decreases to 1.4 m / s, and the altitude increases by 0.3 m; t=1000 ms: The drone accurately reached the target state with a yaw angle of 30°, a horizontal speed of 1.2 m / s, and an altitude of +0.5m, successfully avoiding the corner of the wall, and then resumed cruise flight.

[0040] Example of overall synergistic effect: In a complete scenario, the drone flies along the exterior of a building at a speed of 2.5 m / s. The main control module 140 suddenly detects an open window in front of it (a dynamic obstacle, with the window sash opening outwards). The main control module 140 generates an adjustment signal: "Yaw 45° to the right, decelerate to 1.0 m / s, descend 0.3 m." The motion control unit interprets this command as a yaw rate of 45° / s, a horizontal deceleration of 1.5 m / s², and a vertical velocity of -0.3 m / s. The drive unit further calculates the differentiated speeds of the four motors: the front motor decelerates to generate pitch braking, the right motor accelerates to generate yaw torque, and all motors slightly decelerate to achieve descent. The entire process is completed within 0.8 seconds, and the drone safely passes below the window, avoiding a collision with the open window sash.

[0041] Through the close cooperation between the motion control unit and the drive unit, the execution module 150 can accurately and in real time convert the high-level adjustment signal output by the main control module 140 into the actual action of the UAV motion mechanism, realizing fast and smooth obstacle avoidance of dynamic obstacles in a restricted environment.

[0042] In some embodiments, the vision sensing module 130 includes at least one depth camera (corresponding to a CMOS), which is disposed on the vehicle body of the controlled object and used to acquire depth images of the near-peripheral area of ​​the vehicle body to identify nearby obstacles. The output of the depth camera (corresponding to CMOS) is connected to the input of the microcontroller U101 to receive and acquire depth images and analyze and process the input depth images. For example, a quadcopter drone performs an inspection mission in an underground utility tunnel under low-light conditions, flying at a speed of 1.5 m / s and an altitude of 0.8 m above the ground. The tunnel has a complex internal structure with nearby obstacles such as cable trays, flanges, hanging dripping water crystals, and temporarily stacked maintenance tools.

[0043] Hardware configuration and installation of vision sensing module 130: The drone has a depth camera (RGB-D camera, such as Intel RealSense D435 or similar model) mounted on its bottom and front. Forward depth camera: mounted below the nose of the drone, with a pitch angle tilted downwards by 15°, used to acquire depth images within a range of 0.5m to 5m in front of the drone, focusing on near-field obstacles along the flight path; Downward-looking depth camera: Mounted directly below the fuselage, vertically downward, it is used to acquire depth images within a range of 0.1m to 1.5m around the ground, focusing on the ground conditions at the take-off and landing point and obstacles below (such as protruding pipe flanges).

[0044] Data streams from the two depth cameras are transmitted to the microcontroller U101 in real time at a frequency of 30 Hz.

[0045] In some embodiments, the vision sensing module 130 further includes a data preprocessing subunit (not shown) for real-time filtering and segmentation of the depth image. Median filtering or bilateral filtering can be used to remove outlier noise (such as flying spots caused by light reflection) from depth images. By fitting the ground plane using the RANSAC algorithm, the depth image is segmented into two regions: "ground" and "obstacles," reducing false detections in non-obstacle regions. Euclidean clustering or DBSCAN algorithms are used to aggregate neighboring depth pixels into independent obstacle instances, providing instance-level input for subsequent obstacle tracking and obstacle avoidance decisions.

[0046] In some implementations, the attitude sensing module 120 includes an inertial measurement unit for acquiring the attitude angle of the obstacle in real time, and the main control module 140 determines the interference risk between the controlled object and the surrounding environment based on the attitude angle.

[0047] For example, a hexacopter drone performs low-altitude flight missions in canyons or densely built-up areas, flying at an altitude of 5 meters above the ground and a speed of 4 m / s. The area has complex wind conditions and terrain changes. The drone's own attitude angle and the spatial attitude of surrounding obstacles (such as cliff walls, building facades, suspension bridge cables, etc.) directly affect flight safety and obstacle avoidance decisions.

[0048] Hardware configuration and working principle of attitude sensing module 120: The attitude sensing module 120 includes at least one inertial measurement unit (IMU), which typically consists of a three-axis gyroscope, a three-axis accelerometer, and a three-axis magnetometer. The IMU is mounted at the center of mass of the UAV and acquires the following data in real time at a sampling frequency of 100 Hz to 500 Hz: Three-axis angular velocities (gyroscope): pitch angular velocity, roll angular velocity, yaw angular velocity; Triaxial acceleration (accelerometer): linear acceleration components in the body coordinate system; Magnetic field direction (magnetometer): helps determine the absolute heading angle.

[0049] By integrating angular velocity and fusing acceleration (e.g., using complementary filtering or extended Kalman filtering), the IMU can output the absolute attitude angles (Euler angles) of the UAV in real time: pitch angle θ, roll angle φ, and yaw angle ψ, with an accuracy typically within ±0.5°.

[0050] For example, a drone needs to pass through a vertical gap only 1.2m wide between two buildings. The drone's fuselage is 0.8m wide, and its outstretched arms are 1.1m wide. Under ideal conditions (roll angle = 0°), the drone can pass safely.

[0051] Interference risk assessment process: The attitude sensing module 120 detected in real time that the current roll angle of the drone is 8° (tilting to the right), which is caused by lateral wind pressure. The projection width of the main control module 140 computer body is: actual pass-through width = body width / cos(roll angle) = 0.8m / cos(8°)≈0.81m, and the projection width of the arm = 1.1m / cos(8°)≈1.11m; Comparing the projection width (1.11m) with the gap width (1.2m), the safety margin is only 0.09m. Taking into account attitude fluctuation (±2°) and positioning error (±0.05m), the main control module 140 determines the interference risk level to be high. The generated adjustment signal is as follows: First, the roll angle is corrected to within 2° by adjusting the motor speed difference, and then the crossing action is performed to ensure that the safety margin is increased to more than 0.2m.

[0052] Secondly, a control system for obstacle avoidance in a confined environment detects the dynamic distance between the controlled object and the obstacle in front through at least one radar sensing module 110, acquires the attitude parameters of the obstacle in three-dimensional space in real time through at least one attitude sensing module 120, and detects the morphological parameters of the obstacle in three-dimensional space through at least one visual sensing module 130. The dynamic distance, attitude parameters, and shape parameters are input into a main control module 140, which fuses the dynamic distance, attitude parameters, and shape parameters to generate comprehensive obstacle feature information. The main control module 140 generates adjustment signals based on the comprehensive feature information of the obstacle and the preset obstacle avoidance strategy; The adjustment signal is output to an execution module 150, which corrects the movement path of the controlled object according to the adjustment signal to avoid obstacles.

[0053] For example, the system divides the space in front of the controlled object into different functional areas. For instance, the distance from the vehicle body is 0-0.5 meters, which is designated as the "emergency braking zone", 0.5-2 meters as the "obstacle avoidance adjustment zone", and more than 2 meters as the "far-field detour prediction zone".

[0054] Among them, the lidar is responsible for detecting the precise distance and outline information of obstacles in each area, the visual sensing module identifies the specific type of obstacle (such as pedestrians, stairs, walls) and its dynamic intent, and the ultrasonic sensor focuses on supplementing the detailed information in the near blind spot (such as glass curtain walls, thin columns and other targets that may be missed by vision and laser).

[0055] The main control module 140 performs spatiotemporal synchronization and confidence-weighted fusion of data from the three types of sensors to generate a global environment model that includes geometric boundaries, attribute labels, and motion trends. Based on this model, the decision-making unit can accurately plan the optimal path that balances safety and smoothness. For example, it can initiate detours in advance when static obstacles are detected in the far field, fine-tune the trajectory in real time according to the motion trend of dynamic obstacles in the mid-field, and trigger emergency braking in the near field, thereby achieving layered and precise responses to obstacles of different distances and types.

[0056] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A controller for obstacle avoidance in confined environments, characterized in that, have: At least one radar sensing module is used to detect the dynamic distance between the controlled object and the obstacle in front; At least one attitude sensing module is used to acquire the attitude parameters of the obstacle in three-dimensional space in real time; At least one visual sensing module is used to detect the shape parameters of the obstacle in three-dimensional space; A main control module, whose input terminals are respectively connected to the output terminals of the radar sensing module, the attitude sensing module and the vision sensing module, outputs adjustment signals according to the dynamic distance, the attitude parameters and the shape parameters; An execution module, whose input is connected to the output of the main control module, corrects the movement path of the controlled object according to the adjustment signal to avoid obstacles.

2. The controller for obstacle avoidance in confined environments according to claim 1, characterized in that, The main control module acquires the current speed of the controlled object with a planning period of 100ms to 200ms, uses the dynamic window method to sample multiple feasible trajectories in the current speed space, and performs cost evaluation based on the distance between each feasible trajectory and the obstacle, the motion constraints of the controlled object and the predicted dynamic obstacle motion trend, and selects the optimal trajectory to generate a replanning path.

3. The controller for obstacle avoidance in confined environments according to claim 1 or 2, characterized in that, The main control module includes at least the following: A data processing unit is used to perform spatiotemporal alignment and feature fusion of the dynamic distance, the attitude parameters and the morphological parameters to generate obstacle fusion feature data; A decision-making unit, whose input is connected to the data processing unit, generates the adjustment signal based on the obstacle fusion feature data.

4. The controller for obstacle avoidance in confined environments according to claim 3, characterized in that, The main control module also includes a storage unit. The storage unit contains a pre-stored obstacle avoidance strategy mapping table corresponding to different obstacle types; The decision-making unit identifies the obstacle type based on the obstacle fusion feature data, matches the corresponding obstacle avoidance strategy from the obstacle avoidance strategy mapping table, and generates the adjustment signal.

5. The controller for obstacle avoidance in confined environments according to claim 3, characterized in that, The execution module includes a motion control unit and a drive unit. The motion control unit is used to analyze the adjustment signal and generate speed control commands and steering control commands; The input terminal of the drive unit is connected to the output terminal of the motion control unit, and it drives the motion mechanism of the controlled object according to the speed control command and the steering control command.

6. The controller for obstacle avoidance in confined environments according to claim 1, characterized in that, The visual sensing module includes at least one depth camera, which is installed on the vehicle body of the controlled object to acquire depth images of the near-peripheral area of ​​the vehicle body in order to identify nearby obstacles.

7. The controller for obstacle avoidance in confined environments according to claim 1, characterized in that, The attitude sensing module includes an inertial measurement unit for acquiring the attitude angle of the obstacle in real time, and the main control module determines the interference risk between the controlled object and the surrounding environment based on the attitude angle.

8. A control system for obstacle avoidance in confined environments, characterized in that, The dynamic distance between the controlled object and the obstacle in front is detected by at least one radar sensing module, the attitude parameters of the obstacle in three-dimensional space are obtained in real time by at least one attitude sensing module, and the shape parameters of the obstacle in three-dimensional space are detected by at least one visual sensing module. The dynamic distance, the attitude parameters, and the shape parameters are input into a main control module, and the dynamic distance, attitude parameters, and shape parameters are fused to generate comprehensive obstacle feature information. The main control module generates adjustment signals based on the comprehensive feature information of the obstacle and the preset obstacle avoidance strategy; The adjustment signal is output to an execution module, which then corrects the movement path of the controlled object based on the adjustment signal to avoid the obstacle.

Images