An indoor and outdoor logistics robot navigation system

Abstract

The application relates to the technical field of robot navigation, and discloses an indoor and outdoor logistics robot navigation system. The environment perception module of the system collects three-dimensional point cloud data, visible light image data and global navigation satellite system original observation values around the robot in real time, and realizes multi-dimensional environment information acquisition. The positioning and mapping module fuses and processes the collected environment information, constructs a dense point cloud map containing semantic labels, and calculates the six-degree-of-freedom pose of the robot. The path planning module searches a collision-free passing path in the map according to a task target point and the pose. The motion control module converts the path into a driving motor control instruction, and realizes robot movement. The navigation decision module receives data of each module, evaluates the space-time relationship between dynamic obstacles and the passing path, and generates a speed adjustment or local re-planning instruction. The system realizes efficient and safe navigation of the logistics robot in an indoor and outdoor complex environment.

Description

Technical Field

[0001] This invention relates to the field of robot navigation technology, specifically to an indoor and outdoor logistics robot navigation system. Background Technology

[0002] As public service facilities operating around the clock, hospitals face heavy daily logistics tasks, including drug delivery, medical equipment transport, medical waste collection, and the supply of logistical materials. Traditional hospital logistics relies on manual trolleys or dedicated channels, which suffers from low efficiency, high personnel costs, a high risk of cross-infection, and nighttime deliveries disrupting patients' rest. In recent years, hospitals have begun to introduce logistics robots to replace some manual transportation tasks, but existing robot navigation systems still face many challenges in the complex hospital environment.

[0003] Hospital environments are typically characterized by interconnected indoor and outdoor spaces, and a combination of static and dynamic environments. Indoor areas include the outpatient hall, inpatient corridors, and restricted operating room areas, featuring complex spatial structures and high traffic volumes, potentially leading to temporary congestion during peak periods. Outdoor areas include connecting corridors, hospital roads, and underground parking garages, requiring consideration of weather changes, lighting variations, and GPS signal obstruction. Existing logistics robots mostly employ a single navigation method, such as indoor navigation based on LiDAR or outdoor navigation relying on GNSS, making seamless switching between indoor and outdoor environments difficult. When robots traverse building entrances and exits, mission interruptions often occur due to sudden changes in positioning signals or map discontinuities.

[0004] In terms of dynamic obstacle avoidance, hospital scenarios present numerous moving obstacles, such as hospital bed transport vehicles, medical staff, patients, and their families. Traditional robots often employ reactive obstacle avoidance strategies, only braking or detouring when an obstacle is detected, lacking the ability to predict the movement trends of obstacles. This passive obstacle avoidance method easily leads to disjointed robot movement and frequent starts and stops, not only reducing transportation efficiency but also potentially causing staff stress due to sudden movements. Furthermore, some areas of the hospital (such as the ICU and radiology department) have strict restrictions on robot access paths, requiring avoidance of sensitive equipment or isolation wards, while existing systems lack semantic environment understanding capabilities, making it difficult to achieve fine-grained path constraints.

[0005] Current research attempts to improve navigation robustness through multi-sensor fusion, such as combining visual and laser data to build environmental models. However, these systems typically focus on static map construction, and their handling of dynamic objects remains limited to real-time collision detection. Furthermore, existing navigation decision modules often employ rule-based logic, which cannot adapt to the temporal fluctuations in hospital traffic (such as the difference in obstacle density between lunchtime delivery peaks and nighttime patrols), leading to overly conservative or aggressive robot behavior. Therefore, there is an urgent need for a navigation system capable of navigating both indoor and outdoor environments, perceiving semantic information, and proactively predicting dynamic obstacle behavior to meet the high standards of safety, continuity, and adaptability required by hospital logistics. Summary of the Invention

[0006] The purpose of this invention is to provide an indoor and outdoor logistics robot navigation system to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides an indoor and outdoor logistics robot navigation system, the system comprising:

[0008] The environmental perception module is used to collect environmental information around the robot in real time. The environmental information includes three-dimensional point cloud data, visible light image data and raw observations from the Global Navigation Satellite System.

[0009] The localization and mapping module is used to fuse the environmental information, construct a dense point cloud map containing semantic labels, and calculate the robot's six-degree-of-freedom pose based on the map.

[0010] The path planning module searches for a collision-free path in the dense point cloud map based on the task target point and the six-degree-of-freedom pose. The path consists of a series of path points and their corresponding robot motion postures.

[0011] A motion control module is used to convert the travel path into control commands for the drive motors and execute the commands to make the robot move along the path;

[0012] The navigation decision module receives data from the environment perception module, the positioning and mapping module, the path planning module, and the motion control module. By evaluating the spatiotemporal relationship between the motion trajectory of dynamic obstacles in the environment and the passage path, it generates a decision signal containing speed adjustment instructions and local replanning instructions.

[0013] The navigation decision module performs the following operations: predicts the area occupied by the dynamic obstacles in the environment in the dense point cloud map at future times based on their movement trajectories; calculates the degree of overlap between the travel path and the occupied area; and when the degree of overlap exceeds a preset threshold, generates a speed adjustment command to reduce the robot's movement speed or generates a local replanning command to replan the local path.

[0014] Preferably, the environment sensing module includes:

[0015] A lidar unit is used to emit a laser beam and receive echo signals to generate the three-dimensional point cloud data.

[0016] A visual sensor unit is used to acquire the visible light image data and extract edge features and texture features from the scene.

[0017] The multi-band satellite signal receiving unit is used to receive the raw observation values ​​of the global navigation satellite system and calculate the robot's latitude and longitude coordinates and altitude.

[0018] Preferably, the positioning and mapping module includes:

[0019] The point cloud registration unit is used to iteratively match the three-dimensional point cloud data collected at different times with the constructed dense point cloud map to calculate the robot's position change.

[0020] The visual inertial odometry unit is used to combine the visible light image data and the angular velocity and linear acceleration data of the inertial measurement unit to calculate the continuous motion trajectory of the robot through a nonlinear optimization method.

[0021] The compact navigation unit is used to fuse the raw observations of the global navigation satellite system with the output of the visual inertial odometry unit using Kalman filtering to generate the six-degree-of-freedom pose.

[0022] Preferably, the path planning module includes:

[0023] The global planning unit is used to calculate an initial global path from the robot's current six-degree-of-freedom pose to the task target point in the dense point cloud map using a graph search-based path planning algorithm.

[0024] The local planning unit is used to locally adjust the initial global path based on the temporary obstacle information detected in real time by the environment perception module, and to generate a smooth trajectory that satisfies the robot's kinematic constraints.

[0025] Preferably, the motion control module includes:

[0026] The trajectory tracking unit is used to discretize the smooth trajectory into a series of pose points on a time scale, and calculate the target rotation speed and steering angle of the robot drive wheel at each moment;

[0027] The motor drive unit is used to generate a corresponding motor drive voltage signal based on the target rotation speed and steering angle using pulse width modulation technology.

[0028] The motion feedback unit is used to collect the actual speed and steering angle of the drive motor and compare them with the target value to form a closed-loop control.

[0029] Preferably, the navigation decision module includes:

[0030] The obstacle prediction unit is used to model the motion state of the dynamic obstacles in the environment and use a linear dynamic system to predict their position and velocity distribution in the next few seconds.

[0031] The collision detection unit is used to perform a spatiotemporal synchronous comparison between the smooth trajectory and the position and velocity distribution, and calculate the degree of overlap between the travel path and the occupied area.

[0032] The decision generation unit is used to selectively generate the speed adjustment instruction or the local replanning instruction based on the degree of overlap.

[0033] Preferably, the system further includes:

[0034] The communication management module is used to interact with the central dispatch server via a wireless network, receive the task target point and map update information, and upload the robot's real-time status data and task execution progress.

[0035] The communication management module is configured to switch to a distributed communication mode based on edge computing nodes when the wireless signal strength is lower than a preset threshold, so as to maintain basic navigation functions.

[0036] Preferably, the system further includes:

[0037] The energy management module is used to monitor the remaining power and instantaneous power consumption of the robot's battery, and to estimate the energy consumption of the entire task based on the length and complexity of the smooth trajectory.

[0038] The energy management module is configured to send a charging request signal to the path planning module and plan a path to the nearest charging station when the remaining power is lower than a preset percentage of the power required to complete the current task.

[0039] Preferably, the system further includes:

[0040] The fault diagnosis module is used to periodically detect the operating status of the environmental perception module, the positioning and mapping module, the path planning module, the motion control module, and the navigation decision module;

[0041] The fault diagnosis module is configured to record a fault code and trigger a safety protection mechanism when any module malfunctions, causing the robot to decelerate and stop or return to the starting point along the original path.

[0042] Preferably, the system further includes:

[0043] The data recording module is used to continuously store the raw data collected by the environmental perception module, the map data generated by the positioning and mapping module, the path data calculated by the path planning module, the execution data of the motion control module, and the decision data of the navigation decision module.

[0044] The data recording module is configured to build a complete navigation process dataset in chronological order for subsequent offline analysis and algorithm optimization.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] This system constructs a dense point cloud map with semantic labels through multimodal perception data fusion, enabling the robot to uniformly represent indoor and outdoor environmental characteristics and eliminate localization discrepancies during scene transitions. Semantic labels introduce the ability to identify functional areas of the hospital (such as corridors, elevator lobbies, and hazardous areas), allowing path planning to select compliant routes based on the type of transportation task, avoiding intrusion into sensitive medical areas. The navigation decision module assesses the interaction between dynamic obstacles and the planned path based on spatiotemporal relationships, and adjusts the robot's motion strategy in advance by predicting obstacle trajectories, reducing the number of sudden stops or passive detours and improving movement stability.

[0047] The dynamic obstacle occupancy prediction and path overlap calculation mechanism enables the robot to proactively adopt deceleration or local path reconstruction strategies in densely populated areas, ensuring both passage safety and maintaining task continuity. The synergistic effect of speed adjustment commands and local replanning commands adapts to changes in patient density at different times in the hospital, optimizing transportation efficiency while ensuring safety. The multi-source data fusion-based localization and mapping method mitigates the limitations of single sensors, maintaining pose estimation stability even in scenarios lacking GNSS or visual features, such as glass curtain walls and low-light corridors.

[0048] The system's modular design supports functional expansion, such as embedding special constraints like disinfection areas and negative pressure wards into dense point cloud maps, reserving interfaces for future integration of integrated disinfection and transportation functions. Environmental update data generated by the robot during transportation tasks can be fed back to the map database, enabling continuous optimization of navigation resources throughout the hospital. This navigation capability reduces reliance on pre-set beacons or manual guidance facilities, shortens robot deployment cycles, and adapts to environmental changes brought about by adjustments to hospital department layouts. Attached Figure Description

[0049] Figure 1 This is a schematic diagram illustrating the working principle of the indoor and outdoor logistics robot navigation system described in this invention.

[0050] Figure 2 A flowchart illustrating the working principle of the environmental perception module;

[0051] Figure 3 This is a flowchart illustrating the working principle of the motion control module.

[0052] Figure 4 A graph for trajectory planning and decision analysis in intelligent navigation systems;

[0053] Figure 5This is a diagram for collaborative monitoring and analysis of communication and energy management. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Please see Figure 1 This invention provides an indoor and outdoor logistics robot navigation system, comprising an integrated and collaborative environmental perception module, a localization and mapping module, a path planning module, a motion control module, and a navigation decision module. Specific implementation details are as follows:

[0056] The environmental perception module is responsible for real-time acquisition of environmental information surrounding the robot, including 3D point cloud data, visible light image data, and raw observations from the Global Navigation Satellite System (GNSS). This data is directly acquired and pre-processed by sensor hardware. The localization and mapping module receives the output from the environmental perception module and fuses the 3D point cloud data, visible light image data, and raw GNSS observations. It constructs a dense point cloud map with semantic labels using point cloud registration and visual-inertial odometry (VIO) technology. Simultaneously, it calculates the robot's six-DOF pose using a compact combination navigation method to ensure consistent positioning accuracy in both indoor and outdoor environments. The path planning module, based on the task target point and the six-DOF pose, uses a graph search-based algorithm to search for a collision-free path in the dense point cloud map. This path consists of a series of path points and their corresponding robot motion postures, adapting to different terrains and obstacle distributions. The motion control module converts the path into control commands for the drive motors. Through trajectory tracking and motor drive unit execution of these commands, the robot moves along the path, and closed-loop control is achieved using a motion feedback unit. The navigation decision module receives data from all modules and generates decision signals containing speed adjustment and local replanning instructions by evaluating the spatiotemporal relationship between the movement trajectory of dynamic obstacles in the environment and the travel path. Specifically, the navigation decision module predicts the area occupied by the dynamic obstacles in the environment in the dense point cloud map at future times based on their movement trajectory, calculates the degree of overlap between the travel path and the occupied area, and automatically triggers the speed adjustment or local replanning mechanism when the degree of overlap exceeds a preset threshold to ensure navigation safety and efficiency.

[0057] Example 1: See Figure 2The environmental perception module includes a lidar unit, a vision sensor unit, and a multi-band satellite signal receiving unit. The lidar unit generates 3D point cloud data by emitting laser beams and receiving echo signals. The lidar unit operates at a frequency of 10 Hz to ensure real-time acquisition. The lidar unit uses 16-line lidar hardware, and the output data format includes spatial coordinates and reflection intensity information. The vision sensor unit acquires visible light image data. The vision sensor unit uses a high-definition camera to capture images at a rate of 30 frames per second and integrates image processing algorithms to extract edge features and texture features in the scene. The multi-band satellite signal receiving unit receives raw observations from the Global Navigation Satellite System. The multi-band satellite signal receiving unit supports both GPS and BeiDou systems and obtains the robot's latitude and longitude coordinates and altitude by calculating pseudorange and carrier phase data. In practice, the units of the environmental perception module are synchronized through a unified data interface. The data acquisition cycle is aligned with the system clock to avoid timestamp misalignment. The point cloud data of the lidar unit is transmitted via Ethernet, the image data of the vision sensor unit is sent via USB interface, and the data of the multi-band satellite signal receiving unit is transmitted via serial communication. All raw data is buffered within the environmental perception module and then forwarded to the positioning and mapping module.

[0058] The localization and mapping module includes a point cloud registration unit, a visual inertial odometry unit, and a compact integrated navigation unit. The point cloud registration unit iteratively matches the 3D point cloud data collected at different times with the constructed dense point cloud map. The point cloud registration unit uses an iterative nearest-point algorithm to calculate the robot's position change. The matching process uses the dense point cloud map as a reference frame and optimizes the transformation matrix by minimizing the point-to-point distance. The visual inertial odometry unit combines visible light image data and angular velocity and linear acceleration data from the inertial measurement unit. The visual inertial odometry unit calculates the robot's continuous motion trajectory based on a nonlinear optimization method. The optimization process uses a graph optimization framework to process visual feature points and inertial data residuals. The compact integrated navigation unit fuses the original observations from the global navigation satellite system with the output of the visual inertial odometry unit using Kalman filtering. The compact integrated navigation unit uses an extended Kalman filter to process sensor noise and generate a six-degree-of-freedom pose output. In specific implementations, the matching frequency of the point cloud registration unit is set to 5 Hz, the optimization frequency of the visual inertial odometry unit is 10 Hz, and the fusion period of the compact navigation unit is 100 milliseconds. During the initialization phase, the point cloud registration unit loads a priori dense point cloud map. Upon startup, the visual inertial odometry unit executes a calibration procedure to calibrate sensor deviations. When satellite signals are lost, the compact navigation unit relies on the visual inertial odometry unit to maintain pose output. In some embodiments, the point cloud registration unit supports multi-resolution map matching to improve efficiency in large scenes, the visual inertial odometry unit integrates a keyframe management mechanism to reduce computational load, and the compact navigation unit can configure filtering parameters to adapt to different environmental dynamics.

[0059] In practical implementation, the environmental perception module and the positioning and mapping module exchange data through the ROS middleware. The environmental perception module publishes topic data on point clouds, images, and satellite data, while the positioning and mapping module subscribes to these topics and publishes pose and map updates. Point cloud data from the LiDAR unit is filtered to remove noise, image data from the visual sensor unit undergoes grayscale conversion and feature extraction preprocessing, and data from the multi-band satellite signal receiving unit is decoded and converted into a unified coordinate system. The point cloud registration unit uses kd-trees to accelerate nearest-point search during the matching process, the visual inertial odometry unit uses ORB features to describe and track image sequences, and the Kalman filter state vector of the compactly combined navigation unit contains position, velocity, and attitude angles. It can be understood that the synchronization mechanism of the environmental perception module is based on hardware trigger signals to ensure spatiotemporal alignment of multi-sensor data, and the fusion strategy of the positioning and mapping module prioritizes high-confidence sensor data; for example, the visual inertial odometry unit has a higher weight in indoor environments, while raw observations from the global navigation satellite system dominate the fusion results outdoors. Optionally, the lidar unit can be replaced with a solid-state lidar to reduce size, the vision sensor unit can be equipped with an infrared camera to enhance low-light performance, and the multi-band satellite signal receiving unit can be integrated with an inertial navigation system as a backup.

[0060] In practical implementation, the iterative nearest point algorithm of the point cloud registration unit includes two modes: point-to-point matching and point-to-plane matching, which are automatically switched according to the scene complexity. The nonlinear optimization of the visual inertial odometry unit uses the g2o library to solve the bundle adjustment problem. The Kalman filter prediction and update steps of the compactly integrated navigation unit are processed separately to prevent numerical instability. The data flow of the environment perception module is managed using a circular buffer to avoid data loss. The map update mechanism of the localization and mapping module triggers reconstruction only when the pose change exceeds a threshold to reduce computational overhead. In some embodiments, the point cloud registration unit supports incremental map updates, the visual inertial odometry unit can enable loop closure detection to correct accumulated errors, and the compactly integrated navigation unit can fuse differential signals from multiple base stations to improve accuracy. It can be understood that the power consumption of the environment perception module is controlled by dynamically adjusting the sampling rate, the real-time performance of the localization and mapping module is ensured by multi-threaded parallel processing, and the outputs of the point cloud registration unit and the visual inertial odometry unit are fused after time alignment in the compactly integrated navigation unit. Optionally, point cloud data from the lidar unit can be compressed for transmission to save bandwidth, image data from the vision sensor unit can be processed at reduced resolution to improve speed, and data from the multi-band satellite signal receiving unit can be cached and recalculated offline.

[0061] In practical implementation, the calibration procedure of the environmental perception module runs automatically upon system startup. The extrinsic parameters of the lidar and visual sensor units are obtained through checkerboard calibration. The antenna phase center deviation of the multi-band satellite signal receiving unit is pre-compensated. The initialization of the positioning and mapping module requires at least 10 seconds of sensor data to establish a stable trajectory. The map loading time of the point cloud registration unit is optimized to milliseconds. Feature tracking of the visual inertial odometry unit continuously verifies matching consistency. The filtering initialization of the compact navigation unit uses static detection to avoid drift. Fault handling for the environmental perception module includes data timeout checks and sensor heartbeat monitoring. The anomaly recovery mechanism of the positioning and mapping module reinitializes failed units. In practical implementation, the coordinate transformation of 3D point cloud data is unified to the world coordinate system. Feature extraction of visible light image data uses the SIFT algorithm. The calculation of the original observations from the global navigation satellite system uses the least squares method. The matching error threshold for the point cloud registration unit is set to 0.1 meters. The optimization iteration count of the visual inertial odometry unit is limited to 50 times. The filtering covariance matrix of the compact navigation unit is periodically reset. It is understandable that the environmental perception module performs time interpolation before fusing multi-sensor data, the pose output frequency of the localization and mapping module matches the requirements of the path planning module, and the intermediate result logs of the point cloud registration unit and the visual inertial odometry unit are used for debugging. Optionally, the lidar unit can be configured with scanning modes to adapt to different distances, the visual sensor unit can adjust exposure parameters to optimize image quality, and the multi-band satellite signal receiving unit can switch satellite system priorities.

[0062] In practical implementation, the data acquisition link of the environmental perception module is scheduled by the main controller. Interrupt service routines of the LiDAR unit, visual sensor unit, and multi-band satellite signal receiving unit process real-time data. The computational tasks of the positioning and mapping module are allocated to a dedicated processor. The point cloud registration unit uses a GPU to accelerate the iterative nearest-point algorithm. The visual inertial odometry unit runs optimized code on a DSP. The filtering calculations of the compact navigation unit are implemented with low latency on an FPGA. The power management of the environmental perception module dynamically shuts down idle sensors. The data storage of the positioning and mapping module uses non-volatile memory to save map snapshots. In practical implementation, the 3D point cloud data format is standardized to PCD files, visible light image data is encoded as a JPEG stream, raw observations from the Global Navigation Satellite System are stored in RINEX format, the matching results of the point cloud registration unit output a transformation matrix, the trajectory data of the visual inertial odometry unit includes timestamps and covariance, and the six-DOF pose publication of the compact navigation unit is a TF transform. It is understandable that the interface protocol between the environmental perception module and the positioning and mapping module defines strict message types to ensure interoperability between modules. Data synchronization between the point cloud registration unit and the visual inertial odometry unit is based on hardware timestamps, and the fusion output of the tightly coupled navigation unit undergoes smoothing filtering to reduce jitter. Optionally, the lidar unit can be equipped with a heater to prevent condensation, the visual sensor unit can integrate autofocus functionality, and the multi-band satellite signal receiving unit can support external clock input.

[0063] In practical implementation, the deployment of the environmental perception module considers mechanical structure vibration reduction; the installation positions of the lidar unit and visual sensor unit optimize field-of-view overlap; the antenna arrangement of the multi-band satellite signal receiving unit is far away from interference sources; the algorithm parameters of the positioning and mapping module are adjusted through configuration files; the matching threshold of the point cloud registration unit dynamically adapts to the environment; the feature quantity limit of the visual inertial odometry unit balances accuracy and speed; and the filtering noise model of the compact navigation unit is updated online. Maintenance of the environmental perception module includes regularly cleaning the sensor lenses and the positioning and mapping module's ground... Figure 1 Consistency checks prevent corruption. In specific implementations, the laser beam divergence angle of the LiDAR unit is set to 0.1 degrees, the image resolution of the visual sensor unit is selected as 1280x720 pixels, the update rate of the multi-band satellite signal receiving unit is 1 Hz, the iterative convergence condition of the point cloud registration unit is an error change of less than 1e-5, the optimization tolerance of the visual inertial odometry unit is set to 1e-6, and the state vector dimension of the compact navigation unit is 16. It can be understood that the data integrity of the environmental perception module is guaranteed by CRC check, the real-time performance of the positioning and mapping module is verified by benchmark testing, and the output delay measurement of the point cloud registration unit and the visual inertial odometry unit ensures system response. Optionally, the environmental perception module can be expanded with a temperature sensor to monitor hardware status, and the positioning and mapping module can integrate a semantic segmentation network to enrich map labels.

[0064] Example 2: See Figure 3 The path planning module comprises a global planning unit and a local planning unit. The global planning unit employs a graph search-based path planning algorithm in a dense point cloud map to calculate an initial global path from the robot's current six-DOF pose to the target point. Specifically, the global planning unit implements either the A* algorithm or Dijkstra's algorithm for path search. During the search process, the dense point cloud map is discretized into a grid network, with each grid containing passage cost information. The initial global path output is a sequence of path points, each containing 3D coordinates and the robot's orientation angle. The local planning unit, based on temporary obstacle information detected in real-time by the environmental perception module, uses a spline curve-based trajectory optimization method to locally adjust the initial global path. The local planning unit uses a B-spline curve fitting algorithm to generate a smooth trajectory that satisfies the robot's kinematic constraints, taking into account the robot's maximum turning radius and acceleration limitations. The motion control module includes a trajectory tracking unit, a motor drive unit, and a motion feedback unit. The trajectory tracking unit discretizes the smooth trajectory into a series of pose points on a time scale. The trajectory tracking unit calculates the target speed and steering angle of the robot's drive wheels at each moment using an interpolation algorithm. The motor drive unit generates the corresponding motor drive voltage signal based on the target speed and steering angle using pulse width modulation technology. The motion feedback unit collects the actual speed and steering angle of the drive motor and compares them with the target values ​​to form closed-loop control.

[0065] In practical implementation, the global planning unit's map resolution is set to 5 cm, the path search step size is fixed at 10 cm, the A* algorithm's cost function comprehensively considers path length and terrain slope, the heuristic function uses Euclidean distance estimation, and the global planning unit's execution frequency is 1 Hz, but replanning is triggered immediately upon receiving a new task target point. The local planning unit's trajectory optimization frequency is set to 20 Hz, the number of control points for the B-spline curve is dynamically adjusted according to the path length, typically maintained between 10 and 20 points, and the local planning unit subscribes to obstacle topics published by the environmental perception module in real time; when a temporary obstacle is detected, trajectory re-optimization is completed within 200 milliseconds. The trajectory tracking unit's control cycle is configured to 1 millisecond, and the pose point interpolation uses a cubic spline interpolation algorithm to ensure the continuity of the velocity curve. The motor drive unit's pulse width modulation frequency is set to 20 kHz to reduce motor torque ripple. The motion feedback unit's encoder data sampling rate is 100 kHz, and the closed-loop control algorithm uses an incremental PID controller; the proportional, integral, and derivative coefficients are obtained through experimental calibration. The path planning module and the motion control module exchange data through a shared memory area. The initial global path output by the global planning unit is written into a circular buffer, and the local planning unit reads data from it for local optimization. The smooth trajectory is sent to the motion control module through a message queue.

[0066] In some embodiments, the global planning unit supports multi-target point path planning, capable of calculating the optimal path sequence passing through multiple intermediate points at once. The trajectory optimization algorithm of the local planning unit integrates the dynamic window method to evaluate the feasibility of trajectory velocity and acceleration in real time. The trajectory tracking unit of the motion control module can be configured with a feedforward control mode to compensate for tracking errors caused by system inertia in advance. In a specific implementation, the initialization process of the path planning module requires loading a pre-built dense point cloud map. The map data is stored in memory in an octree structure. Before path search, the global planning unit performs obstacle expansion processing, with the expansion radius set to 1.2 times the radius of the robot's circumcircle. The B-spline curve fitting of the local planning unit uses a constraint optimization algorithm. Hard constraints include maximum curvature limits and obstacle avoidance distances, while soft constraints focus on trajectory smoothness optimization. The discretization process of the trajectory tracking unit divides the smooth trajectory into 10-millisecond intervals, with each pose point containing a timestamp, position coordinates, and attitude quaternions. The pulse width modulation signal of the motor drive unit drives the brushless DC motor through an H-bridge circuit, and the encoder signal of the motion feedback unit is processed by a quadruple frequency decoding circuit to improve measurement resolution.

[0067] It can be understood that the global planning unit and local planning unit of the path planning module constitute a hierarchical planning architecture. The global planning unit solves the macroscopic path optimization problem, while the local planning unit handles the microscopic dynamic obstacle avoidance task. The trajectory tracking unit, motor drive unit, and motion feedback unit of the motion control module form a cascaded control loop. The trajectory tracking unit acts as a position loop controller, the motor drive unit acts as a velocity loop actuator, and the motion feedback unit constitutes the feedback loop. In specific implementation, the path search algorithm of the global planning unit pre-calculates the reachability map at startup to accelerate the online search process. The trajectory optimization of the local planning unit uses the Jacobian matrix to analytically calculate the gradient, improving the convergence speed. The interpolation algorithm of the trajectory tracking unit considers the differential drive characteristics of the robot, converting the pose point into left and right wheel speed commands. The pulse width modulation duty cycle of the motor drive unit is linearly related to the target rotational speed. The PID controller output of the motion feedback unit is amplitude-limited to prevent integral saturation.

[0068] In practical implementation, the path planning module interacts with the upper-level navigation decision module through ROS service calls. Upon receiving a local replanning instruction, the local planning unit immediately interrupts the current trajectory generation process and replans the trajectory starting from the latest robot pose. The motion control module and the underlying motor driver use the CAN bus communication protocol. The control commands issued by the trajectory tracking unit include a checksum field to ensure data transmission reliability. The path point sequence of the global planning unit is stored as a linked list structure, supporting dynamic insertion and deletion operations. The smoothed trajectory data of the local planning unit includes a covariance matrix, representing the uncertainty of the trajectory points. The pulse width modulation waveform of the motor drive unit is generated by timer hardware, and the encoder counter of the motion feedback unit uses a 32-bit register to avoid overflow during long-term operation. In some embodiments, the global planning unit can switch to a fast random tree algorithm to handle complex terrain, and the local planning unit can enable an elastic band optimization method to enhance obstacle avoidance capabilities. The trajectory tracking unit of the motion control module can integrate a model predictive controller to improve tracking accuracy.

[0069] Optionally, the path planning module can be configured with a path pruning function to remove redundant path points in the initial global path. The local planning unit can be equipped with a trajectory prediction module to estimate the trajectory execution status in the next few seconds. The motion control module can be equipped with a temperature monitoring unit to automatically derate when the motor overheats. In specific implementations, the heuristic weight factor of the global planning unit is set to 1.0 to balance search efficiency and path quality; the obstacle avoidance distance threshold of the local planning unit is set to 0.5 meters; and the maximum number of trajectory optimization iterations is 100. The pose point cache queue length of the trajectory tracking unit is 1000; the pulse width modulation resolution of the motor drive unit is set to 12 bits; and the data filtering of the motion feedback unit uses a first-order low-pass filter with a cutoff frequency of 100 Hz. The path planning module's memory management employs memory pool technology to avoid fragmentation caused by frequent dynamic allocation; and the interrupt service routine of the motion control module is set to the highest priority to ensure accurate control timing.

[0070] In practical implementation, the global planning unit performs a connectivity check before path search to confirm that the starting point and target point are in the same connected region. The trajectory optimization algorithm of the local planning unit uses the quasi-Newton method to solve the nonlinear least squares problem. The forward prediction module of the trajectory tracking unit predicts the trajectory points at future moments based on historical pose data. The short-circuit protection circuit of the motor drive unit monitors the current magnitude in real time, and the zero-position calibration program of the motion feedback unit is automatically executed when the system starts. It can be understood that the performance of the path planning module depends on the accuracy and completeness of the dense point cloud map, and the response speed of the motion control module directly affects the stability of trajectory tracking. The global planning unit establishes a tight coupling relationship with the localization and mapping module, receiving six-degree-of-freedom pose updates in real time. The local planning unit maintains high-frequency data interaction with the environmental perception module, responding promptly to dynamic obstacle changes. The trajectory tracking unit works collaboratively with the navigation decision module, dynamically scaling the tracking trajectory according to speed adjustment commands.

[0071] Optionally, the global planning unit can output multiple alternative paths for decision-making, while the local planning unit can integrate visual and semantic information to optimize trajectory generation. The motion control module can incorporate a vibration suppression algorithm to actively counteract vibrations caused by mechanical transmission. In practical implementation, the path planning module's anomaly handling mechanism monitors path planning timeouts, and the motion control module's safety protection circuit detects motor stall events. The global planning unit uses convolutional filtering for path smoothing post-processing, and the real-time performance of the local planning unit is ensured through computation time monitoring. The trajectory tracking unit's control parameters can be self-tuned online, and the motor drive unit's fault diagnosis code records its operating status. The motion feedback unit's accuracy calibration uses a laser interferometer to ensure accurate and reliable measurement data.

[0072] In practical implementation, the integration testing of the path planning module and motion control module is conducted in a simulation environment to verify the end-to-end performance from path planning to motor drive. The search efficiency of the global planning unit is evaluated by the number of path points, and the trajectory quality of the local planning unit is checked by curvature continuity. The tracking error of the trajectory tracking unit is quantified using the root mean square value, and the response characteristics of the motor drive unit are tested using step response. The delay time of the motion feedback unit is measured by timestamp comparison to ensure that the stability of the entire control loop meets navigation requirements.

[0073] Example 3: The navigation decision module includes an obstacle prediction unit, a conflict detection unit, and a decision generation unit. The obstacle prediction unit models the motion state of dynamic obstacles in the environment and uses a linear dynamic system to predict their position and velocity distribution within the next few seconds. The obstacle prediction unit processes the dynamic obstacle trajectory data provided by the environmental perception module based on the Kalman filter algorithm, with a prediction time window set to 3 seconds and an update frequency of 10 Hz. The conflict detection unit performs a spatiotemporal synchronous comparison between the smooth trajectory output by the path planning module and the position and velocity distribution obtained by the obstacle prediction unit. The conflict detection unit calculates the degree of overlap between the travel path and the occupied area. The degree of overlap is calculated by integrating the intersection area of ​​the smooth trajectory and the obstacle prediction area at a future time point. The decision generation unit selectively generates speed adjustment instructions or local replanning instructions based on the degree of overlap. The decision generation unit sets two levels of decision thresholds: when the degree of overlap is lower than the first threshold, a speed adjustment instruction is generated; when the degree of overlap is higher than the second threshold, a local replanning instruction is generated. In practical implementation, the navigation decision module's software architecture adopts a multi-threaded design. The main thread is responsible for inter-module communication and data scheduling, the obstacle prediction thread is dedicated to motion prediction calculation, the conflict detection thread performs spatiotemporal comparison analysis, and the decision generation thread processes instruction logic. Each thread exchanges data through shared memory and message queues to ensure real-time response capability. The input data of the obstacle prediction unit comes from the dynamic obstacle list of the environmental perception module. Each obstacle identifier corresponds to a tracking sequence, including timestamp, position coordinates, and velocity vector. The obstacle prediction unit maintains an independent Kalman filter instance for each active obstacle. The filter state vector includes four dimensions: x-position, y-position, x-direction velocity, and y-direction velocity. The state transition matrix is ​​constructed based on a uniform motion model. The process noise covariance matrix is ​​calibrated according to sensor characteristics, and the observation noise covariance matrix is ​​set according to the measurement accuracy of the environmental perception module. The input to the conflict detection unit includes the smooth trajectory message published by the path planning module and the predicted distribution output by the obstacle prediction unit. The smooth trajectory message contains a sequence of path points and corresponding timestamps. The predicted distribution represents the uncertainty of the future position of the obstacle in the form of a covariance ellipse. The execution cycle of the conflict detection unit is synchronized with that of the obstacle prediction unit, both at 10 Hz. The conflict detection unit discretizes the prediction time window into 30 time segments, each segment lasting 0.1 seconds. Within each time segment, it calculates the spatial relationship between the robot's expected position and the predicted obstacle area. The decision generation unit's decision logic is implemented as a finite state machine, containing three states: normal driving state, deceleration and avoidance state, and emergency replanning state. The state transition condition is based on the comparison result of the overlap degree value and a threshold. The first threshold is set to 0.3, and the second threshold is set to 0.7. The output instructions of the decision generation unit are published through ROS topics. The speed adjustment instruction contains a target speed ratio coefficient, and the local replanning instruction contains parameters of the replanning area range.

[0074] In practical implementation, the Kalman filter implementation of the obstacle prediction unit adopts a numerically stable square root Kalman filter algorithm to avoid the covariance matrix losing its positive definiteness. The filter prediction step uses the state transition matrix to project the current state, and the update step fuses new observations to adjust the state estimate. The obstacle prediction unit supports online adaptive model parameter adjustment, automatically adjusting process noise parameters based on prediction errors to improve prediction accuracy. The overlap calculation of the collision detection unit uses a geometric method, projecting the robot body as a two-dimensional polygon and modeling the obstacle prediction region as an ellipse or polygon. The intersection area is calculated using a polygon clipping algorithm, and the overlap quantification formula is defined as an average time integral form:

[0075]

[0076] in: Indicates the degree of overlap. Indicates the length of the prediction time window. Indicates the current moment. This represents the projected area of ​​the robot on a two-dimensional plane. This represents the predicted area occupied by the obstacle at time t.

[0077] The collision detection unit utilizes spatial indexing for computational acceleration, dividing the workspace into a grid structure to quickly retrieve potentially interacting obstacles. Overlap results are smoothed using low-pass filtering to eliminate fluctuations caused by sensor noise. The finite state machine implementation of the decision generation unit includes a timeout protection mechanism to prevent the system from remaining in a certain state. The decision generation unit records historical instructions and execution effects for subsequent analysis and optimization. In some embodiments, the obstacle prediction unit can be expanded into an interactive multi-model system, simultaneously running a uniform velocity model, a uniform acceleration model, and a turning model. The optimal prediction model is selected based on the obstacle's motion characteristics. The collision detection unit can incorporate a velocity-obstacle method for auxiliary collision detection, enhancing the robustness of risk assessment. The decision generation unit can integrate reinforcement learning algorithms to adaptively adjust decision thresholds based on historical data. In practical implementation, the initialization process of the navigation decision module loads the configuration file, sets filter parameters, decision thresholds and algorithm options. After the navigation decision module starts, it enters calibration mode, collects several seconds of environmental data to establish an initial state estimate, the obstacle prediction unit maintains a dynamic memory pool, pre-allocates a fixed number of filter objects to avoid dynamic memory allocation at runtime, the collision detection unit uses a high-performance geometry calculation library to optimize polygon operation efficiency, and the output instructions of the decision generation unit are stamped with timestamps and sequence numbers to ensure synchronization with the receiving module.

[0078] In some embodiments, the obstacle prediction unit can fuse visual trajectory prediction information and improve the motion model by combining the appearance features of obstacles. The collision detection unit can add a probabilistic risk assessment method to calculate the risk values ​​of different collision avoidance strategies. The decision generation unit can add a decision verification step to predict the execution effect of commands through rapid simulation. In specific implementations, the integration of the navigation decision module with other modules of the system is achieved through ROS middleware. The navigation decision module subscribes to dynamic obstacle topics published by the environment perception module and smooth trajectory topics published by the path planning module. The decision topics published by the navigation decision module are subscribed to by the motion control module and the path planning module. The data processing of the obstacle prediction unit adopts a pipelined architecture to process the prediction tasks of multiple obstacles in parallel. The computational load of the collision detection unit is dynamically adjusted, and the computational accuracy is reduced to ensure real-time performance when system resources are scarce. The command issuance mechanism of the decision generation unit includes priority scheduling, where high-priority commands can interrupt low-priority commands. It can be understood that the performance of the navigation decision module depends on the tracking accuracy of the environment perception module and the trajectory quality of the path planning module. The prediction accuracy of the obstacle prediction unit affects the reliability of collision detection. The computational efficiency of the collision detection unit determines the system response speed. The rationality of the decision generation unit's decisions is related to the overall navigation safety. In practical implementation, the obstacle prediction unit is implemented using the C++ programming language and utilizes the Eigen library for matrix operations. The collision detection unit integrates the CGAL library to handle computational geometry problems. The state machine of the decision generation unit is built based on the Boost state graph library. The navigation decision module's runtime log records prediction results, detection data, and decision-making processes in detail for offline analysis and debugging. Optionally, the navigation decision module can be configured with a decision result visualization interface to display the obstacle prediction area, smooth trajectory, and overlap degree change curves in real time, assisting developers in verifying the algorithm's correctness. In practical implementation, the Kalman filter parameters of the obstacle prediction unit are calibrated through extensive experimental data. The process noise covariance matrix reflects the uncertainty of the motion model, and the observation noise covariance matrix matches the sensor error characteristics. The overlap degree calculation of the collision detection unit supports multi-resolution modes, and the time discretization granularity can be adjusted as needed. The threshold settings of the decision generation unit have been optimized through simulation testing to balance safety and efficiency. In some embodiments, the obstacle prediction unit can add an obstacle interaction model to consider the mutual influence between multiple obstacles, the conflict detection unit can expand the spatiotemporal corridor check to ensure the safety of the entire trajectory, and the decision generation unit can achieve multi-objective optimization while considering collision avoidance, energy consumption and comfort indicators.In practical implementation, the navigation decision module's anomaly handling mechanism monitors the validity of input data. When anomalies are detected, it switches to a degradation mode, employing conservative prediction and decision-making strategies. The obstacle prediction unit performs data integrity checks, verifying the continuity and rationality of the input trajectory. The conflict detection unit includes numerical stability protection to prevent floating-point calculation errors. The decision generation unit has a timeout handling function, outputting a safe stop command when the decision-making process times out. It is understandable that the navigation decision module has strict real-time requirements; the computational complexity of the obstacle prediction unit must match the control cycle; the algorithm selection for the conflict detection unit considers computational resource limitations; and the response latency of the decision generation unit must meet the demands of the dynamic environment.

[0079] See Figure 4 This diagram demonstrates the trajectory planning and decision-making process of an intelligent navigation system in a dynamic environment. The blue trajectory line represents the robot's planned path, while the red elliptical area represents the predicted area occupied by dynamic obstacles, the size of which reflects the degree of uncertainty in the predicted position. The system models the obstacle's motion state using a Kalman filter, predicting its position distribution within a future time window. Different colored markers represent real-time decisions made by the system based on overlap: green points correspond to normal driving, orange points indicate the system detects a potential conflict and initiates a deceleration and avoidance strategy, and red points trigger an emergency replanning command. The purple dashed line marks the decision threshold boundary; when the overlap between the trajectory and the predicted obstacle area exceeds a set threshold, the system automatically adjusts its navigation strategy. This multi-layered decision-making mechanism ensures the robot's safe navigation in complex environments while maintaining travel efficiency. The smoothness of the trajectory and the distribution of decision points reflect the system's effective response to dynamic obstacles and its collision avoidance capabilities.

[0080] Example 4: The communication management module interacts with the central scheduling server via a wireless network, receives task target points and map update information, and uploads the robot's real-time status data and task execution progress. The communication management module is configured to switch to a distributed communication mode based on edge computing nodes when the wireless signal strength is lower than a preset threshold, maintaining basic navigation functions. The energy management module monitors the robot's remaining battery power and instantaneous power consumption, and estimates the energy consumption of the entire task based on the length and complexity of the smooth trajectory output by the path planning module. The energy management module is configured to send a charging request signal to the path planning module and plan a path to the nearest charging station when the remaining battery power is lower than a preset proportion of the power required to complete the current task. In its implementation, the communication management module adopts a dual-mode communication architecture. The primary communication link is based on a Wi-Fi network using the IEEE 802.11ac standard, while the backup communication link uses an LTE Cat-M1 cellular network. The software implementation of the communication management module includes a communication state machine, a data compression unit, and a link quality monitoring unit. The communication state machine manages the switching logic between different communication modes. The data compression unit performs lossless compression on transmitted data to reduce bandwidth usage. The link quality monitoring unit evaluates the wireless channel quality in real time, with a preset signal strength threshold of -85dBm. When the signal strength remains below this threshold for 3 consecutive seconds, a communication mode switch is triggered. The hardware foundation of the energy management module is an intelligent battery management system, including a coulomb counter chip, voltage and current sampling circuits, and a temperature sensor. The software algorithm of the energy management module implements a power consumption prediction model and charging decision logic. The power consumption prediction model is based on machine learning methods, considering factors such as the length of the smooth trajectory, terrain slope, and expected speed. The preset charging decision ratio is set to 20%. When the remaining battery power is detected to be less than 20% of the estimated power consumption to complete the current task, the energy management module sends a charging request to the path planning module via ROS service calls.

[0081] The communication management module employs a customized lightweight communication protocol for data interaction. The data packet structure includes a frame header, message type, data payload, and cyclic redundancy check (CRC) code. A heartbeat mechanism is established between the communication management module and the central scheduling server, exchanging heartbeat packets every 5 seconds to confirm the connection status. In distributed communication mode, the communication management module communicates with edge computing nodes via a point-to-point direct connection, forming a self-organizing network using the IEEE 802.11s mesh network protocol. The energy management module uses a piecewise linear regression model for energy consumption prediction. Model parameters are obtained through training with historical operating data. The energy management module maintains a charging station location database, containing the coordinates, interface type, and availability information of charging stations. When a charging request is triggered, the energy management module calculates the optimal charging station selection based on the current location and remaining battery power. In implementation, the communication management module uses a priority scheduling strategy for message queue management, with navigation control commands having the highest priority, status data having medium priority, and log information having low priority. Data encryption in the communication management module uses the AES-256 algorithm to ensure communication security. The energy management module's battery health monitoring includes cycle count statistics and capacity decay assessment. The energy management module's temperature protection function automatically reduces the maximum discharge current when the battery temperature exceeds 45 degrees Celsius.

[0082] The communication management module achieves seamless link switching, pre-establishing backup connections when a signal strength decline is detected. It also supports a retransmission mechanism, automatically retransmitting important data packets without acknowledgment. The energy management module uses a 10Hz power consumption sampling frequency, with the sampled data filtered by a moving average before being used for power calculation. Its charging strategy supports emergency charging and planned charging modes. Emergency charging is executed immediately when power is severely insufficient, while planned charging is performed during task intervals. It is understandable that the reliability of the communication management module directly affects the collaborative efficiency between the robot and the control center, while the accuracy of the energy management module relates to the robot's continuous working capability. The collaborative work of these two modules ensures the robot's long-term autonomous operation in complex environments.

[0083] In practical implementation, the network status monitoring of the communication management module includes multiple quality indicators, such as network latency, packet loss rate, and link stability, in addition to signal strength. The distributed communication mode of the communication management module supports multi-hop relay functionality, allowing the robot to relay data through adjacent nodes. The energy management module's online energy consumption model learning function can dynamically adjust prediction parameters based on actual power consumption. The charging station database of the energy management module supports remote updates, receiving server-pushed updates when the charging station status changes. The resource management of the communication management module adopts a dynamic bandwidth allocation mechanism, adjusting data transmission volume according to current task requirements. The communication log of the communication management module records all connection events and data transmission statistics. The battery protection logic of the energy management module includes over-discharge prevention functionality, forcing the robot into a sleep state when the battery level falls below a safe threshold. The power consumption analysis report of the energy management module generates a detailed task energy consumption analysis. Referring to Table 1, the signal strength monitoring of the communication management module uses a sliding window averaging algorithm to avoid erroneous switching due to instantaneous fluctuations. The switching decision of the communication management module considers multiple factors.

[0084] Table 1: Correspondence between Communication Quality Assessment Parameters and Mode Switching Decisions

[0085] Monitoring parameters Threshold conditions Weighting coefficient Trigger Action Signal Strength RSSI <-85dBm for 3 seconds 0.4 Startup mode switching Network latency >500ms lasts for 5 seconds 0.3 Startup mode switching Data packet loss rate >10% lasts for 5 seconds 0.2 Startup mode switching Link stability index <0.6 0.1 Startup mode switching

[0086] The comprehensive evaluation value of the communication management module is calculated using weighted averages. A communication mode switch is triggered when the comprehensive evaluation value falls below 0.5. The charging decision algorithm of the energy management module comprehensively considers factors such as remaining battery power, task urgency, and distance to charging stations. The path planning request of the energy management module includes current battery status and expected power consumption information. In some embodiments, the communication management module can be extended to support 5G network slicing technology, providing differentiated quality of service guarantees for different services. The energy management module can integrate solar charging functionality to extend outdoor working endurance. In specific implementations, the anomaly handling mechanism of the communication management module detects network interface faults and switches to degraded mode when a hardware fault is detected. The fault detection of the energy management module includes sensor anomaly diagnosis and circuit fault detection to ensure the reliability of monitoring data.

[0087] Optionally, the communication management module can add a data caching mechanism to temporarily store critical data during network interruptions and transmit it after recovery. The energy management module can add a power consumption warning function to notify the dispatch system in advance before the battery level reaches a threshold. In specific implementation, the communication quality report of the communication management module is sent to the central dispatch server periodically for network optimization and analysis. The battery performance data of the energy management module is recorded for a complete charge-discharge cycle for health status assessment. The time synchronization function of the communication management module uses a network time protocol to maintain clock consistency with the server, while the timer of the energy management module runs independently to avoid system clock drift affecting power calculation. The multi-link aggregation function of the communication management module can use multiple network interfaces simultaneously to improve throughput when the signal is good. The power consumption optimization function of the energy management module automatically reduces unnecessary system load when low battery is detected. The implementation of the communication management module is based on the Linux network stack, uses the TCP / IP protocol stack for reliable transmission, and uses the UDP protocol for real-time data transmission to reduce latency. The configuration parameters of the communication management module are set through a configuration file and support dynamic adjustment at runtime. The energy management module's hardware platform is based on an ARM Cortex-M series microcontroller, enabling low-power operation. The software components of the energy management module communicate with the main system via a serial peripheral interface. In practice, the communication management module's resource monitoring function tracks network bandwidth usage to prevent network congestion, and its security authentication uses two-way certificate verification to ensure connection legitimacy. The energy management module's calibration function periodically calibrates the accuracy of power metering, and its self-protection function automatically disconnects the discharge circuit upon detecting abnormally high current.

[0088] Optionally, the communication management module can support software-defined networking technology to enable flexible scheduling of network resources, and the energy management module can add interfaces for hybrid power supply systems such as fuel cells to expand energy supply methods. In specific implementation, the performance indicators of the communication management module include connection establishment time, data transmission rate, and handover interruption duration. The stability test of the communication management module includes long-term continuous operation tests, the accuracy verification of the energy management module is carried out through standard load tests, and the reliability test of the energy management module simulates various abnormal power supply conditions. The collaborative operation of the communication management module and the energy management module is realized through the system event bus. When communication is interrupted, the energy management module adjusts its power consumption strategy; when power is insufficient, the communication management module reduces data transmission. This collaborative mechanism ensures the basic operational capability of the system under adverse conditions.

[0089] See Figure 5This diagram illustrates the collaborative operation of the communication management module and the energy management module. The blue signal strength curve shows the quality changes of the wireless communication link. When the signal strength falls below a preset threshold, the system automatically switches to distributed communication mode to maintain basic navigation functions. The green network latency curve reflects the real-time performance of data transmission; latency peaks may occur during signal switching or network congestion. The red battery power curve shows the energy consumption trend. The system estimates energy consumption needs based on path planning and task complexity, and automatically triggers a charging request when the remaining battery power is lower than a preset percentage required to complete the current task. The orange power consumption curve shows the energy consumption characteristics of the system under different operating conditions, with peaks typically occurring during acceleration, turning, or communication mode switching. The vertical purple dashed line marks critical moments of communication mode switching, while the brown dashed line indicates the trigger point for charging requests. This collaborative monitoring mechanism ensures the robot's continuous and reliable operation under conditions of wireless signal fluctuations and energy constraints, optimizing system performance by dynamically adjusting communication strategies and energy allocation.

[0090] Example 5: The fault diagnosis module is used to periodically detect the operating status of the environment perception module, localization and mapping module, path planning module, motion control module, and navigation decision module. The fault diagnosis module is configured to record a fault code and trigger a safety protection mechanism when any abnormality is detected in any module, causing the robot to decelerate and stop or return to the starting point along the original path. The data recording module is used to continuously store the raw data collected by the environment perception module, the map data generated by the localization and mapping module, the path data calculated by the path planning module, the execution data of the motion control module, and the decision data of the navigation decision module. The data recording module is configured to establish a complete navigation process dataset in chronological order for subsequent offline analysis and algorithm optimization. In its implementation, the fault diagnosis module's software architecture adopts a distributed monitoring agent design. Each monitored module embeds a lightweight diagnostic agent, which periodically collects internal status indicators and exchanges data via shared memory. The main control unit of the fault diagnosis module polls each diagnostic agent at a frequency of 1 Hz. Status indicators include CPU utilization, memory usage, message publishing frequency, data validity, and algorithm convergence status. The fault diagnosis module's anomaly judgment rules are based on multi-threshold judgments. For example, if the LiDAR unit fails to publish point cloud data for five consecutive cycles, the environmental perception module is deemed abnormal; if the pose covariance determinant value of the localization and mapping module exceeds a threshold of 10, the localization is deemed abnormal. The data recording module's storage architecture adopts a multi-level caching design. Raw data is first stored in a circular memory buffer, then, after timestamp alignment and format standardization, is written in batches to the solid-state drive. The data recording module defines independent storage partitions for each data type. Point cloud data is compressed and stored in PCD format, image data uses JPEG2000 lossy compression, and navigation decision data uses Protocol Buffers serialization format.

[0091] The fault diagnosis module's safety protection mechanism is implemented using a tiered response strategy. Level 1 response addresses minor anomalies such as momentary data loss, triggering data retransmission and algorithm reinitialization. Level 2 response addresses persistent anomalies such as sensor failure, triggering deceleration and spare parts switching. Level 3 response addresses severe anomalies such as control system failure, immediately triggering an emergency stop and reporting error codes. The data logging module's dataset organization structure is indexed by task number and timestamp. Each navigation task generates an independent data packet containing a metadata description file and multiple data stream files. In implementation, the fault code encoding of the fault diagnosis module uses 16-bit binary code: the high 4 bits represent the fault module identifier, the middle 8 bits represent the specific fault type, and the low 4 bits represent the fault severity level. In addition to being stored in local flash memory, fault records from the fault diagnosis module are also uploaded to the central dispatch server in real time via the communication management module. The data compression algorithm of the data logging module adaptively selects based on data type. Point cloud data uses octree encoding compression, while path data uses differential encoding to reduce storage space. The data logging module's storage monitoring function monitors the remaining storage capacity in real time, automatically initiating an old data cleanup strategy when the capacity falls below a threshold. The periodic testing process of the fault diagnosis module includes self-testing and mutual testing procedures. The self-testing procedure is executed by the internal diagnostic agent of each module to check the integrity of its own functions. The mutual testing procedure is executed by the main control unit of the fault diagnosis module to verify data consistency between modules. The time synchronization mechanism of the data recording module adopts a high-precision network time protocol, and all data records are accompanied by microsecond-level timestamps. The data integrity of the data recording module is verified using SHA-256 hash values. It can be understood that the real-time requirements of the fault diagnosis module determine the system's response speed to faults, while the reliability of the data recording module directly affects the quality of subsequent data analysis. The collaborative work of these two modules provides a foundation for fault tracing and performance optimization.

[0092] In practical implementation, the sensor health monitoring of the fault diagnosis module includes LiDAR light intensity attenuation detection, camera focal blur detection, and IMU zero-bias stability monitoring. The algorithm status monitoring of the fault diagnosis module includes feature point tracking count statistics for the localization and mapping module, path search time monitoring for the path planning module, and trajectory tracking error analysis for the motion control module. The data recording module's storage strategy supports multiple storage modes: normal mode records the complete dataset, energy-saving mode records only key indicators, and debugging mode records detailed intermediate results. The fault prediction function of the fault diagnosis module is based on time series analysis, identifying potential faults in advance by monitoring the changing trends of performance indicators. The data retrieval interface of the data recording module supports multi-condition queries by time range, data type, and spatial location. The data export function of the data recording module supports multiple formats including CSV, JSON, and ROSBag.

[0093] In some embodiments, the fault diagnosis module can integrate a machine learning classifier to improve fault identification accuracy through training on historical fault data. The data recording module can add real-time data analysis capabilities to generate online navigation performance reports. In specific implementations, test cases for the fault diagnosis module include simulating LiDAR data loss scenarios to verify the abnormal handling process of the environmental perception module, and simulating positioning jump scenarios to verify the fault recovery mechanism of the positioning and mapping module. The stress test for the data recording module includes continuously recording data for 72 hours to verify the stability of the storage system. The emergency stop interface between the fault diagnosis module and the motion control module uses a direct hardware connection, bypassing the software middleware layer to ensure real-time response. The important data marking function of the data recording module allows manual marking of special event points.

[0094] The fault injection testing platform of the fault diagnosis module supports simulating various abnormal conditions, such as artificially created communication delays, data anomalies, and hardware failures, to verify the completeness of the diagnostic system. The data playback tool of the data logging module supports accelerated playback and paused analysis, facilitating developers to reproduce problem scenarios. In practical implementation, the diagnostic rule base of the fault diagnosis module supports online updates, allowing modification of detection thresholds and rules without restarting the system. The statistical function of the fault diagnosis module records the frequency and distribution patterns of various faults. The automatic archiving function of the data logging module periodically transfers historical data to a cloud server, freeing up local storage space, and the privacy protection function of the data logging module encrypts and stores sensitive data.

[0095] Optionally, the fault diagnosis module can be equipped with an audible alarm to alert the user when a serious fault is detected. The data logging module can integrate a data visualization component to display key data curves in real time. In practical implementation, the performance optimization of the fault diagnosis module adopts a multi-resolution monitoring strategy, with high-frequency monitoring of key parameters and low-frequency sampling of secondary parameters. The log records of the fault diagnosis module contain complete fault context information. The storage encryption of the data logging module uses the AES-256 algorithm to ensure data security, and the damage detection function of the data logging module periodically scans the storage medium for bad sectors. The self-healing mechanism of the fault diagnosis module is designed with an automatic recovery process for transient faults, reducing the need for manual intervention.

[0096] Optionally, the fault diagnosis module supports remote diagnostics, allowing technical support personnel to analyze system status online. The data logging module can add data quality assessment indicators and automatically identify abnormal data segments. In practical implementation, the redundancy design of the fault diagnosis module includes dual diagnostic unit hot backup; the backup unit automatically takes over when the primary diagnostic unit fails, and the fault diagnosis module's startup self-test process verifies the integrity of all diagnostic components. The data logging module's storage load balancing function distributes data writes across multiple storage devices, improving storage system throughput. The fault diagnosis module's case learning function extracts features from historical faults to optimize diagnostic rules.

[0097] Optionally, the fault diagnosis module can be extended to support third-party module diagnostic interfaces, and the data recording module can provide a data sharing interface for other systems to call. In specific implementation, the fault diagnosis module generates a standardized format report, including the fault time, module identifier, fault type, and handling suggestions. The fault diagnosis module's early warning function provides advance warnings when indicators approach thresholds. The data recording module's data lifecycle management automatically cleans up expired data, and its version management records the history of data format changes. The collaborative work between the fault diagnosis module and the data recording module is reflected in the fact that when a fault occurs, the data recording module automatically marks the data before and after the fault time point, forming a complete fault dataset for analysis.

[0098] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0099] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An indoor / outdoor logistics robot navigation system, characterized in that, include: The environmental perception module is used to collect environmental information around the robot in real time. The environmental information includes three-dimensional point cloud data, visible light image data and raw observations from the Global Navigation Satellite System. The localization and mapping module is used to fuse the environmental information, construct a dense point cloud map containing semantic labels, and calculate the robot's six-degree-of-freedom pose based on the map. The path planning module searches for a collision-free path in the dense point cloud map based on the task target point and the six-degree-of-freedom pose. The path consists of a series of path points and their corresponding robot motion postures. A motion control module is used to convert the travel path into control commands for the drive motors and execute the commands to make the robot move along the path; The navigation decision module receives data from the environment perception module, the positioning and mapping module, the path planning module, and the motion control module. By evaluating the spatiotemporal relationship between the motion trajectory of dynamic obstacles in the environment and the passage path, it generates a decision signal containing speed adjustment instructions and local replanning instructions. The navigation decision module performs the following operations: predicts the area occupied by the dynamic obstacles in the environment in the dense point cloud map at future times based on their movement trajectories; calculates the degree of overlap between the travel path and the occupied area; and when the degree of overlap exceeds a preset threshold, generates a speed adjustment command to reduce the robot's movement speed or generates a local replanning command to replan the local path.

2. The indoor / outdoor logistics robot navigation system according to claim 1, characterized in that, The environment sensing module includes: A lidar unit is used to emit a laser beam and receive echo signals to generate the three-dimensional point cloud data. A visual sensor unit is used to acquire the visible light image data and extract edge features and texture features from the scene. The multi-band satellite signal receiving unit is used to receive the raw observation values ​​of the global navigation satellite system and calculate the robot's latitude and longitude coordinates and altitude.

3. The indoor / outdoor logistics robot navigation system according to claim 2, characterized in that, The positioning and mapping module includes: The point cloud registration unit is used to iteratively match the three-dimensional point cloud data collected at different times with the constructed dense point cloud map to calculate the robot's position change. The visual inertial odometry unit is used to combine the visible light image data and the angular velocity and linear acceleration data of the inertial measurement unit to calculate the continuous motion trajectory of the robot through a nonlinear optimization method. The compact navigation unit is used to fuse the raw observations of the global navigation satellite system with the output of the visual inertial odometry unit using Kalman filtering to generate the six-degree-of-freedom pose.

4. The indoor / outdoor logistics robot navigation system according to claim 3, characterized in that, The path planning module includes: The global planning unit is used to calculate an initial global path from the robot's current six-degree-of-freedom pose to the task target point in the dense point cloud map using a graph search-based path planning algorithm. The local planning unit is used to locally adjust the initial global path based on the temporary obstacle information detected in real time by the environment perception module, and to generate a smooth trajectory that satisfies the robot's kinematic constraints.

5. The indoor / outdoor logistics robot navigation system according to claim 4, characterized in that, The motion control module includes: The trajectory tracking unit is used to discretize the smooth trajectory into a series of pose points on a time scale, and calculate the target rotation speed and steering angle of the robot drive wheel at each moment; The motor drive unit is used to generate a corresponding motor drive voltage signal based on the target rotation speed and steering angle using pulse width modulation technology. The motion feedback unit is used to collect the actual speed and steering angle of the drive motor and compare them with the target value to form a closed-loop control.

6. The indoor / outdoor logistics robot navigation system according to claim 5, characterized in that, The navigation decision module includes: The obstacle prediction unit is used to model the motion state of the dynamic obstacles in the environment and use a linear dynamic system to predict their position and velocity distribution in the next few seconds. The collision detection unit is used to perform a spatiotemporal synchronous comparison between the smooth trajectory and the position and velocity distribution, and calculate the degree of overlap between the travel path and the occupied area. The decision generation unit is used to selectively generate the speed adjustment instruction or the local replanning instruction based on the degree of overlap.

7. The indoor / outdoor logistics robot navigation system according to claim 6, characterized in that, Also includes: The communication management module is used to interact with the central dispatch server via a wireless network, receive the task target point and map update information, and upload the robot's real-time status data and task execution progress. The communication management module is configured to switch to a distributed communication mode based on edge computing nodes when the wireless signal strength is lower than a preset threshold, so as to maintain basic navigation functions.

8. The indoor / outdoor logistics robot navigation system according to claim 7, characterized in that, Also includes: The energy management module is used to monitor the remaining power and instantaneous power consumption of the robot's battery, and to estimate the energy consumption of the entire task based on the length and complexity of the smooth trajectory. The energy management module is configured to send a charging request signal to the path planning module and plan a path to the nearest charging station when the remaining power is lower than a preset percentage of the power required to complete the current task.

9. The indoor / outdoor logistics robot navigation system according to claim 8, characterized in that, Also includes: The fault diagnosis module is used to periodically detect the operating status of the environmental perception module, the positioning and mapping module, the path planning module, the motion control module, and the navigation decision module; The fault diagnosis module is configured to record a fault code and trigger a safety protection mechanism when any module malfunctions, causing the robot to decelerate and stop or return to the starting point along the original path.

10. The indoor / outdoor logistics robot navigation system according to claim 9, characterized in that, Also includes: The data recording module is used to continuously store the raw data collected by the environmental perception module, the map data generated by the positioning and mapping module, the path data calculated by the path planning module, the execution data of the motion control module, and the decision data of the navigation decision module. The data recording module is configured to build a complete navigation process dataset in chronological order for subsequent offline analysis and algorithm optimization.

Images