Robot obstacle avoidance control method and system based on spatiotemporal trajectory prediction

Abstract

The application discloses a robot obstacle avoidance control method and system based on space-time trajectory prediction. The application realizes a closed loop from perception to control through the following steps: generating a reliable weight through multi-source perception cross verification, constructing a dynamic obstacle space-time trajectory probability heat map, filtering high-risk nodes by upgrading the path space-time dimension, and recalibrating the timestamp after path optimization. The application converts the uncertainty of future motion of obstacles into an indexable space-time probability distribution, enabling the robot to predict collision risks in advance and actively plan detours. The application solves the problems of traditional reactive obstacle avoidance path oscillation, planning and execution disconnection, and significantly improves the safety, smoothness and efficiency of obstacle avoidance in dynamic environments.

Description

Technical Field

[0001] This application relates to the field of robotics technology, and in particular to a robot obstacle avoidance control method and system based on spatiotemporal trajectory prediction. Background Technology

[0002] With the rapid development of intelligent manufacturing and smart logistics, the demand for collaborative operations of multiple robots in the same scenario is increasing. In dynamic environments, robots need to perceive the movement status of surrounding pedestrians, other robots, and other dynamic obstacles in real time, and adjust their own paths accordingly to complete navigation tasks safely and efficiently.

[0003] Existing robot obstacle avoidance technologies are mainly divided into two categories: The first type is reactive obstacle avoidance. This type of method relies on sensors (such as LiDAR and depth cameras) on the robot to detect obstacles in real time, and triggers an obstacle avoidance action when an obstacle is detected to be within a safe distance.

[0004] The second category is obstacle avoidance methods based on trajectory prediction. These methods use historical trajectory data to predict the future movement of dynamic obstacles and incorporate the prediction results into the path planning stage. This type of method possesses a certain degree of proactive obstacle avoidance capability.

[0005] However, both of these methods have shortcomings in practical deployment. The first type, reactive obstacle avoidance, triggers obstacle avoidance actions solely based on current sensor information, lacking the ability to predict future dynamic environmental changes. In scenarios with dense obstacles or high-speed movement, the robot frequently experiences localized oscillations or sudden stops because it cannot plan detours in advance, significantly reducing path smoothness and passage efficiency. The second type, trajectory prediction-based obstacle avoidance, while possessing some foresight, typically fails to fully consider the underlying kinematic constraints of dynamic obstacles during their movement. This results in some predicted trajectories being physically difficult to execute, leading to motion deviations in obstacle avoidance paths generated based on such trajectories. Furthermore, existing methods do not effectively integrate the time dimension into the collision risk assessment of path points during the path planning stage, lacking the ability to finely retrieve and filter collision probabilities in the continuous spatiotemporal domain. This makes the obstacle avoidance decision insensitive to the dynamic changes in the future spatiotemporal distribution of obstacles. Summary of the Invention

[0006] The purpose of this application is to provide a robot obstacle avoidance control method and system based on spatiotemporal trajectory prediction, which solves the problem of unreliable obstacle avoidance execution caused by time information mismatch after path change.

[0007] To address the aforementioned technical problems, this application provides the following technical solutions: The first aspect of this application provides a robot obstacle avoidance control method based on spatiotemporal trajectory prediction, the method comprising: The system acquires environmental perception data and body state data collected by multiple robots, and performs spatiotemporal alignment under a unified coordinate system to construct a spatiotemporal database; it then performs cross-consistency verification based on spatial constraints on the environmental perception data across terminals to determine the spatiotemporal dynamic reliability weight of each robot's perception data. Based on the environmental perception data, candidate motion trajectories of dynamic obstacles are predicted in future time periods; and the distribution probability of the candidate motion trajectories in spatiotemporal space is calculated to construct a spatiotemporal trajectory probability heatmap of dynamic obstacles. The robot's initial planned path points are synchronously assigned timestamps, and the initial planned path points carrying timestamps are expanded into a spatiotemporal key point sequence containing spatial coordinates and time; the spatiotemporal key point sequence is mapped to the spatiotemporal trajectory probability heatmap. Based on the collision probability of each key point in the spatiotemporal key point sequence in the spatiotemporal trajectory probability heatmap, unsafe nodes with collision probabilities exceeding a preset threshold are replaced with local obstacle avoidance alternative paths to generate optimized spatiotemporal obstacle avoidance paths. Based on the actual length of the optimized spatiotemporal obstacle avoidance path, the timestamps of each spatiotemporal key point on the path are recalibrated, and the robot is controlled to perform dynamic obstacle avoidance actions according to the optimized spatiotemporal obstacle avoidance path.

[0008] A second aspect of this application provides a robot obstacle avoidance control system based on spatiotemporal trajectory prediction, the system comprising: The database construction module is used to acquire environmental perception data and body state data collected by multiple robots, and to construct a spatiotemporal database by spatiotemporal alignment under a unified coordinate system; and to perform cross-consistency verification based on spatial constraints on the environmental perception data across terminals to determine the spatiotemporal dynamic reliability weight of each robot's perception data. The heatmap construction module is used to predict candidate motion trajectories of dynamic obstacles in future time periods based on the environmental perception data; and to calculate the distribution probability of the candidate motion trajectories in spatiotemporal space to construct a spatiotemporal trajectory probability heatmap of dynamic obstacles. The mapping module is used to synchronously assign timestamps to the robot's initial planned path points, expand the initial planned path points carrying timestamps into a spatiotemporal key point sequence containing spatial coordinates and time, and map the spatiotemporal key point sequence to the spatiotemporal trajectory probability heatmap. The spatiotemporal obstacle avoidance path generation module is used to replace unsafe nodes with local obstacle avoidance alternative paths with collision probabilities exceeding a preset threshold in the spatiotemporal trajectory probability heatmap of each key point in the spatiotemporal key point sequence, thereby generating an optimized spatiotemporal obstacle avoidance path. The control module is used to recalibrate the timestamps of each spatiotemporal key point on the optimized spatiotemporal obstacle avoidance path based on the actual length of the path, and control the robot to perform dynamic obstacle avoidance actions according to the optimized spatiotemporal obstacle avoidance path.

[0009] Compared to existing technologies, the robot obstacle avoidance control method and system based on spatiotemporal trajectory prediction provided in this application integrates the robot chassis kinematic constraints into the spatial geometric path and uses high-order continuous velocity planning and time parameterized mapping to accurately solve the absolute timestamp, seamlessly upgrading the traditional pure spatial path to a spatiotemporal key point sequence, overcoming the defects of velocity abrupt change and spatiotemporal misalignment. At the same time, a time-domain interpolation mechanism based on the time offset ratio is introduced to eliminate the probability step error caused by the heat map layer, realizing a high-precision quantitative assessment of collision risk in the continuous time domain. By screening high-risk nodes and locating the front and rear safety anchor points, seamless reconnection and replacement are performed in the spatiotemporal trajectory probability heat map using local alternative paths, taking into account both the computational efficiency of local planning and the safety and executability of the global path, greatly improving the smoothness, passage efficiency and control robustness of the robot's autonomous obstacle avoidance in complex dynamic scenarios. Attached Figure Description

[0010] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, with the same or corresponding reference numerals denoteing the same or corresponding parts, wherein: Figure 1 A flowchart illustrating a robot obstacle avoidance control method based on spatiotemporal trajectory prediction is shown schematically. Figure 2 A schematic diagram of a robot obstacle avoidance control system based on spatiotemporal trajectory prediction is shown. Detailed Implementation

[0011] Exemplary embodiments of this application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.

[0012] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains.

[0013] The methods described in the embodiments of this application will be explained in detail below.

[0014] This embodiment provides a robot obstacle avoidance control method based on spatiotemporal trajectory prediction. This method can be applied to cross-regional navigation tasks of autonomous mobile robots in scenarios such as large factories, hospitals, and warehousing and logistics centers. In this embodiment, the algorithm processing task can be flexibly deployed on a cloud server or an edge computing terminal mounted on the robot body, depending on the actual needs of the scenario. The task is completed collaboratively by the cloud server and the robot terminal, or by the robot terminals themselves, through a wireless communication network.

[0015] In the centralized cloud deployment mode, each robot terminal is responsible for collecting environmental perception data and body status data in real time and uploading them to the cloud server. The cloud server centrally executes all planning and computing tasks, including spatiotemporal alignment and cross-consistency verification, observation feature extraction, multimodal trajectory prediction and spatiotemporal trajectory probability heatmap construction, single-machine spatiotemporal path dimensionality upgrade and collision probability filtering, multi-machine spatiotemporal conflict detection and arbitration scheduling, and distributes the generated optimized spatiotemporal obstacle avoidance path to the corresponding robot terminal.

[0016] In the edge distributed deployment mode, each robot's edge computing terminal independently performs preprocessing and spatiotemporal alignment of its own perception data, shares the aligned data through the inter-machine communication network, and performs distributed cross-consistency verification to determine the trust weight of each terminal's data. Subsequently, each terminal autonomously completes observation feature extraction, trajectory prediction and heat map construction based on the trust weight, independently performs path dimensionality upgrading and insecure node filtering locally, and performs distributed spatiotemporal conflict detection and autonomous negotiation scheduling through inter-machine broadcast.

[0017] Regardless of the deployment mode, after acquiring the optimized spatiotemporal obstacle avoidance path, the robot terminal uses an onboard trajectory tracking controller to generate linear and angular velocity control commands based on local perception information, driving the robot's underlying motion chassis to complete dynamic obstacle avoidance actions. Cloud servers include, but are not limited to, cloud inference nodes, GPU computing clusters, or edge cloud servers; edge computing terminals include, but are not limited to, embedded computing platforms, mobile GPU processors, or FPGA inference accelerator cards. The overall flow of the method in this embodiment is as follows: Figure 1 As shown, it includes the following steps: Step S110: Acquire environmental perception data and body state data collected by multiple robots, and construct a spatiotemporal database by spatiotemporal alignment under a unified coordinate system; perform cross-consistency verification based on spatial constraints on the environmental perception data across terminals to determine the spatiotemporal dynamic reliability weight of each robot's perception data.

[0018] This step completes the spatiotemporal unification and quality assessment of multi-robot perception data. First, spatial coordinate transformation and timestamp synchronization are performed on the environmental perception data and body state data collected by each robot. The spatiotemporal positions of dynamic obstacles and the spatiotemporal poses of each robot are associated and stored in a unified spatiotemporal coordinate system, constructing a spatiotemporal database. Then, cross-consistency verification based on spatial constraints is performed on the cross-terminal environmental perception data—extracting the spatially overlapping observation areas of each robot in the spatiotemporal database, comparing the consistency of cross-terminal observation data within the overlapping areas, eliminating abnormal observation data with deviations exceeding a preset threshold, and determining the spatiotemporal dynamic reliability weight of each robot's perception data based on the degree of consistency of the retained observation data from each terminal. This weight will serve as the weighting basis for data fusion and feature extraction in subsequent step S120.

[0019] In a specific implementation, the step of constructing a spatiotemporal database through spatiotemporal alignment can be achieved by the following steps: Spatial coordinate transformation and timestamp synchronization are performed on environmental perception data and body state data respectively. The spatiotemporal positions of each dynamic obstacle and the spatiotemporal poses of each robot are associated and stored in a unified spatiotemporal coordinate system to construct a spatiotemporal database.

[0020] Specifically, this step first performs spatial coordinate transformation and timestamp synchronization on the environmental perception data and body state data collected by each robot, unifying the multi-source heterogeneous data under the same spatiotemporal reference. Regarding spatial coordinate transformation, the raw data collected by the sensors (such as LiDAR and depth cameras) and UWB positioning tags on each robot are all based on each robot's local coordinate system. This step transforms the environmental perception data (such as obstacle point clouds and target detection boxes) in the local coordinate system to the global world coordinate system using the real-time pose of each robot in the global world coordinate system (provided by UWB base station coordinate matching or synchronous localization and mapping algorithms), obtaining the global position of each dynamic obstacle under the unified spatial reference. Simultaneously, the body state data of each robot (including position, velocity, orientation, etc.) is also uniformly transformed to this global world coordinate system, obtaining the global spatiotemporal pose of each robot.

[0021] Regarding timestamp synchronization, there may be clock discrepancies in the sensor data acquisition times of different robots. This step uses a unified system clock source (such as a cloud server clock or GPS timing signal) as a reference to align the timestamps of the data frames uploaded by each terminal. Through linear interpolation or nearest neighbor matching, the asynchronously acquired data is synchronized to a unified timeline to ensure that the observation data of different robots at the same time have temporal consistency.

[0022] After completing spatial coordinate transformation and timestamp synchronization, this step associates and stores the global spatiotemporal positions of each dynamic obstacle with the global spatiotemporal poses of each robot within the same spatiotemporal coordinate system, constructing a spatiotemporal database. Each data entry in this database includes: a timestamp, a unique identifier for the obstacle / robot, global spatial coordinates, and motion state attributes (velocity, acceleration, orientation, etc.). Through the joint index of timestamps and spatial coordinates, the spatial distribution relationship between each obstacle and each robot can be quickly retrieved at any given time, providing a structured data foundation for subsequent cross-consistency verification and observation feature extraction.

[0023] In a specific implementation, cross-consistency verification of environmental perception data across terminals based on spatial constraints can be achieved through the following steps: Extract the spatially overlapping observation areas of each robot in the spatiotemporal database, and perform consistency comparison on the observation data across terminals within the spatially overlapping areas; remove abnormal observation data with deviations exceeding a preset threshold, and determine the spatiotemporal dynamic reliability weight based on the consistency of the retained observation data from each terminal.

[0024] Specifically, after completing the spatiotemporal database construction, this step further utilizes redundant observations of the same spatial region by multiple robots to assess the reliability of the perception data from each robot. Since the robots are deployed distributed throughout the environment, the fields of view of their onboard sensors inevitably overlap spatially, and the same dynamic obstacle may be observed by multiple robots simultaneously. Therefore, cross-consistency verification of this spatial redundancy is necessary.

[0025] For spatial overlap region extraction, this step calculates the perception coverage area of ​​each robot at each time stamp based on the aligned global spatiotemporal poses of each robot and the field-of-view parameters of each sensor (such as the scanning angle and maximum detection distance of the LiDAR, the field of view angle of the depth camera, etc.) in the spatiotemporal database. When the perception coverage areas of two or more robots overlap, the overlapping area is marked as the spatial overlap observation area. Dynamic obstacles within this overlap area meet the conditions for redundant observation by multiple robots.

[0026] For cross-robot consistency comparison, for each dynamic obstacle within the spatially overlapping region, the observation descriptions of each robot regarding the obstacle (including spatial position, geometric dimensions, movement speed, and other attributes) are extracted from the spatiotemporal database. Using the observation of one robot as a reference benchmark, the spatial position deviation between the observations of other robots and the reference benchmark is calculated. If the deviation is within a reasonable range of sensor measurement noise (i.e., does not exceed a preset spatial threshold), the observations of the two robots are determined to be consistent; otherwise, they are marked as inconsistent. This comparison process is performed pairwise among all robots within the overlapping region, forming a cross-robot observation consistency matrix.

[0027] Regarding outlier removal and reliability weight determination, the frequency of inconsistencies between each robot's observations and those of other robots within spatially overlapping areas is statistically analyzed. If a robot consistently exhibits significant deviations from the observations of most other robots across multiple overlapping areas and timestamps, its current perception data is deemed to be affected by factors such as occlusion, multipath reflection, or sensor malfunction, and is marked as outlier data, removed from the current processing cycle. For the retained valid observation data, a spatiotemporal dynamic reliability weight is calculated for each robot based on the proportion of consistency between its observations and those of other robots, combined with historical consistency statistics. Robots with higher consistency receive higher weights. This reliability weight is a dynamic value, updated in real-time as the perception quality of each robot changes, ensuring that subsequent data fusion and feature extraction processes are always based on the most reliable current perception data source.

[0028] Step S120: Based on environmental perception data, predict candidate motion trajectories of dynamic obstacles in future time periods; and calculate the distribution probability of candidate motion trajectories in spatiotemporal space to construct a spatiotemporal trajectory probability heatmap of dynamic obstacles.

[0029] This step completes trajectory prediction and spatiotemporal probability modeling of dynamic obstacles based on environmental perception data. First, the spatiotemporal dynamic reliability weights output in step S110 are used to adaptively extract features from the environmental perception data, generating observation features that characterize the current global state of the environment. These observation features are then injected as conditional variables into a pre-trained trajectory prediction model. Through feature mapping and decoding calculations, a sequence of positions of dynamic obstacles over multiple consecutive time steps in the future is generated as candidate trajectories. Subsequently, the predicted future time period is divided into multiple consecutive time window layers according to a preset time step. For each layer, the geometric distance from the center point of each spatial grid to the corresponding candidate trajectory position point is calculated. A preset probability decay function is used to convert the geometric distance into a probability value of each grid being occupied by an obstacle. The probability values ​​of each time window layer are then aggregated in a three-dimensional spatiotemporal coordinate system to generate a spatiotemporal trajectory probability heatmap. This heatmap provides a probabilistic basis for path collision detection in step S130.

[0030] In a specific implementation, extracting observation features based on spatiotemporal dynamic reliable weight fusion of environmental perception data can be achieved through the following two steps: By utilizing spatiotemporal dynamic reliability weights, adaptive feature extraction is performed on cross-terminal environmental perception data to obtain observational features that characterize the current global state of the environment. The observed features are injected as conditional variables into the pre-trained trajectory prediction model. Through feature mapping and decoding calculations, the position sequence of dynamic obstacles in multiple consecutive time steps in the future is generated as candidate motion trajectories.

[0031] Specifically, the steps for predicting candidate motion trajectories of dynamic obstacles can be implemented by a deep learning algorithm network through the following process: First, adaptive feature extraction based on trusted weights is performed. After acquiring cross-terminal environmental awareness data, the system inputs it into a feature extraction network (such as a lightweight convolutional network (CNN) or a multilayer perceptron (MLP)) to extract the original feature matrix. Simultaneously, the system assigns the current spatiotemporal dynamic reliability weights to each sensing terminal. Mapped to a weight coefficient matrix that matches the feature dimensions. By applying attention-based weighted regulation to the original features, interference from low-confidence data is actively suppressed, generating an observation feature vector representing the global state of the current environment. Its mathematical expression is:

[0032] in, This represents element-wise multiplication (Hadamard product) to implement feature gating at the spatial or channel level. This represents a nonlinear feature mapping function (including activation functions and pooling dimensionality reduction operations).

[0033] Subsequently, conditional trajectory prediction based on global constraints is performed. This involves refining the highly condensed and noise-filtered observation feature vectors described above. As a global conditional variable, it is injected into the pre-trained trajectory prediction model. The prediction model then uses an internal cross-attention network or feature mapping layer to process the conditional variable. The current state of the obstacle is densely fused with the current state of the obstacle; and an autoregressive decoder is used to perform multi-time-step extrapolation calculations, ultimately outputting a sequence of position coordinates of the obstacle in the future for multiple consecutive time steps, i.e., candidate motion trajectories. :

[0034] in, This refers to the current observed state of the dynamic obstacle (such as its current coordinates, speed, and yaw angle). The decoding process indicates that the observed feature vectors are used to represent the decoding process. Guided generation based on prior conditions; This represents the decoding mapping operation of the prediction model; the output... For the prediction period A sequence of two-dimensional or three-dimensional spatial coordinates at each future time step.

[0035] To achieve attribute mapping from mathematically predicted sequences to physical obstacle avoidance entities, the discrete coordinate point sequence output from the above decoding is formally defined as a candidate motion trajectory through temporal correlation, intent coverage, and geometric reconstruction. Specifically, the system first processes the ordered set of points... According to time step By performing temporal coherence connections, isolated spatial locations are transformed into coherent motion paths that characterize the future displacement trends of obstacles. Simultaneously, to address the uncertainty of dynamic obstacle behavior, the system generates multiple sets of parallel position sequences to exhaustively enumerate high-probability behavioral trends of obstacles under different multimodal intentions such as turning left, going straight, or turning right, thus forming a complete set of "candidate" trajectories. Finally, by performing spline interpolation or linear fitting on the above set of points, a motion geometric envelope containing position, velocity, and acceleration information is constructed, thereby completely defining the dynamic obstacle candidate motion trajectories for robot obstacle avoidance decision-making in the spatiotemporal dimensions.

[0036] In a specific implementation, calculating the distribution probability of candidate motion trajectories in spatiotemporal space and constructing a spatiotemporal trajectory probability heatmap of dynamic obstacles can be achieved through the following three steps: According to the preset time step, the predicted future period is divided into multiple consecutive time window layers, and each time window layer corresponds to a spatial section at a certain moment in the future. For each time window layer, calculate the geometric distance from the center point of each spatial grid in that layer to the candidate motion trajectory location point at the corresponding time. Using a preset probability decay function, the probability value of each spatial grid being occupied by dynamic obstacles is calculated based on the geometric distance. The probability values ​​under each time window layer are then aggregated in a three-dimensional spatiotemporal coordinate system to generate a spatiotemporal trajectory probability heatmap.

[0037] Specifically, the system constructs a heatmap to guide the robot to avoid obstacles by mapping discrete candidate motion trajectories into a continuous spatiotemporal probability field.

[0038] First, the spatiotemporal space is discretized and sliced. The system performs this according to a preset time step. The entire predicted future period is divided into consecutive time window layers ( Each time window layer is physically defined as a specific point in the future. The system uses a two-dimensional spatial section. Through this slicing process, the dynamic trajectory prediction problem can be transformed into a series of static probability distribution calculation problems at independent time points.

[0039] Secondly, perform a geometric distance retrieval based on the grid center point. For any given time... The system first discretizes the spatial section into a two-dimensional spatial grid composed of several square or hexagonal units. Then, the system traverses each grid within the current section. And calculate the coordinates of the center point of the grid. Corresponding time Next candidate motion trajectory prediction location point Euclidean distance between :

[0040] in, This represents the L2 norm operation. The distance... It directly reflects the geometric proximity of a specific spatial unit to the threat of dynamic obstacle movement.

[0041] Next, occupancy probability modeling is performed using a probability decay function. Considering that dynamic obstacles have realistic physical dimensions and that the prediction model suffers from increasing uncertainty errors over time, the system introduces a Gaussian kernel function as the probability decay function. This is based on the geometric distance... Calculate each spatial grid in The probability that the device is constantly occupied by obstacles. :

[0042] in, The peak probability of occupancy of the core area at the center of the obstacle; This is a decay coefficient (variance factor) used to control the amplitude of probability outward diffusion. In this embodiment, The value increases linearly with the time index k, which is used to simulate the divergence of prediction uncertainty as the time step increases.

[0043] Finally, a spatiotemporal trajectory probability heatmap is generated, and a mapping retrieval interface is established. The system arbitrates the probability values ​​of all layers in a three-dimensional spatiotemporal coordinate system according to time sequence. The process involves layering elements to generate a dynamic, time-evolving spatiotemporal trajectory probability heatmap. This heatmap essentially constructs a four-dimensional spatiotemporal safety assessment index. It not only includes spatial occupancy information but also establishes a data interface with "time t" as the retrieval axis. This allows the robot to retrieve timestamped "spatiotemporal key points" in subsequent steps. By directly using the input index and aligning the coordinates on the corresponding time-bound plane in the heatmap, the collision probability value of the path point at a specific time point can be extracted instantly. In this way, the generated spatiotemporal trajectory probability heatmap achieves a direct quantitative mapping from "obstacle motion prediction" to "robot path safety assessment," providing a continuous and accurate basis for subsequent filtering of unsafe path nodes.

[0044] Step S130: Synchronously assign timestamps to the robot's initial planned path points, expand the initial planned path points carrying timestamps into a spatiotemporal key point sequence containing spatial coordinates and time; map the spatiotemporal key point sequence to a spatiotemporal trajectory probability heatmap.

[0045] This step extends the robot's two-dimensional spatial path to a three-dimensional spatiotemporal dimension and establishes a mapping relationship with the heatmap generated in step S120. In a dynamic environment, the safety status of the same spatial location can vary drastically at different times. If the path point only contains spatial coordinates without temporal information, only fuzzy spatial matching can be performed, which cannot answer the crucial question of "whether the obstacle is still there when I expect to reach this location".

[0046] To this end, this step first constructs a high-order continuous velocity planning curve based on the kinematic constraint parameters of the robot chassis, and performs time parameterization mapping and step-by-step accumulation by combining the spatial geometric distance between adjacent path points. It accurately assigns absolute timestamps to each initial planned path point, transforming the static spatial path into a spatiotemporal key point sequence containing spatial coordinates and time, thus overcoming the defects of velocity mutation and spatiotemporal misalignment in traditional planning from the source.

[0047] Subsequently, the spatiotemporal key point sequence is mapped to a spatiotemporal trajectory probability heatmap. The heatmap is a three-dimensional structure, storing the probability that "position (X,Y) is occupied by an obstacle at time T" using a spatiotemporal grid. Path points carrying timestamps can be directly indexed by the time dimension to retrieve their collision risk corresponding to the expected arrival time. To further improve query accuracy, this step calculates the collision probability of each key point using temporal interpolation as the mapping result. When the timestamp falls between two discrete time window layers, interpolation is performed based on the time offset ratio to eliminate the probability step error caused by discretization. This achieves a high-precision quantitative assessment of collision risk in a continuous spatiotemporal domain, providing a reliable basis for subsequent steps to accurately screen and filter out high-risk nodes.

[0048] In a specific implementation, timestamps are synchronously assigned to the robot's initial planned path points, and the initial planned path points carrying timestamps are expanded into a sequence of spatiotemporal key points containing spatial coordinates and time. This can be achieved through the following three steps: Extract the chassis kinematic constraint parameters of the robot, and based on the chassis kinematic constraint parameters and the spatial geometric distance between adjacent initial planned path points, construct a high-order continuous velocity planning curve; By using a high-order continuous velocity planning curve to perform time parameterization mapping on spatial geometric distance, the relative time interval required for smooth transition between adjacent nodes can be calculated. Based on the progressive accumulation of each relative time interval and the system's initial time, an absolute timestamp is assigned to each initially planned path point to generate a spatiotemporal key point sequence.

[0049] Specifically, timestamps are synchronously assigned to the robot's initial planned path points, and these timestamped initial planned path points are expanded into a sequence of spatiotemporal key points containing spatial coordinates and time. This is achieved through the following three steps: Extract kinematic constraints and construct high-order continuous velocity programming curves First, the system extracts the robot's chassis kinematic constraint parameters from the chassis control module. To prevent sudden stops or starts during robot movement that could cause mechanical impact or slippage on the chassis, the kinematic constraint parameters must include at least the maximum operating speed. Maximum acceleration and maximum jerk .

[0050] Subsequently, the system calculates the spatial geometric distance between two adjacent initially planned path points. Based on the aforementioned constraint parameters, the system maps and constructs a high-order continuous velocity planning curve to connect the two points. To ensure extremely smooth chassis movement, this curve at any given time... All must strictly satisfy the following physical boundary constraints:

[0051] in, Represents the time variable of continuous motion; express Instantaneous velocity at a given moment; express Instantaneous acceleration at a given moment (i.e., the first derivative of the velocity); express The instantaneous jerk at time t (i.e., the second derivative of the velocity); This represents the maximum operating speed constraint of the chassis extracted by the system; This indicates the maximum acceleration constraint of the chassis; This indicates the maximum jerk constraint of the chassis.

[0052] Temporal parameterization mapping of spatial distance and solution of relative time interval After constructing the high-order continuous velocity planning curve, the system performs time-parameterized mapping of spatial geometric distances. The physical essence of this mapping process is: solving for the robot's speed while satisfying the aforementioned high-order continuous velocity curve. Smoothly traverse the geometric distance under boundary constraints. The time required.

[0053] The system establishes an equation relating distance and time by performing definite integral operations on the speed planning curve:

[0054] in, The spatial geometric distance between adjacent nodes is obtained above; For integration time; To construct a high-order continuous velocity programming curve function; This represents the unknown variable, namely the relative time interval required for the robot to smoothly transition along the geometric distance of this space.

[0055] By solving the equations for each kinematic stage of the curve (such as the acceleration-increasing segment, the uniform acceleration segment, and the acceleration-decreasing segment), the system can uniquely calculate the relative time interval. .Should It ensures extremely smooth chassis movement while maximizing the robot's mobility.

[0056] The generation of spatiotemporal key point sequences through the progressive accumulation of relative time. In order to align the robot's trajectory with the "spatiotemporal trajectory probability heatmap" of the dynamic obstacles on the same absolute time axis, the system uses the current system time obtained by the robot at the starting path point. As an initial benchmark.

[0057] Subsequently, the system follows the planned path, using a simple recursive formula to cascade and accumulate the relative time intervals between adjacent nodes, accurately assigning a corresponding absolute timestamp to each initially planned path point:

[0058] in, This is the absolute timestamp of the current node. The absolute timestamp of the previous node This refers to a relative time interval.

[0059] Through the above calculations, the geometric path points that originally only contained spatial coordinates were successfully expanded and upgraded to spatiotemporal key points that included both spatial coordinates and the absolute time dimension:

[0060] in, Indicates the first A key point in time and space; and This represents its corresponding two-dimensional spatial coordinates; This indicates its corresponding absolute timestamp.

[0061] By combining all nodes in sequence, a complete sequence of spatiotemporal key points is finally generated:

[0062] in, This represents the final set of spatiotemporal key point sequences. to This represents each key point in spacetime that carries an absolute timestamp; This represents the index of the total number of path points in the initial planning.

[0063] In summary, this embodiment first constructs a high-order continuous velocity planning curve by combining chassis kinematic constraints and geometric paths. Then, it transforms the spatial span into a relative time interval by solving the time parameterized definite integral. Finally, it performs cascade accumulation based on the current time and assigns an absolute timestamp to each path point. This seamlessly reconstructs the initial planned path points in pure geometric space into a sequence of spatiotemporal key points containing location and predicted arrival time, providing a control basis for subsequent dynamic obstacle avoidance in the spatiotemporal trajectory probability heatmap.

[0064] In a specific implementation, mapping the spatiotemporal key point sequence to a spatiotemporal trajectory probability heatmap can be achieved through the following two steps: Determine the two adjacent time window layers in the spatiotemporal trajectory probability heatmap of any spatiotemporal key point with its absolute timestamp. Based on the time offset ratio of the absolute timestamp relative to the two time window layers, the temporal domain interpolation is performed on the grid occupancy probability corresponding to the spatiotemporal key point under the two time window layers to calculate the collision probability of the spatiotemporal key point, which is used as the mapping result of the spatiotemporal key point to the spatiotemporal trajectory probability heatmap.

[0065] Specifically, in actual engineering operations, the spatiotemporal trajectory probability heatmap of dynamic obstacles is composed of a series of discrete "time window layers" stacked at fixed time steps (e.g., predicting one layer at 0.1s); while the absolute timestamp carried by the spatiotemporal key points of the robot is a continuously evolving precise physical time (e.g., 0.13s).

[0066] To eliminate this misalignment between discrete layers and continuous time, the system first analyzes the temporal and spatial attributes of any target spatiotemporal key point. Then, it places the absolute timestamp into the time axis of the spatiotemporal trajectory probability heatmap for retrieval, precisely locating the two adjacent time window layers (i.e., the previous time-step layer and the next time-step layer) that "wrap" the absolute timestamp in the middle.

[0067] After locating two adjacent time window layers, the system uses the spatial two-dimensional coordinates contained in the spatiotemporal keypoint to extract the "grid occupancy probability" (i.e., the original collision risk probability in a discrete state) of its corresponding position under these two time window layers. To obtain the true probability of the keypoint at a precise time, the system calculates the time difference between the absolute timestamp and the previous time window layer, and divides it by the total time step between the two layers to obtain a time offset ratio with a value between [0,1]. Finally, the system uses this time offset ratio. Using the weights, the occupancy probabilities of the two extracted grids are calculated using temporal linear interpolation. The specific simplified mapping logic is as follows:

[0068] in, This represents the calculated precise collision probability of the spatiotemporal keypoint in a continuous temporal dimension. and These represent the probability of the spatial coordinates being occupied by the grid extracted in the previous time window layer and the next time window layer, respectively. This is the time offset ratio calculated above.

[0069] The continuous collision probability calculated above This serves as the final mapping result of spatiotemporal key points in the spatiotemporal trajectory probability heatmap. This dynamic interpolation mapping mechanism based on the time offset ratio eliminates the probability step error caused by the time discretization of the heatmap, enabling the collision risk obtained by the robot planning layer to maintain extremely high smoothness and accuracy on the time axis.

[0070] Step S140: Based on the collision probability of each key point in the spatiotemporal key point sequence in the spatiotemporal trajectory probability heatmap, replace unsafe nodes with local obstacle avoidance alternative paths with collision probabilities exceeding a preset threshold, and generate optimized spatiotemporal obstacle avoidance paths.

[0071] This step performs safety filtering and optimized reconstruction of the path based on the collision probabilities of each spatiotemporal key point output in step S130. Through a local replacement strategy, this step retains the safe portions of the original path while selectively bypassing high-risk areas, providing an optimized path foundation for timestamp calibration and path execution in step S150.

[0072] In a specific implementation, step S140 can be achieved by the following four steps: Traverse the spatiotemporal key point sequence and mark key points with a collision probability greater than a preset threshold as unsafe nodes; The nearest safe node found by backtracking forward and searching backward in the spatiotemporal key point sequence for the unsafe node is determined and used as the previous and subsequent anchor points, respectively. In the spatiotemporal trajectory probability heatmap, plan a local alternative path that connects the preceding anchor point and the following anchor point, and satisfies the robot kinematic constraints and the collision probability of each spatiotemporal grid is lower than a preset threshold. Replace the corresponding unsafe nodes with local alternative paths and update the spatiotemporal key point sequence to generate an optimized spatiotemporal obstacle avoidance path.

[0073] Specifically, the system traverses the spatiotemporal key point sequence in chronological order, comparing the collision probability of each key point with a set safety threshold. If the collision probability of a key point exceeds the limit, it is marked as an "unsafe node," thereby accurately identifying the high-risk spatiotemporal region where a collision may occur in the future. After identifying the dangerous region, the system uses the unsafe node as the origin in the sequence and traverses both the historically planned path (backward tracing) and the future path to be traveled (backward search) to find the absolutely safe node closest to the dangerous region. This node is then established as the starting point (i.e., the preceding anchor point) and the ending point (i.e., the following anchor point) of the local detour, respectively.

[0074] Within the spatiotemporal trajectory probability heatmap, the system initiates a local path search, starting from the preceding anchor point and ending at the following anchor point. The system imposes dual constraints on the search algorithm: the newly generated local alternative path must simultaneously satisfy both "compliance with chassis dynamics constraints" and "collision probabilities throughout the entire path being below a preset threshold," thereby ensuring the physical feasibility and absolute safety of the new road segment. Finally, the system removes dangerous road segments located between the preceding and following anchor points in the original sequence and inserts the newly planned "local alternative path" in its original position. Through this replacement operation, the system achieves seamless stitching of the underlying data sequence, ultimately outputting an optimized spatiotemporal obstacle avoidance path that circumvents all dynamic risks.

[0075] Step S150: Based on the actual length of the optimized spatiotemporal obstacle avoidance path, recalibrate the timestamps of each spatiotemporal key point on the path, and control the robot to perform dynamic obstacle avoidance actions according to the optimized spatiotemporal obstacle avoidance path.

[0076] After generating the optimized obstacle avoidance path in step S140, some unsafe nodes in the original path are replaced by local alternative paths, resulting in a discrepancy between the actual geometric length of the optimized path and the original initial path. Since the original timestamps of each spatiotemporal key point are calculated and allocated based on the initial path length, continuing to use the original timestamps after the path change will cause a mismatch between the time information of the spatiotemporal key points and the spatial path length. This will prevent the robot from reaching the predetermined position at the predetermined time, and subsequent multi-robot collaborative scheduling and low-level control execution will fail. Therefore, this step requires recalibrating the timestamps of each spatiotemporal key point based on the actual length of the optimized path.

[0077] In a specific implementation, step S150 can be achieved by the following four steps: Calculate the cumulative geometric arc length of the optimized spatiotemporal obstacle avoidance path; By combining the physical boundaries of the robot's maximum speed and maximum acceleration, a time-optimal trajectory parameterization algorithm is applied to the cumulative geometric arc length to generate a nonlinear velocity planning curve that conforms to the chassis kinematic constraints. Based on the nonlinear velocity programming curve, the timestamps of each spatiotemporal key point on the optimized spatiotemporal obstacle avoidance path are reassigned to complete the recalibration of the timestamps of each spatiotemporal key point on the optimized spatiotemporal obstacle avoidance path and generate a dynamic executable reference trajectory. The dynamic executable reference trajectory is input to the preset trajectory tracking controller, the trajectory tracking deviation of the robot in the current control cycle is calculated, and the linear velocity and angular velocity control commands that satisfy the chassis kinematic constraints are generated based on the trajectory tracking deviation to drive the robot to complete dynamic obstacle avoidance.

[0078] First, the cumulative geometric arc length of the optimized path is calculated to obtain the actual length after the path change. Second, combining the robot's maximum speed and maximum acceleration physical boundaries, a time-optimal trajectory parameterization algorithm is applied to the cumulative arc length to generate a nonlinear velocity planning curve that conforms to the chassis kinematic constraints. Third, based on this velocity planning curve, the timestamps of each spatiotemporal key point on the optimized path are reassigned to complete the timestamp recalibration and generate a dynamically executable reference trajectory. Finally, the reference trajectory is input into a preset trajectory tracking controller to calculate the tracking deviation and generate linear and angular velocity control commands to drive the robot to complete dynamic obstacle avoidance. Through the above steps, this method realizes a complete closed loop from spatiotemporal path planning to underlying motion control.

[0079] This application also includes a multi-robot spatiotemporal conflict collaborative scheduling mechanism. In a multi-robot collaborative operation scenario, each robot broadcasts and shares its optimized spatiotemporal obstacle avoidance path generated in step S150 in the multi-robot network, and monitors the path information of other robots in real time.

[0080] The aforementioned multi-machine spatiotemporal conflict collaborative scheduling mechanism can be implemented in the following four steps: Each robot broadcasts and shares its generated optimized spatiotemporal obstacle avoidance path in the multi-robot network; If a trajectory conflict is detected between the spatiotemporal paths of multiple robots within an overlapping spatiotemporal grid, then the collision probability of the spatiotemporal grid where the conflict occurs is queried in the spatiotemporal trajectory probability heatmap constructed by each of the robots involved in the conflict. Obtain the task priority and remaining battery power of each conflicting robot, and then weight and fuse the collision probability, task priority and remaining battery power to calculate the comprehensive arbitration weight of each conflicting robot. Robots with lower overall arbitration weights are controlled to execute waiting loop strategies or deceleration and hovering actions, while robots with higher overall arbitration weights are given priority to pass through conflict areas.

[0081] When multiple robots are detected to have trajectory conflicts within overlapping spatiotemporal grids, this mechanism first queries the collision probability in the spatiotemporal trajectory probability heatmap generated in step S120 for the spatiotemporal grid where the conflict occurs, serving as a quantitative basis for the urgency of the conflict risk. Subsequently, it acquires the task priority and remaining battery power of each conflicting robot, weights and fuses the collision probability, task priority, and remaining battery power to calculate the comprehensive arbitration weight for each conflicting robot. Finally, the robot with the lower comprehensive arbitration weight is controlled to execute a waiting loop strategy or a deceleration and hovering action, while the robot with the higher comprehensive arbitration weight is given priority to pass through the conflict area. This mechanism incorporates the collision probability from the heatmap into the arbitration decision, enabling the scheduling result to consider not only task attributes and energy status but also dynamically correlated with the urgency of the current environmental risk, achieving a complete closed loop from single-machine path planning to multi-machine collaborative scheduling.

[0082] Based on the same inventive concept, as an implementation of the above-mentioned robot obstacle avoidance control method based on spatiotemporal trajectory prediction, this application also provides a robot obstacle avoidance control system based on spatiotemporal trajectory prediction. Figure 2 This is a structural diagram of the system in an embodiment of this application. See also... Figure 2 As shown, the system includes: The database construction module is used to acquire environmental perception data and body state data collected by multiple robots, and to construct a spatiotemporal database by spatiotemporal alignment under a unified coordinate system; it performs cross-consistency verification based on spatial constraints on cross-terminal environmental perception data to determine the spatiotemporal dynamic reliability weight of each robot's perception data. The heatmap construction module is used to predict candidate motion trajectories of dynamic obstacles in future time periods based on environmental perception data; and to calculate the distribution probability of candidate motion trajectories in spatiotemporal space to construct a spatiotemporal trajectory probability heatmap of dynamic obstacles. The mapping module is used to synchronously assign timestamps to the robot's initial planned path points, expand the initial planned path points carrying timestamps into a spatiotemporal key point sequence containing spatial coordinates and time, and map the spatiotemporal key point sequence to a spatiotemporal trajectory probability heatmap. The spatiotemporal obstacle avoidance path generation module is used to filter out unsafe nodes whose collision probability exceeds a preset threshold based on the collision probability of each key point in the spatiotemporal key point sequence in the spatiotemporal trajectory probability heatmap, and generate an optimized spatiotemporal obstacle avoidance path. The control module is used to recalibrate the timestamps of each spatiotemporal key point on the optimized spatiotemporal obstacle avoidance path based on the actual length of the path, and control the robot to perform dynamic obstacle avoidance actions according to the optimized spatiotemporal obstacle avoidance path.

[0083] This application integrates robot chassis kinematic constraints into spatial geometric paths and utilizes high-order continuous velocity planning and time parameterized mapping to accurately calculate absolute timestamps, seamlessly upgrading traditional pure spatial paths into a spatiotemporal key point sequence, overcoming the defects of velocity abrupt changes and spatiotemporal misalignment. Simultaneously, it introduces a time-domain interpolation mechanism based on time offset ratios to eliminate probabilistic step errors caused by the heat map layer, achieving high-precision quantitative assessment of collision risks in the continuous time domain. By screening high-risk nodes and locating preceding and following safety anchors, seamless reconnection and replacement are performed in the spatiotemporal trajectory probability heat map using local alternative paths, balancing the computational efficiency of local planning with the safe and executable nature of global paths. This significantly improves the smoothness, passage efficiency, and control robustness of the robot's autonomous obstacle avoidance in complex dynamic scenarios.

[0084] This application also provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a program for a robot obstacle avoidance control method based on spatiotemporal trajectory prediction. When the processor executes the computer program, it implements all the steps of the aforementioned robot obstacle avoidance control method based on spatiotemporal trajectory prediction, such as S110 to S150 and their sub-steps. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the aforementioned system embodiments.

[0085] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete this application. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0086] The electronic device may be a computing device such as a server or cloud server deployed on a robot obstacle avoidance control system based on spatiotemporal trajectory prediction. The electronic device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include data acquisition cards and output cards for connecting sensors and actuators, network communication devices, buses, etc.

[0087] The processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (OPGs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the entire electronic device via various interfaces and lines.

[0088] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created according to system usage, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart memory card, secure digital card, flash memory card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0089] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0090] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A robot obstacle avoidance control method based on spatiotemporal trajectory prediction, characterized in that, The method includes: The system acquires environmental perception data and body state data collected by multiple robots, and performs spatiotemporal alignment under a unified coordinate system to construct a spatiotemporal database; it then performs cross-consistency verification based on spatial constraints on the environmental perception data across terminals to determine the spatiotemporal dynamic reliability weight of each robot's perception data. Based on the environmental perception data, candidate motion trajectories of dynamic obstacles are predicted in future time periods; and the distribution probability of the candidate motion trajectories in spatiotemporal space is calculated to construct a spatiotemporal trajectory probability heatmap of dynamic obstacles. The robot's initial planned path points are synchronously assigned timestamps, and the initial planned path points carrying timestamps are expanded into a spatiotemporal key point sequence containing spatial coordinates and time; the spatiotemporal key point sequence is mapped to the spatiotemporal trajectory probability heatmap. Based on the collision probability of each key point in the spatiotemporal key point sequence in the spatiotemporal trajectory probability heatmap, unsafe nodes with collision probabilities exceeding a preset threshold are replaced with local obstacle avoidance alternative paths to generate optimized spatiotemporal obstacle avoidance paths. Based on the actual length of the optimized spatiotemporal obstacle avoidance path, the timestamps of each spatiotemporal key point on the path are recalibrated, and the robot is controlled to perform dynamic obstacle avoidance actions according to the optimized spatiotemporal obstacle avoidance path.

2. The method according to claim 1, characterized in that, The construction of a spatiotemporal database through spatiotemporal alignment in a unified coordinate system includes: Spatial coordinate transformation and timestamp synchronization are performed on the environmental perception data and the body state data respectively. The spatiotemporal positions of each dynamic obstacle and the spatiotemporal poses of each robot are associated and stored in a unified spatiotemporal coordinate system to construct the spatiotemporal database. The cross-consistency check based on spatial constraints on the environmental perception data across terminals includes: Extract the observation areas of each robot that have spatial overlap in the spatiotemporal database, and perform consistency comparison on the observation data across terminals within the spatially overlapping areas; remove abnormal observation data with deviations exceeding a preset threshold, and determine the spatiotemporal dynamic reliability weight based on the consistency of the retained observation data of each terminal.

3. The robot obstacle avoidance control method according to claim 1, characterized in that, The step of predicting candidate motion trajectories of dynamic obstacles in future time periods based on the environmental perception data includes: The spatiotemporal dynamic reliability weight is used to adaptively extract features from cross-terminal environmental perception data to obtain observation features that characterize the current global state of the environment. The observed features are injected as conditional variables into the pre-trained trajectory prediction model. Through feature mapping and decoding calculations, a sequence of positions of dynamic obstacles in multiple consecutive time steps in the future is generated as candidate motion trajectories.

4. The method according to claim 1, characterized in that, The calculation of the distribution probability of the candidate motion trajectory in spatiotemporal space, and the construction of a spatiotemporal trajectory probability heatmap of the dynamic obstacle, includes: According to the preset time step, the predicted future period is divided into multiple consecutive time window layers, and each time window layer corresponds to a spatial section at a certain moment in the future. For each time window layer, calculate the geometric distance from the center point of each spatial grid in the layer to the candidate motion trajectory position point at the corresponding time. Using a preset probability decay function, the probability value of each spatial grid being occupied by the dynamic obstacle is calculated based on the geometric distance, and the probability values ​​under each time window layer are aggregated in a three-dimensional spatiotemporal coordinate system to generate the spatiotemporal trajectory probability heatmap.

5. The method according to claim 1, characterized in that, The process involves synchronously assigning timestamps to the robot's initial planned path points, and expanding these timestamped initial planned path points into a sequence of spatiotemporal key points containing spatial coordinates and time, including: Extract the chassis kinematic constraint parameters of the robot, and based on the chassis kinematic constraint parameters and the spatial geometric distance between adjacent initial planned path points, construct a high-order continuous velocity planning curve; The spatial geometric distance is mapped to a time parameter using the high-order continuous velocity planning curve, and the relative time interval required for a smooth transition between adjacent nodes is calculated. Based on the progressive accumulation of the relative time intervals and the system initial time, an absolute timestamp is assigned to each of the initial planned path points to generate the spatiotemporal key point sequence.

6. The method according to claim 5, characterized in that, The step of mapping the spatiotemporal key point sequence to the spatiotemporal trajectory probability heatmap includes: Determine the absolute timestamp of any of the aforementioned spatiotemporal key points to two adjacent time window layers in the spatiotemporal trajectory probability heatmap; Based on the time offset ratio of the absolute timestamp relative to the two time window layers, the temporal domain interpolation is performed on the grid occupancy probability corresponding to the spatiotemporal key point under the two time window layers to calculate the collision probability of the spatiotemporal key point, which is used as the mapping result of the spatiotemporal key point to the spatiotemporal trajectory probability heatmap.

7. The method according to claim 1, characterized in that, The step of generating an optimized spatiotemporal obstacle avoidance path by replacing unsafe nodes with local obstacle avoidance alternative paths with collision probabilities exceeding a preset threshold in the spatiotemporal trajectory probability heatmap of each key point in the spatiotemporal key point sequence, includes: Traverse the spatiotemporal key point sequence and mark key points with a collision probability greater than a preset threshold as unsafe nodes; The nearest safe node found by backtracking forward and searching backward in the spatiotemporal key point sequence for the unsafe node is determined and used as the previous and subsequent anchor points, respectively. In the spatiotemporal trajectory probability heatmap, a local alternative path is planned that connects the preceding anchor point and the following anchor point, and satisfies the robot kinematic constraints and the collision probability of each spatiotemporal grid is lower than the preset threshold. The corresponding unsafe nodes are replaced using the local alternative path, and the spatiotemporal key point sequence is updated to generate the optimized spatiotemporal obstacle avoidance path.

8. The method according to claim 1, characterized in that, Based on the actual length of the optimized spatiotemporal obstacle avoidance path, the timestamps of each spatiotemporal key point on the path are recalibrated, and the robot is controlled to perform dynamic obstacle avoidance actions according to the optimized spatiotemporal obstacle avoidance path, including: Calculate the cumulative geometric arc length of the optimized spatiotemporal obstacle avoidance path; Combining the robot's maximum speed and maximum acceleration physical boundaries, a time-optimal trajectory parameterization algorithm is applied to the cumulative geometric arc length to generate a nonlinear velocity planning curve that conforms to the chassis kinematic constraints; Based on the nonlinear velocity planning curve, the timestamps of each spatiotemporal key point on the optimized spatiotemporal obstacle avoidance path are reassigned to complete the recalibration of the timestamps of each spatiotemporal key point on the optimized spatiotemporal obstacle avoidance path and generate a dynamic executable reference trajectory. The dynamic executable reference trajectory is input to a preset trajectory tracking controller, the trajectory tracking deviation of the robot in the current control cycle is calculated, and linear velocity and angular velocity control commands that satisfy the chassis kinematic constraints are generated based on the trajectory tracking deviation to drive the robot to complete dynamic obstacle avoidance.

9. The method according to any one of claims 1-8, characterized in that, The method further includes: Each robot broadcasts and shares its generated optimized spatiotemporal obstacle avoidance path in the multi-robot network; If a trajectory conflict is detected between the spatiotemporal paths of multiple robots within an overlapping spatiotemporal grid, then the collision probability of the spatiotemporal grid where the conflict occurs is queried in the spatiotemporal trajectory probability heatmap constructed by each of the robots involved in the conflict. Obtain the task priority and remaining battery power of each conflicting robot, and then weight and fuse the collision probability, task priority and remaining battery power to calculate the comprehensive arbitration weight of each conflicting robot. Robots with lower overall arbitration weights are controlled to execute waiting loop strategies or deceleration and hovering actions, while robots with higher overall arbitration weights are given priority to pass through conflict areas.

10. A robot obstacle avoidance control system based on spatiotemporal trajectory prediction, characterized in that, The system includes: The database construction module is used to acquire environmental perception data and body state data collected by multiple robots, and to construct a spatiotemporal database by spatiotemporal alignment under a unified coordinate system; and to perform cross-consistency verification based on spatial constraints on the environmental perception data across terminals to determine the spatiotemporal dynamic reliability weight of each robot's perception data. The heatmap construction module is used to predict candidate motion trajectories of dynamic obstacles in future time periods based on the environmental perception data; and to calculate the distribution probability of the candidate motion trajectories in spatiotemporal space to construct a spatiotemporal trajectory probability heatmap of dynamic obstacles. The mapping module is used to synchronously assign timestamps to the robot's initial planned path points, expand the initial planned path points carrying timestamps into a spatiotemporal key point sequence containing spatial coordinates and time, and map the spatiotemporal key point sequence to the spatiotemporal trajectory probability heatmap. The spatiotemporal obstacle avoidance path generation module is used to replace unsafe nodes with local obstacle avoidance alternative paths with collision probabilities exceeding a preset threshold in the spatiotemporal trajectory probability heatmap of each key point in the spatiotemporal key point sequence, thereby generating an optimized spatiotemporal obstacle avoidance path. The control module is used to recalibrate the timestamps of each spatiotemporal key point on the optimized spatiotemporal obstacle avoidance path based on the actual length of the path, and control the robot to perform dynamic obstacle avoidance actions according to the optimized spatiotemporal obstacle avoidance path.

Images