Priority-based centralized multi-agv multi-path lane-changing decision planning method

Abstract

A priority-based centralized multi-AGV multi-path lane change decision planning method, comprising the following steps: S100: according to the pre-established multi-path lane passing rule, reasonably allocate path resources, and confirm the lane change decision result of whether the AGV under the multi-path lane needs to change lane; S200: according to the priority order of each AGV, receive the lane change decision result, plan the behavior level of the current AGV, plan the specific semantic action of lane change, and generate a time-space channel in the cloud for a single AGV. The method can make the multi-AGV system have stronger adaptability and higher efficiency on wide paths, and can reduce congestion and solve sudden situations on wide paths of the multi-AGV system.

Description

Technical Field

[0001] This disclosure belongs to the field of AGV intelligent control technology, and specifically relates to a priority-based centralized multi-AGV multi-path channel lane change decision-making and planning method. Background Technology

[0002] Automated Guided Vehicles (AGVs) are transportation devices capable of automatically handling materials. They typically use radio waves, cameras, lidar, or magnetic strips, nails, and QR codes marked on the ground for navigation. Compared to other logistics and transportation equipment, AGVs offer advantages such as high adaptability, high automation, reduced labor costs, and ease of maintenance. With rising labor costs and increasingly diverse production patterns, more and more companies are adopting highly automated production systems, and AGVs are a crucial component of these systems. Therefore, the design and research of AGVs are of great significance for enterprises to improve production efficiency and reduce production costs.

[0003] Compared to single AGVs, multi-AGV systems have a stronger ability to perform material handling operations. Furthermore, facing complex and ever-changing factory environments, multi-AGV systems can quickly respond to changes in the external environment and complete tasks flexibly and efficiently. Therefore, the development of multi-AGV systems has become inevitable. Compared to single AGVs, multi-AGV systems need to solve problems such as multi-machine collaboration, information interaction, and conflict resolution, making their design more complex. Currently, designing multi-AGV systems with high swarm dynamics, self-organization, and adaptability has become a research hotspot, mainly including the following research directions: research on motion analysis and control algorithms, research on autonomous sensing and networking algorithms, research on multi-machine positioning algorithms, research on multi-machine autonomous dynamic decision-making and path planning algorithms, and research on multi-machine formation combination and disbanding and biomimetic behavior simulation algorithms.

[0004] In recent years, little attention has been paid to the collaboration of multiple AGV systems on wide paths to reduce congestion and handle emergencies. Some existing methods can achieve cooperation between multiple AGVs to safely navigate complex road environments, but they struggle to efficiently utilize wide path space. Other methods can maximize the space utilization of a single AGV in complex environments, but they do not consider collaboration between multiple AGVs and are therefore difficult to apply to multi-AGV systems. Therefore, to enhance the adaptability and efficiency of multi-AGV systems on wide paths, a method suitable for reducing congestion and handling emergencies on wide paths is crucial. Summary of the Invention

[0005] To address the aforementioned technical issues, this disclosure presents a priority-based centralized multi-AGV multi-path channel lane-changing decision-making and planning method, comprising the following steps:

[0006] S100: Based on the pre-defined multi-path passage rules, rationally allocate path resources and confirm the lane change decision result of whether the AGV needs to change lanes under the multi-path passage.

[0007] S200: According to the priority order of each AGV, receive the lane change decision results, perform behavioral-level planning for the current AGV, plan the specific semantic actions of lane change, and generate a spatiotemporal channel in the cloud for the single AGV.

[0008] The above technical solution can efficiently utilize the advantages of multiple lanes in complex multi-path environments containing multiple cooperative AGVs, multiple non-cooperative AGVs, and obstacles, thereby maximizing the operating efficiency of the multi-AGV system and ensuring safety. Attached Figure Description

[0009] Figure 1 This is a priority-based centralized multi-AGV multi-path channel lane change decision-making and planning method provided in one embodiment of the present disclosure;

[0010] Figure 2 This is a schematic diagram of the structure of a multipath channel and its connection with an external path provided in one embodiment of this disclosure;

[0011] Figure 3 This is a detailed design diagram of a multi-machine collaborative decision-making planning provided in one embodiment of this disclosure. Detailed Implementation

[0012] To enable those skilled in the art to understand the technical solutions disclosed herein, the following will describe them in conjunction with embodiments and related appendices. Figures 1 to 3 The technical solutions of various embodiments are described herein, and the described embodiments are only a part of the embodiments of this disclosure, not all of them. The terms "first," "second," etc., used in this disclosure are used to distinguish different objects, not to describe a specific order. Furthermore, "comprising" and "having," and any variations thereof, are intended to be omnipresent and non-exclusive. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, systems, products, or devices.

[0013] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this disclosure. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. Those skilled in the art will understand that the embodiments described herein can be combined with other embodiments.

[0014] See Figure 1 In one embodiment, this disclosure discloses a priority-based centralized multi-AGV multi-path channel lane-changing decision-making and planning method, including the following steps:

[0015] S100: Based on the pre-defined multi-path passage rules, rationally allocate path resources and confirm the lane change decision result of whether the AGV needs to change lanes under the multi-path passage.

[0016] S200: According to the priority order of each AGV, receive the lane change decision results, perform behavioral-level planning for the current AGV, plan the specific semantic actions of lane change, and generate a spatiotemporal channel in the cloud for the single AGV.

[0017] In this embodiment, the method designs a new set of multipath channel passage rules, and a corresponding cloud-based centralized method for making lane-changing decisions for multiple cooperative AGVs based on priority, and for planning subsequent actions.

[0018] The path resource allocation section primarily receives real-time dynamic scene information streams and the real-time positions of each AGV from the previous module, and allocates path resources rationally according to pre-defined multi-path channel traffic rules. This method defines a relatively wide road segment as a multi-path channel, i.e., a path containing multiple parallel lanes, each lane wide enough to accommodate one AGV. On the multi-path channel, this method sets the two lanes at the two ends as fixed-direction lanes with opposite travel directions, while the other lanes are set as variable-direction lanes with variable travel directions. By considering the travel directions and real-time positions of all AGVs on the entire multi-path channel, the method autonomously adjusts the travel direction on the variable-direction lanes in real time. Simultaneously, this method also ensures AGV safety by setting entry and exit buffer zones and corresponding entry and exit rules at the start and end points of the multi-path channel. Based on the multi-path channel rules proposed in this method, the AGV autonomously makes real-time lane-changing decisions according to relevant information, such as whether to adjust the direction of the variable-direction lane, calculating the priority of each AGV, and whether the AGV needs to change lanes to an adjacent lane, whether to change lanes to the left or right. The path resource allocation module will send semantic information-level behaviors to the behavior-level planning module.

[0019] The behavior-level planning component is another part of the priority-based centralized multi-AGV multi-path channel lane-changing decision-making and planning method. Following the priority order of each AGV, the behavior planning component receives the lane-changing decision results (left lane change, right lane change, or remain unchanged, etc.) from the path resource allocation component, plans the next behavior of the current AGV, such as whether to start changing lanes immediately or to travel straight for a certain distance before changing lanes, and generates a spatiotemporal channel to send to the next module.

[0020] In another embodiment, step S100 further includes the following steps:

[0021] S101: Update lane direction by collecting information within the scene; the information within the scene includes: the position, speed, size and type of non-cooperative targets and the position and type of cooperative targets;

[0022] S102: Determining the priority of cooperation objectives;

[0023] S103: Perform lane change assessment according to pre-set rules.

[0024] In this embodiment, the path resource allocation part mainly operates in the multipath channel. Therefore, in order to maximize efficiency, this method designs a new multipath channel passage rule.

[0025] To make full use of the space of a wider path, alleviate congestion caused by multiple AGVs passing through the same path, and support flexible obstacle avoidance, so as to maximize the overall operating efficiency of the multi-AGV system while ensuring safety, this method defines the wider road segment as a multi-path path, which is divided into multiple parallel lanes that can accommodate a single AGV. AGVs can change lanes between adjacent lanes.

[0026] To accommodate access to multi-path channels and other routes, the traffic rules are explained using a multi-path channel with five lanes, each connected to a single-path channel at the beginning and end, as an example. A schematic diagram is shown below. Figure 2 As shown.

[0027] exist Figure 2 The central section is a multi-path channel, with single-path channels at its left and right ends, each accommodating only one AGV at a time. Both channels allow AGVs to enter and exit, but only one AGV can enter or exit at a time on each side. To ensure AGVs can freely traverse the path left or right under any circumstances, the two edge lanes are designated as fixed-direction lanes with opposite and unchangeable directions, shown as lanes ① and ⑤ in the diagram. When entering the multi-path channel from left to right, entry is only possible through lane ①; when entering from right to left, entry is only possible through lane ⑤. There is no need to wait in the buffer zone upon entry. Simultaneously, to fully utilize the remaining space, the other lanes, ②, ③, and ④, are designated as variable-direction lanes, meaning the travel direction can be dynamically adjusted according to actual conditions. In other words, throughout operation, the travel direction of AGVs on fixed-direction lanes ① and ⑤ remains constant; the travel direction of AGVs on variable-direction lanes ②, ③, and ④ may change.

[0028] This method sets the rules for leaving the multipath channel as follows: For AGVs leaving the multipath channel from right to left (i.e., from the left end), lanes ②, ③, ④, and ⑤ can be selected; for AGVs leaving the multipath channel from left to right (i.e., from the right end), lanes ①, ②, ③, and ④ can be selected. When an AGV enters the buffer zone, it will report to the external scheduling system. Only when the scheduling system determines that it can enter the external path and issues the subsequent path and release signal can the AGV leave the buffer zone and exit the multipath channel.

[0029] To efficiently utilize the advantages of multiple lanes, maximize the operational efficiency of multi-AGV systems, and ensure operational safety in complex multi-path environments containing multiple cooperative AGVs, multiple non-cooperative AGVs, and obstacles, this method provides a cloud-based approach for centrally managing lane-changing decisions based on priority for multiple cooperative AGVs and planning their subsequent actions. This method mainly consists of two parts: path resource allocation and behavior planning.

[0030] like Figure 3 As shown, the path resource allocation section needs to receive two types of information to determine whether a lane change is necessary: ​​1) the position, speed, size, and type of non-cooperative targets. This information is mainly obtained by the perception fusion module, which collects information about non-cooperative targets through target detection and tracking; 2) information about cooperative targets. This information mainly relies on the server to aggregate the position, type, and other information reported by all AGVs in the scene. Both types of information are sent in the format MOT_info, which includes the target's position, speed, size, and type.

[0031] Specifically, this part can be divided into three processes, such as... Figure 3 As shown, this includes lane direction updates, priority determination of cooperative targets, and lane change assessment.

[0032] By collecting the aforementioned information within the scene, the path resource allocation section first updates the path direction to... Figure 2 For example, it determines whether each variable-direction lane (lanes ②, ③, and ④) is currently traveling from left to right or from right to left according to pre-set rules. Initially, the number of lanes traveling from left to right is l. l And the number of lanes from right to left l r The AGVs are equal or approximately equal in number, each distributed at both ends, traveling in the same direction as the fixed-direction lanes at the edges. Each time an AGV enters or leaves the multipath channel, the number of AGVs (n) from left to right in the multipath channel is counted. l And the number of AGVs from right to left, n r If there is:

[0033]

[0034] Where μ < 1 is the set lane direction update threshold, the number of lanes in the direction with fewer AGVs per unit lane is reduced by one, and the number of lanes in the other direction is increased by one. That is, the direction of the variable-direction lane in the direction with fewer AGVs per unit lane at the boundary between the two driving directions is changed until only the outermost fixed-direction lane remains. The value of μ can be 0.8, and can be adjusted according to the actual situation.

[0035] Next, the priority of the cooperation objectives is determined. This method requires first evaluating the lane change and planning the behavior of the highest priority AGV. After receiving confirmation from the corresponding AGV, the priority of the remaining cooperation AGVs is determined, and the highest priority AGV is selected to proceed with the subsequent work.

[0036] Finally, lane change evaluation is performed. This method evaluates lane change decisions according to the following rules: If the current lane has been adjusted in the current lane direction update, the current AGV is directly set to change lanes to an adjacent lane in the same direction. If the current lane direction does not change, the number of AGVs in its adjacent lanes in the same direction is checked. It is assumed that both adjacent lanes can be used for lane change, and the current AGV is in lane j. j Its distance from the outermost lane in the same direction is x. j The number of AGVs in this lane is m. j The adjacent lanes in the same direction are called lanes. j+1 Its distance from the outermost lane in the same direction is x. j+1 The number of AGVs on the lane is m j+1 The other adjacent lane in the same direction is lane j-1 Its distance from the outermost lane in the same direction is x. j-1 The number of AGVs on the lane is m j-1 Then from lane j Lane change towards lane j+1 The lane's rating is:

[0037]

[0038] towards lane j-1 The lane's rating is:

[0039]

[0040] Where c is a constant, n is the total number of AGVs in this direction, l is the number of lanes in this direction, and X is the total width of the free path. When

[0041] max(score j+1 score j-1 )>τ

[0042] or

[0043]

[0044] Where τ>0 is the lane change threshold. The AGV only needs to change lanes when the nearest obstacle threshold is reached, and it changes lanes to the side with the higher score. Intuitively speaking, the above formula means choosing the adjacent lane with fewer AGVs to change lanes. At the same time, the term after the plus sign drives the AGV to change lanes to a more distant lane, so that the lanes near the edge of the same direction can be cleared, making it easier for subsequent AGVs entering the free path to change lanes.

[0045] If the lane with the highest score becomes a lane in the same direction during this round of lane direction updates, it is necessary to ensure that there are no AGVs traveling in the opposite direction on the remaining path of the target lane in the current AGV's direction of travel. If there are, and the obstacle distance is greater than the nearest obstacle threshold, then skip this round of lane change; otherwise, the AGV stops and waits. If there are no, then it can change lanes to enter.

[0046] In another embodiment, the priority in step S102 is defined as the reciprocal of the smaller of the length of the AGV's current position from the end of its target multipath channel and the length of the nearest obstacle.

[0047] In this embodiment, the priority is set as the reciprocal of the smaller of the length of the AGV's current position from the end of its target multipath channel and the length of the nearest obstacle. This represents the distance of the i-th AGV from the end point of its target multipath channel. This represents the distance of the i-th AGV to the nearest obstacle. If there is no obstacle in front, it is set to infinity. The priority P of the i-th AGV is then... i This can be represented as:

[0048]

[0049] Where c is a constant term to avoid the denominator being 0.

[0050] In another embodiment, the lane change assessment in step S103 is determined by the priority of the lane change assessment, with the lane change assessment performed first for lane changes with higher priority.

[0051] In this embodiment, lane change evaluation is performed first based on priority. The function of this process is to determine whether the current AGV needs to change lanes, and whether to change lanes to the left or right, and then send this result to the behavior planning part to plan semantic-level actions.

[0052] In another embodiment, the behavior-level planning in step S200 is divided into three parts: guiding branch, scenario implementation, and behavior evaluation. The guiding branch part is responsible for expanding the action sequence according to the AGV's predefined strategy based on the priority of the local machine, and predicting the future trajectories of other cooperative and non-cooperative vehicles. In the scenario implementation, the strategy obtained from the guiding branch needs to be positively simulated with the trajectories of other cooperative and non-cooperative AGVs to generate all possible trajectories and simulate the scenario of executing the entire strategy. The behavior evaluation part evaluates the decision and selects the best one to execute.

[0053] In this embodiment, when the path resource allocation part issues a command requiring a lane change, the behavior-level planning part needs to plan the specific semantic actions for the current AGV to change lanes and generate a Space-time channel for the individual AGV. The behavior planning part also runs on the server-side CPU, and its relationship with other parts is as follows: Figure 3 As shown, behavior planning is completed based on the commands sent by the upper-level module and the aggregated trajectories of all cooperating AGVs currently executing.

[0054] This method divides behavior planning into three processes: guided branching, scenario implementation, and behavior evaluation. Finally, based on the semantic-level action sequence, a Space-time channel is generated in the cloud. Ultimately, as shown... Figure 3 As shown, the server will publish a Space-time channel to the standalone AGV to implement the AGV's specific motion planning. The Space-time channel ensures that the AGV can safely complete lane changes within a certain time period and range without interference from other cooperative or non-cooperative targets. The main method is to expand the AGV state points predicted by behavior planning in the three-dimensional space of horizontal, vertical, and time, using obstacles as the termination boundary of the expansion, thus generating an absolutely safe channel, i.e., a space-time channel. Based on the characteristics of Bézier curve generation, control points of the Bézier curve are selected in this channel. The Bézier curve is completely constrained within the convex hull formed by the control points, ensuring the generation of an absolutely safe and smooth Bézier curve, which serves as the path for the AGV. The formula for defining an m-order Bézier curve using m+1 control points on a fixed interval t∈[0,1] is as follows:

[0055]

[0056] Where, p i Indicates control points, This represents a Bernstein polynomial.

[0057] In another embodiment, the guiding branch obtains all possible semantic-level action sequences as lane-changing strategies by introducing semantic-level operations and closed-loop decision trees.

[0058] In this embodiment, the guiding branch process introduces semantic-level operations, such as changing lanes to the left, changing lanes to the right, and maintaining the current lane. A closed-loop decision tree is then used to obtain all possible sequences of semantic-level actions. Each node in the closed-loop decision tree represents a predefined semantic action, directed edges represent the execution order in time, and expanding downwards by one level represents a decision cycle. Each decision cycle contains at most one behavioral change. Starting from the root node, child nodes containing predefined actions are generated, and this process continues downwards. After establishing the closed-loop decision tree, the entire decision space is created. By traversing all paths from the root node to all leaf nodes, all sequences of semantic-level actions, also known as lane-changing strategies, can be obtained. Subsequent filtering and evaluation of all strategies yields the optimal lane-changing strategy. Furthermore, since the evaluation of each strategy is independent, strategies can be evaluated in parallel.

[0059] In another embodiment, the scenario implementation uses the previously generated trajectories of cooperative AGVs with higher priority than the current AGV for scenario implementation; while for cooperative AGVs with lower priority than the current AGV, the scenario implementation is based on the trajectories recorded in the previous round. If the trajectory of the cooperative AGV cannot fully cover the planning cycle of scenario implementation, the insufficient part is predicted; and for non-cooperative AGVs, their future is predicted based on their historical motion trajectories, assuming that their speed and acceleration remain unchanged in the future, to obtain their future trajectory.

[0060] In this embodiment, during scenario implementation, the strategy obtained from the guiding branch needs to be forward simulated with the trajectories of other cooperative AGVs and non-cooperative AGVs to generate all possible trajectories and simulate the scenario of executing the entire strategy. Unlike existing methods, in this method, for cooperative AGVs with higher priority than the current AGV, since they have already completed the current round of planning, their generated trajectories can be directly used for scenario implementation; while for cooperative AGVs with lower priority than the current AGV, they have not yet completed the current round of planning, so it is assumed that they move completely according to the trajectory obtained in the previous round, and the scenario implementation is based on the trajectory recorded in the previous round. Since the trajectory of the required strategy is long, the actual trajectories of other cooperative AGVs may not fully meet the needs of the entire scenario implementation process. For moments not covered by actual trajectories, the future trajectory is predicted based on its historical trajectory. For non-cooperative AGVs, this method predicts their future lateral velocity, acceleration, longitudinal velocity, and acceleration based on their historical motion trajectory, and assumes that their velocity and acceleration remain unchanged in the future to obtain the future trajectory.

[0061] This method models the hybrid forward simulation problem involving both cooperative and non-cooperative AGVs using a Partially Observable Markov Decision Process (POMDP). The POMDP model can be defined as a tuple. Where χ represents the state space. For behavioral space, For the observation space, T(x) t-1 a t x t )=p(x t |x t-1 a t ) is a state transition probability model, O(x t , z t )=p(z t |x t R(x) is the observation model. t-1 a t ) is in state x t-1 Perform action a t The reward function is determined by the condition. Since some states in the real world cannot be directly observed, such as the intention of a non-cooperative AGV, POMDP considers the current state x... t Maintain a confidence level The confidence level can be inferred using Bayes' theorem after the agent performs an action and the results are observed. t (x t )=p(x t |z t a t b t-1 The goal of POMDP is to find an optimal policy that maps from the confidence space to the action space. It can maximize the expected total discounted reward throughout the entire planning period.

[0062] Specifically, in this problem, firstly, this method requires the predicted behavioral trajectory from time step t to time step t+H. Compared to a single best move, This includes more comprehensive information about the environment and future predictions, which is crucial for subsequent movement planning. By maximizing the confidence level for the next moment, the state at the next moment can be obtained:

[0063]

[0064] Secondly, the setup of multiple agents needs to be taken into account.

[0065] During each long-term behavioral planning process, only the currently cooperating AGV can control its actions. The state transition model is defined as follows:

[0066]

[0067] Since cooperative AGVs record their specific behavioral trajectories, it can be assumed that the state of the first part is fixed during the entire planning period, while the state of the second part depends on assumptions and predictions. The state of non-cooperative AGVs, on the other hand, requires assumptions and predictions. Therefore, the state transition model under multi-agent interaction can be written as:

[0068]

[0069] Where 1 to M represent currently cooperating AGVs, and M+1 to N represent non-cooperating AGVs. For cooperative AGVs, the status is as follows Become a state State transition probability model:

[0070]

[0071] t K The time corresponding to the last trajectory point of the behavior planning trajectory saved for the cooperative AGV.

[0072] To simplify calculations, semantic-level actions are also introduced here, replacing the directly executed action 'a'. t Each semantic action lasts at most a few seconds, which restricts the exploration to a high likelihood region, effectively reducing computational complexity to a relatively small level while obtaining a sufficiently large planning horizon. Therefore, the new joint state transition model formula is:

[0073]

[0074] For other AGVs, semantic-level actions can also be introduced. First, for cooperative AGVs, this is not considered if the recorded trajectory still contains valid trajectory points; otherwise, its semantic action is considered to be maintaining its current lane and moving straight. Therefore, the state transition probability model for cooperative AGVs can be written as:

[0075]

[0076] semantic actions to maintain straight-ahead movement in the current lane

[0077] Although non-cooperative AGVs are uncontrollable, their intentions can be predicted, and semantic-level actions can be obtained. Their action generation model can be written as:

[0078]

[0079] In this method, for other cooperative AGVs, their intentions are directly generated from the preserved real trajectories; while for non-cooperative AGVs, their intentions—whether to change lanes or stay in the current lane—are predicted based on their historical poses and trajectories. In other words, the intention is determined based on the confidence level. Predict its entire planning cycle

[0080] Based on the above, the confidence level at the end of the long-term behavioral planning can be obtained. The calculation formula is as follows:

[0081]

[0082] Based on the above formula, a forward simulation was performed that combined cooperative AGVs and non-cooperative AGVs, and a safety mechanism was used to ensure the safety of the planning results.

[0083] Because real-world interference often causes deviations in predicted trajectories, safety mechanisms are needed to ensure sufficient safety. The first safety mechanism calculates safe distances in both the lateral and longitudinal directions when predicting the future trajectory of non-cooperative AGVs, and verifies whether the current strategy and the predicted trajectory meet these safe distance requirements. The second mechanism sets up safe alternative strategies in the strategy selection process to enhance the robustness of the decision-making layer. That is, when other AGVs do not cooperate with the lane-changing action, the lane-changing action can be canceled, and the corresponding action of the alternative strategy can be implemented. Each strategy has at least one alternative strategy; if none is available, it means that the behavior planning has failed, the system will display a warning signal, and trigger protective measures, such as emergency braking.

[0084] In another embodiment, the behavior evaluation assesses each strategy from three aspects: efficiency, safety, and the difference between the semantic action sequence and the lane change target. The optimal strategy is selected as the semantic action sequence to complete the lane change target, and the result of the behavior planning at this time is recorded for use in the behavior planning of other AGVs.

[0085] In another embodiment, behavior planning is only required for all AGVs inside the multipath channel when an AGV leaves or enters the multipath channel, selecting the highest priority AGV each time.

[0086] In this embodiment, to reduce computational overhead, this method does not perform lane-changing decisions and behavior planning in real time. Instead, it only performs behavior planning for all AGVs within the multipath channel when an AGV leaves or enters it. The method prioritizes the highest-priority AGV and plans its behavior sequentially. Furthermore, to ensure driving safety, the server requires confirmation of a new message from the AGV before re-evaluating priorities and selecting the highest-priority AGV for behavior planning. Otherwise, behavior planning for subsequent AGVs is halted until another AGV leaves or enters the multipath channel.

[0087] In another embodiment, path resource allocation is performed on the server-side CPU.

[0088] The path resource allocation section runs on the server's CPU, and its main function is to determine whether AGVs need to change lanes in multi-path channels. This section receives dynamic scene information and the real-time position of each AGV, and allocates path resources reasonably according to pre-defined multi-path channel passage rules. This aims to fully utilize the space of wider roads, alleviate congestion caused by multiple AGVs passing through the same road segment, and maximize the overall operating efficiency of AGVs.

[0089] Although embodiments of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the specific embodiments and application fields described above. The specific embodiments described above are merely illustrative and instructive, and not restrictive. Those skilled in the art can make many other forms based on the guidance of this specification and without departing from the scope of protection of the claims of the present invention, and all of these are within the scope of protection of the present invention.

Claims

1. A priority-based centralized multi-AGV multi-path lane changing decision planning method, comprising the following steps: S100: According to the pre-established multi-path lane passing rule, the path resource is reasonably allocated, and the lane changing decision result of whether the AGV needs to change lane under the multi-path lane is confirmed; S200: According to the priority order of each AGV, the lane changing decision result is received, the behavior level planning of the current AGV is carried out, the specific semantic action of lane changing is planned, and the space-time channel is generated in the cloud for the single AGV; The behavior level planning in step S200 is divided into three parts of guiding branch, scene implementation and behavior evaluation, the guiding branch part is responsible for expanding the action sequence according to the pre-defined strategy of the AGV according to the priority of the local machine, and predicting the future trajectory of other cooperative and non-cooperative vehicles; in the scene implementation, the strategy obtained by the guiding branch needs to be simulated with the trajectory of other cooperative AGVs and non-cooperative AGVs to generate all possible trajectories, simulate the scene of executing the whole strategy process; the behavior evaluation part evaluates the decision and executes the optimal one; The scene implementation uses the trajectory of the cooperative AGV with higher priority than the current AGV to implement the scene; while for the cooperative AGV with lower priority than the current AGV, the trajectory in the last round of record is used to implement the scene, if the trajectory of the cooperative AGV cannot completely cover the planning period of the scene implementation, the insufficient part is predicted; and for the non-cooperative AGV, its future is predicted according to its historical motion trajectory, and it is assumed that its speed and acceleration remain unchanged in the future to obtain the future trajectory.

2. The method of claim 1, wherein step S100 further comprises the following steps: S101: updating the lane direction by collecting information in the scene; wherein, The information in the scene includes the position, speed, size and type of the non-cooperative target and the position and type of the cooperative target; S102: Determination of the priority of the cooperative target; S103: Lane changing evaluation according to the pre-set rules.

3. The method of claim 2, wherein the priority in step S102 is defined as the reciprocal of the smaller length between the length of the current position of the AGV to the end point of the target multi-path lane and the length to the nearest obstacle.

4. The method of claim 2, wherein the lane changing evaluation in step S103 is to determine the evaluation order according to the priority, and the lane changing evaluation of the AGV with higher priority is performed first.

5. The method of claim 1, wherein the guiding branch obtains all possible semantic level action sequences as the lane changing strategy by introducing semantic level operations and closed loop decision trees.

6. The method of claim 1, wherein the behavior evaluation evaluates each strategy from the aspects of efficiency, safety, difference between the semantic action sequence and the lane changing target, selects the optimal strategy as the semantic level action sequence for completing the lane changing target, and records the result of the behavior planning at this time for use by other AGVs when planning behaviors.

7. The method of claim 1, wherein only when an AGV leaves or enters the multi-path lane, the AGVs inside the multi-path lane need to be selected in turn according to the priority, and the behavior planning is carried out.

8. The method of claim 1, wherein the path resource allocation is accomplished in a CPU at the server side.

Images