CN115016506A - Heterogeneous multi-agent network cooperative scheduling planning method based on autonomous obstacle avoidance - Google Patents

Abstract

The invention discloses a heterogeneous multi-agent networking cooperative scheduling planning method based on autonomous obstacle avoidance, which comprises the following steps: carrying out system configuration and transportation task management of multi-agent networking cooperative scheduling planning; each intelligent agent which is going to execute the transportation task in the system carries out collaborative path search to obtain a corresponding path list; the intelligent agent carries out motion planning to obtain an execution path; calculating to obtain a driving speed according to the execution path; the intelligent agent performs autonomous obstacle detouring in the operation process; and adjusting the time window. The invention realizes the cooperative solution of conflict between the intelligent agents and the autonomous obstacle avoidance by adopting a time window algorithm and introducing an autonomous obstacle avoidance technology of the intelligent agents, and compared with the prior art, the invention has the advantages of higher efficiency, flexibility and robustness, and more intelligent agents can be accommodated by the system.

Description

Heterogeneous multi-agent network cooperative scheduling planning method based on autonomous obstacle avoidance

technical field

The invention belongs to the technical field of multi-agent scheduling and path planning, and in particular relates to a heterogeneous multi-agent network-connected collaborative scheduling planning method based on autonomous obstacle avoidance technology. The distributed and integrated heterogeneous multi-agent network-connected collaborative scheduling planning method realizes efficient, flexible, robust and highly scalable heterogeneous multi-agent scheduling planning by exploring the autonomous intelligence capabilities of the agents, which can be used for Scheduling and path planning of multi-agents in complex scenarios such as smart logistics and smart warehousing.

Background technique

In recent years, due to the emergence of flexible manufacturing systems, the increasing demand for customized robots, and the increasing demand for industrial automation in small and medium-sized enterprises, mobile agents have developed rapidly in the fields of manufacturing, medical care, and intelligent logistics, and have been widely used.

In order to maximize production efficiency, users often expect multiple agents to run in the scene at the same time, which poses a challenge to the security of the agents in the scene, ensuring that multiple agents operate in the scene without conflict and efficiently. Performing transportation tasks efficiently is the primary problem that multi-agent systems need to solve. To accomplish this task is the scheduling and planning system, which needs to coordinate the operation of multiple agents in the scene, and improve the operating efficiency of the system as much as possible on the premise of ensuring safety.

At present, the prior art mainly adopts two scheduling planning systems with different architectures, namely, a central architecture system and a distributed architecture system. Compared with the centralized architecture system, the distributed architecture system has the advantages of strong flexibility and scalability. However, in this architecture system, each agent "governs its own affairs", which lays a hidden danger for the security of the system. And it is not conducive to the user's unified management and control of the agents in the scene, so the current distributed architecture system is rarely used in real environments such as smart logistics and smart warehousing. The central architecture system is currently widely used. It has the advantages of being easy to control and easy to obtain the optimal solution. However, the intelligent capabilities of the agents in the existing central architecture system are not fully utilized, and the control center needs to complete all calculations in the system. task, and all the agents in the system need to exchange information with the control center continuously, which causes the control center of the existing central architecture system to have high computing pressure and heavy network communication burden, which further limits the number of agents that the system can accommodate. Moreover, this system has the defect of "single point of failure", and the flexibility and robustness are poor.

SUMMARY OF THE INVENTION

The invention provides a heterogeneous multi-agent network-connected collaborative scheduling planning method based on autonomous obstacle circumvention technology, which integrates central and distributed integration. The advantages of the distributed architecture realize a more flexible, efficient, robust and heterogeneous scheduling planning method that can accommodate a larger number of agents.

The invention includes a control center, an intelligent body and a data sharing terminal. The intelligent body refers to an intelligent robot that performs transportation tasks in an application scenario. The algorithm part of the control center is deployed on a computer, and the data sharing terminal is deployed on a server. Among them, task assignment and cooperative path search are performed by the control center, the task assignment adopts the basic nearby assignment logic, and the cooperative path search is realized based on the Dijkstra algorithm. If the paths of the two agents contain the same road segment, the road segment is a duplicate road segment of the two agents. If the paths of the two agents both contain the same road segment and the directions passing through the road segment are opposite , then the road section is a repeated road section and an opposite driving section of the two agents; the agent uses the existing time window algorithm to plan the motion according to the result of the path search of the control center and follows the path. When the agent encounters an obstacle , the agent uses the existing local path planning algorithm carried by itself to bypass the obstacle, and after the obstacle is circumvented, the time window check and adjustment strategy designed by the present invention is executed to ensure the collaborative relationship between the agents in the scene in the subsequent operation process; The data sharing terminal is used to store the time window data structure required for collision avoidance between agents, and each agent can independently access or modify the data stored in the data sharing terminal.

The method of the invention mainly includes five parts: a task management part, a cooperative path search part, a motion planning part, an autonomous obstacle circumvention part and a time window adjustment part; wherein the task management part and the cooperative path search part are realized in the control center, and the motion planning part , The autonomous obstacle circumvention part and the time window adjustment part are implemented separately at each agent end.

The task management part realizes the dynamic addition of tasks and task allocation; the collaborative path search part is realized based on the Dijkstra algorithm, and at the same time, the safety hazards brought by the repeated sections and the opposite driving sections in the path for the driving of the intelligent agent are considered, so that the method of the present invention is an intelligent agent planning. The resulting path is safer and more reliable; in the motion planning part, each agent uses the time window algorithm for each path according to the results of the cooperative path search of the control center to avoid conflicts with other agents in the scene, and obtain specific speed. The present invention innovatively introduces the autonomous planning of the agent into the multi-agent scheduling and planning method and the multi-agent cooperative application system. When the agent encounters an obstacle during driving, the agent adopts the The local path planning algorithm of the autonomous obstacle avoidance part performs autonomous obstacle avoidance; autonomous obstacle avoidance will break the driving behavior planned by the motion planning part, so the time window adjustment part needs to check and adjust the time window after the autonomous obstacle avoidance.

The heterogeneous multi-agent networked collaborative scheduling planning method based on the autonomous obstacle circumvention technology provided by the present invention includes the following specific steps:

Step 1: Carry out system configuration and transportation task management of multi-agent network-connected collaborative scheduling planning;

The system includes a control center, an agent and a data sharing terminal; the task management part is used to add transportation tasks to the system and assign tasks.

The agent is an intelligent robot that performs transportation tasks in the scene; the control center is used for task assignment and collaborative path search; the data sharing terminal is used to store the time window data structure required for collision avoidance between agents; each agent can be accessed independently Or modify the data stored in the data sharing terminal.

The configuration of the multi-agent networked cooperative scheduling planning system specifically includes the startup of the control center and each agent, the setting of the scene map in the control center and each agent, and the initialization of the time window data structure of the data sharing terminal. The scene map includes node information and road section information. Nodes need to be positioned at fixed intervals (such as 10% of the scene width) according to the length and width of the scene, and some key points in the scene (such as pickup points, delivery points, etc.) It should also be set as a node. If there is an existing driving route in the scene, the position of the node should be set at a fixed interval on the existing driving route; a road segment is a line segment formed by two adjacent nodes. The road segments are connected to form a path.

The transportation task can take any node in the map as the task end point, and the position of the corresponding agent performing the task as the task start point; during the specific implementation, the task assignment can be carried out according to the order in which the tasks are added.

Step 2: Collaborative path search part - perform a collaborative path search for each agent that is about to perform a task, and obtain a corresponding path list.

Based on the dijkstra algorithm, all loop-free paths between the task starting point and the task ending point are obtained and stored in the path list PATH. At the same time, the present invention considers the safety hazards brought by repeated road sections and opposite driving sections for the driving of the agent. If the current agent If the paths of the current agent and other agents contain the same road segment, the road segment is a duplicate road segment of the current agent. If the paths of the current agent and other agents contain the same road segment and the direction of passing through the road segment is opposite, then This road segment is the current agent’s repeating road segment and the opposite driving road segment; in the specific implementation, check each path in the path list PATH, and set a certain path in the path list as path ₀ , if path ₀ is associated with any other agent The ratio of the number of repeated segments of the path being executed to the total number of segments on path ₀ is greater than the set threshold (such as 0.4) or the ratio of the number of opposite travel segments of the path being executed by any other agent to the total number of segments on path ₀ . If the ratio is greater than the set threshold (such as 0.3), then path ₀ is an unavailable path, and path ₀ is removed from the path list PATH.

Step 3: Motion planning part - execute motion planning to get the execution path;

The agent obtains the result of the cooperative path search performed by the control center and uses the existing time window algorithm for motion planning, performs time window insertion operation on each path, and selects the path that can complete the task earliest as the execution path.

Step 4: Motion planning part - calculating the driving speed according to the execution path;

For the execution path obtained in step 3, according to the length of each road segment on the path and the time window insertion result, the driving speed on the road segment is calculated, and sent to the chassis of the agent, and the agent travels at this speed.

Step 5: Autonomous Obstacle Deviance Part - During the operation of the agent, if the agent encounters an obstacle, the agent will execute the obstacle avoidance operation according to the local path planning algorithm of the autonomous obstacle avoidance part. The local path planning algorithm adopts the existing dynamic time window algorithm (Dynamic Window Approach, DWA).

Step 6: Time window adjustment part - Since the time for the agent to bypass the obstacle autonomously cannot be determined, after the autonomous obstacle avoidance of the agent is completed, the time window inspection and adjustment strategy designed by the method of the present invention needs to be adopted to ensure the subsequent operation in the scene. Collaborative relationship between agents.

Through the above steps, multiple agents can flexibly and efficiently perform tasks in the scene.

Compared with the prior art, the beneficial effects of the present invention:

The invention provides a heterogeneous multi-agent network-connected collaborative scheduling planning method based on autonomous obstacle avoidance technology, which integrates central and distributed integration. Compared with the traditional central architecture system, this method is more efficient, flexible, and robust, and the system can accommodate a larger number of agents.

The heterogeneous multi-agent networked cooperative scheduling planning method provided by the present invention has the following advantages:

1) The autonomous local path planning of the agent is introduced, which significantly reduces the computational pressure of the control center.

2) The agent has the ability to bypass obstacles autonomously, and the control center does not need to monitor the running status of the agent at all times, which significantly reduces the burden of network communication.

3) The reduction of the control center's computational pressure and network communication burden further promotes the increase in the number of agents that the system can accommodate.

4) The operation of the system no longer only depends on the control center, that is, the defect of the "single point of failure" of the traditional centralized architecture system is eliminated, and the system is more robust.

5) The agent has the ability to autonomously respond to emergencies and abnormal situations, which is more flexible than the traditional central architecture system where the control center responds to emergencies through re-planning.

Description of drawings

FIG. 1 is a system architecture adopted for the specific implementation of the method of the present invention, including a control center, an intelligent body and a data sharing terminal.

Fig. 2 is a flow chart of the task assignment method of the task management part of the present invention.

FIG. 3 is a schematic diagram of an example used in explaining the specific implementation of the present invention.

FIG. 4 is an example of the time window data structure of the data sharing side of the present invention.

Fig. 5 is the algorithm flow chart of the cooperative path search part of the present invention to check the path

FIG. 6 is an example of the time window insertion operation performed by the motion planning part of the present invention.

FIG. 7 is a schematic diagram of the velocity sampling space of the local path planning part of the present invention.

FIG. 8 is a schematic diagram of an obstacle bypass path autonomously planned by a part of the intelligent body for autonomous obstacle bypass according to the present invention.

Detailed ways

In order to make the above objects, features and advantages of the present invention more clearly understood, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

The present invention provides a heterogeneous multi-agent network-connected collaborative scheduling planning method based on autonomous obstacle bypassing technology, which integrates central and distributed integration, and realizes efficient, flexible, and seamless operation of multiple agents in scenarios such as smart logistics and smart warehousing. Conflicting tasks are performed, and the system architecture adopted for the specific implementation of the method of the present invention is shown in FIG. 1 . The method of the invention mainly includes five parts: a task management part, a cooperative path search part, a motion planning part, an autonomous obstacle avoidance part and a time window adjustment part; wherein the cooperative path search part fully considers the repeated road sections and the opposite driving section as the intelligent body The potential safety hazard brought by driving makes the path planned by the system for the agent more safe and reliable; in addition, the method of the present invention is a heterogeneous method integrating a central architecture and a distributed architecture, making full use of the autonomous intelligence of the agent Compared with the traditional centralized architecture system, the flexibility, robustness and scalability have been greatly improved.

The task management part realizes dynamic addition of tasks and assignment of tasks. The strategy flow of task assignment is shown in Figure 2. The collaborative path search is partially implemented based on the Dijkstra algorithm, and at the same time, the safety hazards brought by the repeated sections and opposite driving sections in the path are considered for the driving of the agent, so that the path planned by the system for the agent is safer and more reliable. The motion planning part realizes the avoidance of conflicts between agents. Each agent accesses the data sharing terminal and performs time window insertion operations to obtain execution instructions specific to the speed level. The autonomous obstacle avoidance part realizes the autonomous avoidance of the agent when it encounters an obstacle, that is, a local path planning algorithm is used to plan a route around the obstacle. The time window adjustment part realizes the inspection and adjustment of the time window after the autonomous obstacle avoidance is completed, so as to ensure the cooperative relationship between the agents in the scene during the subsequent operation.

For ease of understanding, when introducing the specific implementation of the method of the present invention, the scene shown in FIG. 3 will be taken as an example. As can be seen from the figure, there are 4 agents in the scene, namely AGENT ₁ , AGENT ₂ , AGENT ₃ and AGENT ₄ , of which only AGENT ₃ is idle.

The specific implementation of the method of the present invention comprises the following steps:

Step 1: Configure the multi-agent networked collaborative scheduling planning system, the system includes a control center, an agent and a data sharing terminal; task management part - adding tasks to the system and assigning tasks.

The agent is an intelligent robot that performs transportation tasks in the scene; the control center is used for task assignment and collaborative path search; the data sharing terminal is used to store the time window data structure required for collision avoidance between agents; each agent can be accessed independently Or modify the data stored in the data sharing terminal.

The configuration of the multi-agent networked cooperative scheduling planning system specifically includes the startup of the control center and each agent, the setting of the scene map in the control center and each agent, and the initialization of the time window data structure of the data sharing terminal. The startup of the control center and each agent starts the control and deployment of the computer and each agent in the algorithm part of the control center; the setting of the scene map in the control center and each agent refers to the input of the same map file to the control center and each agent to ensure that The control center and each agent use the same scene map; the time window data structure maintains a time window vector for each road segment in the map, and each element in the time window vector corresponds to a period of time that an agent occupies the road segment. The time and time window data structure is empty. Figure 4 shows an example of the time window data structure during the system operation.

Add a task with any node in the map as the end point to the system, and assign tasks according to the order in which the tasks are added. Referring to Figure 2, the specific task assignment strategy is:

1) If there is no idle agent in the current system, the assignment of the task is suspended;

2) If there is an idle agent in the current system, the task is assigned to the agent whose current position is closest to the end point of the task according to the principle of proximity.

Taking the scenario shown in FIG. 3 as an example, if a task whose end point is node 21 is added, the task will be assigned to AGENT ₃ for execution according to the task assignment policy.

Step 2: Collaborative path search part - perform a collaborative path search for each agent that is about to perform a task, and obtain a corresponding path list.

The execution steps of the basic Dijkstra algorithm are as follows:

1) Obtain the task start point s and the task end point e;

2) Define two sets: S and U. S is used to record all nodes in the map for which the shortest path has been found, and U is used to record the nodes for which the shortest path has not yet been found. Define two vectors: D and P. The vector D stores the shortest distance from the starting point s to other nodes in the map in the node number order, and the vector P stores the shortest route from the task starting point s to another node z in the map in the node number order. The previous node of node z in the middle. Initially, there is only the task starting point s in S, and U contains all the nodes in the map except the task starting point s; the value of each element in D is the distance from the task starting point s to other nodes in the map (if a node v in U If it is not adjacent to the task starting point s, the corresponding element value in D is ∞);

3) Select the node k closest to the task starting point s from U, add node k to S, and remove node k from U at the same time;

4) Update the distance from each node in U to the starting point s (considering that the distance from the starting point s of the node k as the intermediate point to a node v in U may decrease), and update the vector D and the vector P at the same time;

5) Repeat steps 3) and 4) until all nodes in the map are traversed;

6) Starting from the end point e, reversely retrieve the vector P to obtain the shortest path from the start point s to the end point e.

本发明基于dijkstra算法进行协同路径搜索的步骤如下：Assuming that there are A agents in the system, except for the current agent, the _i -th agent in the other A-1 agents is currently executing a path length of li (that is, there are _li nodes on the path), then the other A- The path being executed by the ith agent in one agent can be expressed as

The steps of the present invention based on the Dijkstra algorithm to search for a collaborative path are as follows:

1) Use the basic Dijkstra algorithm to find the shortest path between the two points of a task starting point and a task ending point: s→k ₁ →k ₂ →…→k _n →e; where k ₁ ～k _n are all on the path node;

2) Let k ₀ =s, k _n+1 =e, for each node k _m (m=0, 1, 2..., n) on the path except the task end point e, let k _m have r adjacent nodes in total. Node, traverse each node _{v q} (q=1, 2,..., r) adjacent to km; if v _q ≠km ₊₁ , take v _q as the starting point of the task, and take e as the end of the task, using the basic Find the shortest path between v _q and e using the Dijkstra algorithm: v _q →n ₁ →n ₂ →…→n _p →e, where n ₁ ～n _p are all nodes on the path, then another s and e are obtained. The path between: k ₀ →…→k _m →v _q →…→e, add it to the path list PATH.

3) After the execution of step 2), all loop-free paths between the task start point s and the task end point e are stored in the path list PATH. Suppose there are S paths in the path list PATH, and for each path PATH _z (z = 1, 2, ..., S), respectively use β _z and γ _z to record the maximum value of the ratio of the repeated sections of the path PATH _z and the paths being executed by other agents and the opposite driving sections to the total number of sections of the path PATH _z . The algorithm flow is shown in Figure 5.

和节点

连接而成的路段记为arc_h，及节点

和节点

连接而成的路段记为arc_w，若arc_h＝arc_w，则路径PATH_z与路径N_i的重复路段数累加1，若arc_h＝arc_w且

则路径PATH_z与路径N_i的对向行驶路段数也累加1。对于其他智能体当前执行路径的集合N，求得路径PATH_z与其中A-1条路径重复路段、对向行驶路段的最大值，进而求得β_z与γ_z，若β_z≥0.4或γ_z≥0.3，则认为路径PATH_z与当前系统中正在执行任务的其他智能体有较大的产生冲突的风险，则将路径PATH_z从路径列表中移除。Let the length of the path PATH _z be l (that is, there are l nodes on the path PATH _z ), for h=1, 2, ..., l-1, check the paths N _i of other A-1 agents in the system except the current agent ( _i =1, 2,...A-1), then for _w =1, 2,..., li -1, _li is the length of the path Ni (that is, there are _li nodes on the path); from the map information Query by node respectively

and node

The connected road segment is recorded as arc _h , and the node

and node

The connected road section is denoted as arc _w , if arc _h =arc _w , then the number of repeated road sections of path PATH _z and path _Ni is accumulated by 1, if arc _h =arc _w and

Then, the number of opposite travel sections of the path PATH _z and the path N _i is also accumulated by 1. For the set N of current execution paths of other agents, obtain the maximum value of the path PATH _z and the repeated sections and opposite driving sections of the A-1 paths, and then obtain β _z and γ _z , if β _z ≥ 0.4 or γ _z ≥ 0.3, it is considered that the path PATH _z has a greater risk of conflict with other agents that are executing tasks in the current system, and the path PATH _z is removed from the path list.

In the scenario shown in Figure 3, it is assumed that the current execution paths of AGENT ₁ , AGENT ₂ and AGENT ₄ are: 10→3→24→12→11→18→19, 19→15→13→14 and 16→21 →26→23→24→25, the path list PATH _E planned for AGENT ₃ includes the following three paths:

①22→24→3→2→9→16→21

②22→18→14→13→12→19→24→23→26→21

③22→24→12→11→18→23→26→21

After the path inspection of these three paths, there is only one feasible path in the path list PATH _E , that is, path ① and path ② are removed because the opposite driving section and the total path between them and the path of AGENT ₄ are. The ratio of the number of road segments is greater than 0.3, and the reason for the removal of route ③ is that the ratio of the duplicate road segments between it and the route of AGENT ₁ to the total number of road segments of the route is greater than 0.4.

Step 3: Motion planning part - the agent obtains the result of the collaborative path search performed by the control center (the path list obtained in step 2) and uses the existing time window algorithm for motion planning, and performs a time window insertion operation for each path. The path that can complete the task earliest is selected as the execution path.

After the control center sends the results of the collaborative path search to the agent, the agent first needs to access the data sharing terminal to obtain the current time window structure. The time window data structure maintains a time window vector for each road segment in the map. An element corresponds to a period of time that an agent occupies the road segment, and then for each path received by the agent, the following operations are performed:

For each segment α _m (m=1, 2, . . . , n) on each path:

1) Insert the time window with the shortest length (the length of the shortest time window is the ratio of the length of the road segment a _m to the maximum driving speed of the agent). Finds the insertion position that satisfies both of the following conditions:

① In the time window vector of road segment a _m , if the difference between the start time of the next time period of a certain position and the end time of the previous time period of the position is enough to accommodate the length of the time window of the shortest length, then the A position can be used as an insertion position.

②If m≠1, the time to enter this road segment should be after the time when the agent releases the previous road segment on its path.

2) If m≠1, judge whether the time for the agent to release the previous section on its path is equal to the time for entering this section. If they are not equal, extend the release time of the previous section to make it equal to the entry time of this section.

3) Check for overlap between time windows. If there is no overlap, the insertion is successful, and the operation is continued to the next segment. If it overlaps, go back to the previous segment on the route and go back to step 1).

The time window insertion operation on a path is specifically, a path contains multiple road _segments am , and the time window insertion operation of this path is successful after finding the insertion position that satisfies the condition in the time window vector for all the road segments on the path.

After the time window insertion operation on a path is successful, the time when the agent releases the last segment on the path is the estimated time when the agent uses the path to perform the task. The time window insertion operation is performed on each path received by the agent, and then a path that is expected to complete the task at the earliest is selected as the final path for the agent to perform the task.

By performing the time window insertion operation on each path received by the agent, after selecting the final execution path, the agent needs to send a message to the data sharing side, and update the time window insertion result of the selected execution path to the data sharing side synchronously .

Still taking the scene shown in Figure 3 as an example, in the collaborative path search part in step 2, it is found that there is only one candidate path for AGENT ₃ , that is, 22→24→3→2→9→16→21, and the sections on the path are set as follows: a ₉ →a ₁₆ →α ₇ →α ₁ →a ₁₇ →a ₂₂ , suppose that after AGENT ₃ receives the candidate path sent by the control center, before performing motion planning (that is, performing the time window insertion operation), the time of the above road segment The window data structure is shown in (1) in FIG. 6 , and after the motion planning is performed, the time window data structure of the above road section is shown in (2) in FIG. 6 .

Step 4: Motion planning part - for the execution path obtained in step 3, according to the length of each road section on the path and the time window insertion result to calculate the driving speed on this road section, and send it to the chassis of the agent, intelligent The body travels at this speed.

After the agent completes the time window insertion operation according to the current time window structure, it can obtain the length of each section on the execution path and the travel time on the section, and further calculate that the agent should travel on each section in the execution path. The speed of the intelligent body can be sent to the chassis actuator of the intelligent body to realize the tracking driving of the intelligent body.

Step 5: The part of autonomous obstacle avoidance - when the agent encounters an obstacle, the agent uses the local path planning algorithm to autonomously avoid the obstacle.

When the agent encounters an obstacle in front of the driving process, different from the traditional central architecture system that only relies on the control center for re-planning, the agent in the method of the present invention will use the local path planning algorithm to autonomously go around. The target point of autonomous obstacle avoidance is the next node on the running path of the node where the agent is currently located. In particular, when the next node on the running path is the last node of the path, the target point of autonomous obstacle avoidance is the task end. During implementation, the method of the present invention specifically adopts an existing dynamic time window algorithm (Dynamic Window Approach, DWA), which is a dynamic decision-making algorithm based on a limited observation range, and the algorithm comprises three steps:

1) Speed sampling

The speed sampling space of the agent is mainly limited by the following factors: the maximum speed and the minimum speed limit of the agent itself; the limit of the maximum acceleration and maximum deceleration of the agent, as shown in Figure 7, which leads to the agent in the following The speed that can be achieved in one forward simulation cycle is limited to a dynamic time window.

2) Trajectory scoring

In the sampled velocity group, several groups of trajectories are feasible, so the evaluation function is used to evaluate each trajectory. The evaluation function used is as follows:

H(u,ω)=α·goal(v,ω)+β·dist(v,ω)+γ·path(v,ω) (4.5)

Among them, goal(v, ω) is the target evaluation function, which is used to evaluate the position of the robot when it reaches the end of the simulated trajectory and the position of the target point, and the orientation angle and target orientation when it reaches the end of the simulated trajectory under the currently set sampling speed. The gap between the angles, the larger the gap, the lower the score; dist(v, ω) is the distance evaluation function, which is used to evaluate the distance between the trajectory and the nearest obstacle, and the smaller the distance to the nearest obstacle, The larger this value is; path(v, ω) is the path evaluation function, which is used to evaluate the closest distance between the points on the sampling trajectory and the global path. This evaluation function is mainly used to guide the agent to search according to the global path as much as possible. Proceed to reach the target point earlier.

It should be noted that since the parameters of the three evaluation functions such as goal(v, ω), dist(v, ω), and path(v, ω) are different, before calculating the final evaluation value H(v, ω), it is necessary to The evaluation function values of the three parts are normalized, and then multiplied by the corresponding coefficients α, β, and γ, respectively, and then summed to obtain the score of the trajectory of this group of speeds in the next forward simulation period. In the trajectory, a set of speeds with the highest score is selected as the execution speed of the agent in the next cycle.

3) Speed release

After sampling multiple sets of velocities in the velocity space and simulating the trajectories of multiple sets of velocities in the next forward simulation cycle, all trajectories are scored and the trajectory with the highest score is selected, and then the velocity corresponding to the trajectory with the highest score is sent to the intelligent The agent will drive at this speed in the next cycle, and the above process will be repeated until the agent reaches the target point.

Step 6: Time window adjustment part - after the agent has completed the obstacle avoidance, the time window checking and adjustment strategy designed by the method of the present invention is adopted to ensure the cooperative relationship between the agents in the scene in the subsequent running process. .

Since the autonomous obstacle avoidance of the agent is a time-consuming operation and the required time cannot be estimated in advance, if the agent has not reached the end of the task after the autonomous obstacle avoidance, it is necessary to check the time consumed by the obstacle after the obstacle avoidance. If the time consumed exceeds half of the time window protection time, the time window needs to be adjusted. Let the time spent on autonomously circumventing obstacles be δ, and the time window protection time is w (the time window protection time refers to the minimum time difference between their time windows if two agents both need to pass through a road section). The strategy is as follows:

1)

In this case, the time consumed by autonomous obstacle avoidance is less than or equal to half of the time window protection time. In the subsequent driving, the agent can make up for the time consumption by adjusting the driving speed, so there is no need to adjust the time window.

2)

In this case, the time consumed by autonomous obstacle avoidance exceeds half of the time window protection time. If only relying on the agent to compensate for the time error by adjusting the speed in the subsequent driving, there will be a greater risk of conflict, so this It is necessary to adjust the time window and re-plan the motion.

The specific method of adjusting the time window and re-planning the motion is as follows:

1) Access the data sharing terminal and clear the time window data of the current agent.

2) Obtain the time window structure after clearing the current agent time window data.

3) According to the current position of the agent, go to step 3 to determine the subsequent driving path of the agent and re-plan the motion.

所以接下来AGENT₃应清除当前时间窗数据结构中自身的时间窗，而后对后续行驶路径(3→2→9→16→21)重新进行运动规划。Still taking the scene shown in Figure 3 as an example, when AGENT ₃ travels to the location of node 24, it detects that there is an obstacle on the driving path ahead, and the agent uses the local path planning algorithm in step 5 to autonomously circumvent the obstacle, as shown in Figure 8 It is a kind of obstacle bypass path that AGENT ₃ may plan; when AGENT ₃ reaches the location of node 3 after the obstacle bypass, if the obstacle bypass consumes an additional 5.2s, that is, δ=5.2, set the time window protection time to 8s, that is, w = 8, since

Therefore, AGENT ₃ should clear its own time window in the current time window data structure, and then perform motion planning again for the subsequent travel path (3→2→9→16→21).

It should be noted that the purpose of publishing the embodiments is to help further understanding of the present invention, but those skilled in the art can understand that various replacements and modifications are possible without departing from the spirit and scope of the present invention and the appended claims of. Therefore, the present invention should not be limited to the contents disclosed in the embodiments, and the scope of protection of the present invention shall be subject to the scope defined by the claims.

Claims (5)

1. A heterogeneous multi-agent networking cooperative scheduling planning method based on autonomous obstacle avoidance is characterized by comprising the following steps:

step 1: carrying out system configuration and transportation task management of multi-agent networking cooperative scheduling planning;

the system comprises a control center, an intelligent agent and a data sharing end; the intelligent agent is an intelligent robot for executing a transportation task in a scene; the control center is used for performing task allocation and collaborative path search; the data sharing end is used for storing a time window data structure required by collision avoidance between the intelligent agents; each agent can independently access or modify the data stored in the data sharing end;

the system configuration of the multi-agent networking cooperative scheduling planning comprises the following steps: starting a control center and each intelligent agent, setting a scene map in the control center and each intelligent agent and initializing a time window data structure of a data sharing end;

the scene map comprises nodes and road sections; setting the key points of the scene as nodes; the key points of the scene comprise a goods picking point and a goods throwing point; in addition, the positions of the nodes are set at fixed intervals according to the length and the width of the scene; if the existing driving route exists in the scene, setting the positions of the nodes at fixed intervals on the existing driving route; the road sections of the scene are line sections formed by two adjacent nodes, and a plurality of road sections are connected to form a path;

the transportation task management is to point to a system to add a transportation task and distribute the task; the transportation task takes any node in the scene map as a task end point and takes the position of a corresponding agent executing the task as a task starting point;

step 2: each intelligent agent which is about to execute the transportation task in the system scene carries out collaborative path search to obtain a corresponding path list; the method comprises the following steps:

21) and (3) solving the shortest path between a task starting point and a task end point, and recording as: s → k ₁ →k ₂ →…→k _n → e; wherein s is a task starting point; e is a task end point; k is a radical of ₁ ～k _n All nodes are nodes on the path;

22) let k ₀ ＝s，k _n+1 E, for each point k on the path except the task end point e _m Where m is 0,1,2 …, n, k _m Total r neighboring nodes, traverse k _m Each adjacent node v _q Q is 1,2, …, r; if v is _q ≠k _m+1 In v with _q As a task starting point, and e as a task end point, v is obtained _q Shortest path to e: v. of _q →n ₁ →n ₂ →…→n _p →e，n ₁ ～n _p If all the nodes are nodes on the path, obtaining another path between s and e: k is a radical of ₀ →…→k _m →v _q → … → e, add it to the path list;

23) step 22), after the execution is finished, all loop-free paths from the task starting point s to the task ending point e are stored in the path list; for each path, calculating and recording the maximum value of the ratio of repeated road sections and opposite driving road sections of the path and the paths being executed by other agents to the total number of the road sections of the path;

the repeated section means that: if the paths of the current agent and other agents contain the same section of road, the road is the repeated section of the current agent;

the repeated section and the opposite direction driving section refer to: if the paths of the current agent and other agents comprise the same section of road and the direction of the road is opposite, the road is the repeated road and the opposite driving road of the current agent;

the method comprises the following steps:

231) let the PATH list PATH have S PATHs in total, and for each PATH _z 1,2, …, S, respectively, beta _z And gamma _z To record the PATH _z PATH for overlapping road section and opposite driving road section of PATH being executed by other agent _z The maximum value of the ratio of the total number of stages;

232) path setting PATH _z Has a length of l, i.e. the PATH _z The number of the nodes is l; for h-1, 2, …, l-1, a-1 agent paths N in the examination system in addition to the current agent _i I 1,2, … a-1, then w 1,2, …, l _i -1，l _i Is path N _i Respectively finding the route nodes from the map

And node

The link formed by the connection is denoted arc _h And a routing node

And node

The road section formed by the connection is denoted as arc _w ；

233) If arc _h ＝arc _w Then PATH _z And path N _i 1 is accumulated in the number of the repeated sections; if arc _h ＝arc _w And is provided with

Then PATH _z And path N _i The number of opposite driving sections is also accumulated to be 1;

234) solving PATH for the set N of current execution PATHs of other agents _z Calculating beta according to the maximum value of the repeated road section of the A-1 paths and the maximum value of the opposite driving road section _z And gamma _z (ii) a For beta is _z And gamma _z Respectively setting threshold values; if beta is _z Exceeding the corresponding threshold value or gamma _z If the corresponding threshold value is exceeded, the PATH PATH is set _z Removing from the path list;

and step 3: the intelligent agent carries out motion planning to obtain an execution path;

according to a path list result obtained by collaborative path search, performing motion planning by adopting a time window algorithm, performing time window insertion operation on each path, and selecting a path capable of completing a task earliest as an execution path;

and 4, step 4: calculating to obtain a driving speed according to the execution path;

calculating the running speed on each road section according to the length of each road section on the execution path obtained in the step 3 and the time window insertion result, and sending the running speed to a chassis of the intelligent agent, wherein the intelligent agent runs according to the speed;

and 5: the intelligent agent carries out autonomous obstacle avoidance in the operation process: if the intelligent agent meets the obstacle, the intelligent agent bypasses the obstacle according to a local path planning algorithm and reaches an autonomous obstacle-bypassing target point; the target point of the autonomous obstacle detouring is the next node of the node where the intelligent agent is located on the operation path;

step 6, time window adjustment: after the autonomous obstacle avoidance of the intelligent agent is finished, if the intelligent agent does not reach a task end point, a time window checking and adjusting strategy is designed and adopted to ensure the cooperative scheduling among the intelligent agents in a scene in the subsequent operation process of the intelligent agent; the method comprises the following steps:

6A) when the time consumed by the intelligent agent autonomous obstacle avoidance is less than or equal to half of the time window protection time, the time window does not need to be adjusted; in the subsequent running process, the intelligent agent can compensate the consumption of time by adjusting the running speed;

the time window protection time refers to the time which is required to be separated from the time windows of two intelligent agents at least if the two intelligent agents need to pass through a road section;

6B) when the time consumed by the autonomous obstacle avoidance exceeds half of the time window protection time, the time window is adjusted according to the following method:

accessing a data sharing end, and clearing time window data of the current agent;

obtaining a time window structure after the current agent time window data is cleared;

turning to the step 3 to determine a subsequent driving path of the intelligent agent and carrying out motion planning again according to the current position of the intelligent agent;

through the steps, the heterogeneous multi-agent networking cooperative scheduling planning based on autonomous obstacle avoidance can be realized.

2. The heterogeneous multi-agent networking cooperative scheduling planning method based on autonomous obstacle avoidance as claimed in claim 1, wherein in step 234), β is set _z 0.4 of the threshold value; setting up gamma _z The threshold value of (2) is 0.3.

3. The heterogeneous multi-agent networking cooperative scheduling planning method based on autonomous obstacle avoidance as claimed in claim 1, wherein step 21) specifically uses dijkstra algorithm to find a shortest path between two points.

4. The heterogeneous multi-agent networking cooperative scheduling planning method based on autonomous obstacle avoidance as claimed in claim 1, wherein step 3 is to perform motion planning by using a time window algorithm and calculate to obtain an execution path; the method comprises the following steps:

firstly, the intelligent agent accesses a data sharing end to obtain a current time window data structure; the time window data structure maintains a time window vector for each road section in the map, and each element in the time window vector corresponds to a period of time for which an agent occupies the road section; then, for each path received by the agent, the following operations are executed:

for each segment of road a on each path _m ，m＝1,2,…,n：

31) Inserting with a time window of shortest length; the shortest length of the time window is the road section a _m The length of the intelligent agent and the maximum driving speed of the intelligent agent; finding an insertion position satisfying the following two conditions:

(ii) on a road section a _m If the difference between the starting time of the time period after a certain position and the ending time of the time period before the position is enough to accommodate the length of the time window with the shortest length, the position can be used as an insertion position;

if m is not equal to 1, the time for entering the section of road section is after the time for releasing the section of road section on the path of the intelligent agent;

32) if m is not equal to 1, judging whether the time for releasing the last section of road section on the path of the intelligent agent is equal to the time for entering the section of road section; if not, prolonging the release time of the previous section of road section to be equal to the entry time of the section of road section;

33) checking whether there is an overlap between the time windows; if the sections are not overlapped, the insertion is successful, and the next section of road section is continuously operated; if the road sections are overlapped, returning to the previous section of the road section on the path and returning to the step 31);

and after selecting the final execution path by performing time window insertion operation on each path received by the intelligent agent, the intelligent agent sends a message to the data sharing end, and the time window insertion result of the selected execution path is synchronously updated to the data sharing end.

5. The heterogeneous multi-agent networking cooperative scheduling planning method based on autonomous obstacle avoidance of claim 1, wherein in step 5, the agent performs autonomous obstacle avoidance in the operation process according to a local path planning algorithm; specifically, a dynamic time window algorithm DWA is adopted, and the method comprises the following steps:

51) sampling speed;

52) and (3) track scoring:

scoring each trajectory by using an evaluation function; the evaluation indexes include: the distance between the position of the tail end of the track and the position of the target point, the difference between the orientation angle when the tail end of the track is reached and the target azimuth angle, the distance between the track and the nearest obstacle, and the nearest distance between a point on the sampling track and the global path;

53) and (3) speed release:

scoring all tracks and selecting the track with the highest score;

then sending the speed corresponding to the track with the highest score to the intelligent agent;

the intelligent agent runs at the speed in the next period;

and repeating the process until the intelligent agent reaches the target point of autonomous obstacle avoidance.

Images