Multi-agent-based object transportation method and device

Abstract

This invention discloses a multi-agent object transportation method and apparatus. It establishes a scene environment in a simulation environment based on physical rules; based on path planning within the scene environment, it trains the individual decision network of a single agent to assess the agent's ability to reach a designated location within the scene environment; and in the scene environment, it trains the cooperative decision network of multiple agents through reinforcement learning to facilitate collaborative object transportation. The technical solution of this invention accelerates the training speed of reinforcement learning by separating motion control from top-level planning, enabling multi-agent tasks in complex, unstructured large-scale scenes. By integrating historical observations through recurrent neural networks and collecting the expected actions of neighboring agents through multi-head attention mechanisms, it achieves cooperative completion of transportation tasks under partially observed conditions. This enables multi-robot systems to cooperatively transport objects in complex, unstructured scenes.

Description

Technical Field

[0001] This invention relates to the field of task planning technology, and in particular to a method and apparatus for object transportation based on multiple agents. Background Technology

[0002] With the development of science and technology and industry, robots are playing an increasingly important role in human production and life. A key goal of artificial intelligence is to create intelligent agents with autonomous decision-making and behavioral capabilities. Compared to single-robot systems, multi-robot systems have attracted widespread attention due to their lower cost, stronger robustness, and higher scalability. As a well-known use case for multi-robot systems, cooperative object transport shows enormous application potential in the real world. Robots collaboratively push an object through an obstacle-filled environment, which has applications in warehouse automation and disaster relief. These tasks require multiple agents to coordinate their respective actions to achieve specific goals. In the box-pushing problem, the benefits of multi-robot collaboration are twofold: sometimes the box is too heavy for a single robot to push on its own; through cooperation and joint action, the total number of actions performed by the robots to move the box to the target location is greatly reduced.

[0003] Existing algorithms based on manually defined rules require specific scenario-based rule setting, resulting in insufficient generalization. Furthermore, they necessitate manual programming by professionals in new scenarios, hindering the widespread adoption of robot services. Centralized decision-making methods, on the other hand, are limited by communication bandwidth and other constraints in large-scale, multi-robot scenarios. None of these methods can conveniently and efficiently endow multiple robots with the ability to move objects in complex, unstructured environments. Summary of the Invention

[0004] One of the objectives of this invention is to overcome the shortcomings of the prior art. In view of the problem that existing algorithms for object transportation cannot conveniently and efficiently endow multiple robots with the ability to transport objects in complex unstructured scenarios, this invention provides an object transportation method and device based on multiple intelligent agents.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0006] In a first aspect, the present invention provides a multi-agent-based object transportation method, the method comprising:

[0007] S1: Establishing a scene environment in a simulation environment based on physical rules;

[0008] S2: Based on path planning in the scene environment, train the individual decision network of a single agent to assess the agent's ability to reach a specified location in the scene environment;

[0009] S3: In the described scenario, for the collaborative object transportation of multiple agents, reinforcement learning training is performed on the cooperative decision-making network of multiple agents.

[0010] In a preferred embodiment of this application, step S1 specifically includes:

[0011] S11: In a simulation environment, multiple intelligent agents and multiple obstacles are randomly set up;

[0012] S12: Set the target task associated with the plurality of obstacles;

[0013] S13: Based on the target task and physical rules, set physical functions for the multiple intelligent agents and set physical attributes for the multiple obstacles;

[0014] S14: Based on the physical function, establish a multi-agent decision-making model for the multiple agents to execute the target task;

[0015] S15: Based on the multiple intelligent agents, the multiple obstacles, the target task, and the multi-agent decision-making model, a scene environment is formed for the multiple intelligent agents to carry out collaborative object transportation.

[0016] In a preferred embodiment of this application, step S2 specifically includes:

[0017] S21: Establish a single-agent decision network to be trained to control individual intelligent agents;

[0018] S22: Randomly initialize the initial position of the agent and the position of the target point to be reached;

[0019] S23: Calculate the optimal path between the initial position and the target point position;

[0020] S24: Collect the agent's demonstration actions based on the optimal path;

[0021] S25: Use the optimal path and the corresponding demonstration action as training data for training the single-unit decision network to be trained;

[0022] S26: Pre-train the individual decision network to be trained based on the training data;

[0023] S27: Based on the pre-training, reinforcement learning algorithm is used to perform reinforcement learning on the individual decision network to be trained;

[0024] S28: After multiple rounds of reinforcement learning, a trained individual decision network is obtained, and the individual decision network is used to control the single agent to perform tasks.

[0025] In a preferred embodiment of this application, step S3 specifically includes:

[0026] S31: Establish a cooperative decision-making network to be trained to control multiple intelligent agents;

[0027] S32: Based on the cooperative decision-making network to be trained, make corresponding final action decisions for each agent;

[0028] S33: In different scenario environments, based on the final action decisions of each intelligent agent, the cooperative decision network to be trained is repeatedly trained until the requirements are met, and then the trained cooperative decision network is obtained.

[0029] In a preferred embodiment of this application, step S31 specifically includes:

[0030] S311: Based on the single-agent decision network, establish corresponding action decision networks for each agent and initialize the first network parameters, while establishing a unified evaluation network and initializing the second network parameters.

[0031] S312: The action decision network is used to plan the actions of the agent, thus forming the agent's actions;

[0032] S313: The value of each agent's actions is evaluated through the evaluation network.

[0033] In a preferred embodiment of this application, step S32 specifically includes:

[0034] S321: Collect comprehensive information observed by each agent and use it as input to the action decision network;

[0035] S322: By communicating with the nearest n agents, the hidden variables in the action decision networks corresponding to the n agents are extracted as the action intentions of the n agents respectively.

[0036] S323: Integrate the multiple action intentions and make a final action decision for the corresponding agent based on the action decision network;

[0037] S324: Control the movement of the agent based on the final action decision through the individual decision network corresponding to the agent.

[0038] In a preferred embodiment of this application, step S33 specifically includes:

[0039] S331: In a single scenario environment, control the multiple intelligent agents to execute the final action decision and form a cluster action;

[0040] S332: As the multiple intelligent agents act in clusters, the scene environment is updated accordingly, and the current scene information is placed into the memory pool;

[0041] S333: Update the first network parameters and the second network parameters using gradient descent.

[0042] S334: Repeatedly train the cooperative decision network to be trained in different scenario environments until the requirements are met, and then obtain the trained cooperative decision network.

[0043] Secondly, the present invention provides a multi-agent-based object transportation device, the device comprising an establishment module, a training module and a learning module connected to each other;

[0044] The establishment module is used to establish a scene environment in a simulation environment based on physical rules;

[0045] The training module is used to train a single agent's decision network based on path planning in the scene environment, targeting the agent's ability to reach a specified location in the scene environment.

[0046] The learning module is used to perform reinforcement learning training on the cooperative decision-making network of multiple agents for collaborative object transportation in the scenario environment.

[0047] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when run on a computer, causes the computer to perform the multi-agent-based object transport method as described in the first aspect.

[0048] Fourthly, the present invention provides a computer program product comprising a computer program that, when run on a computer, causes the computer to perform the multi-agent-based object transport method as described in the first aspect.

[0049] The multi-agent object transport method and apparatus disclosed in this invention accelerates the training speed of reinforcement learning by separating motion control from top-level planning, enabling multi-agent tasks in complex, unstructured large-scale scenes. By integrating historical observations through recurrent neural networks and collecting the expected actions of neighboring agents through multi-head attention mechanisms, cooperative transport tasks can be completed even under partially observable conditions. This gives multi-robot systems the ability to cooperatively transport objects in complex, unstructured scenes. Attached Figure Description

[0050] The present invention is described with reference to the following figures:

[0051] Figure 1This is a flowchart of the multi-agent-based object transportation method in Embodiment 1 of the present invention;

[0052] Figure 2 This is a flowchart of step S1 in the multi-agent-based object transportation method of Embodiment 1 of the present invention.

[0053] Figure 3 This is a schematic diagram of the scene environment established in step S1 in the multi-agent-based object transportation method of Embodiment 1 of the present invention.

[0054] Figure 4 This is a flowchart of step S2 in the multi-agent-based object transportation method of Embodiment 1 of the present invention.

[0055] Figure 5 This is a flowchart of step S3 in the multi-agent-based object transportation method of Embodiment 1 of the present invention.

[0056] Figure 6 This is a flowchart of step S31 in the multi-agent-based object transportation method of Embodiment 1 of the present invention.

[0057] Figure 7 This is a network structure diagram of the cooperative decision-making network in the multi-agent-based object transportation method of Embodiment 1 of the present invention;

[0058] Figure 8 This is a flowchart of step S32 in the multi-agent-based object transportation method of Embodiment 1 of the present invention.

[0059] Figure 9 This is a flowchart of step S33 in the multi-agent-based object transportation method of Embodiment 1 of the present invention.

[0060] Figure 10 This is a schematic diagram of the multi-agent-based object transportation device in Embodiment 2 of the present invention.

[0061] Figure label:

[0062] 100 - Module for building; 200 - Module for training; 300 - Module for learning. Detailed Implementation

[0063] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0064] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0065] Example 1

[0066] Embodiment 1 of the present invention discloses a multi-agent-based object transportation method, which enables a multi-robot system to learn to cooperate in unstructured scenarios to complete object transportation tasks.

[0067] See Figure 1 The multi-agent-based object transportation method in this embodiment 1 includes:

[0068] S1: Establishing a scene environment in a simulation environment based on physical rules;

[0069] S2: Based on path planning in the scene environment, train the individual decision network of a single agent to assess the agent's ability to reach a specified location in the scene environment;

[0070] S3: In the scene environment, for the collaborative object transportation of multiple agents, reinforcement learning training is performed on the cooperative decision network of multiple agents.

[0071] Specifically, the scenario environment is randomly established within a simulation environment. Each time the cooperative decision-making network is trained through reinforcement learning, the scenario environment is different; that is, the number, location, tasks performed, and movement trajectories of the agents in the scenario environment will not be exactly the same. This ensures that the training and reinforcement learning process closely approximates the various environments and variables encountered in real-world applications. An agent is a subject capable of making action decisions independently of human intervention, based on constantly changing information collected in an independent scenario environment; this is typically a robot. The individual decision-making network is the function that determines the action decisions of a robot acting as an independent agent. The cooperative decision-making network is the function that integrates the actions of a robot cluster of all agents in the scenario environment, determining the action decisions for collective action. Using the object transportation method in Embodiment 1, by combining action decisions made for individual robots and action decisions made for cooperation among multiple robots in multiple randomly established scenario environments, and through repeated training and reinforcement learning, a mature and complete model for controlling multi-agent cooperative object transportation is obtained. This endows the multi-robot system with the ability to cooperate to complete object transportation tasks in unstructured scenarios.

[0072] See Figure 2 In the object transportation method of this embodiment 1, step S1 specifically includes:

[0073] S11: In a simulation environment, multiple intelligent agents and multiple obstacles are randomly set up;

[0074] S12: Set the target task associated with multiple obstacles;

[0075] S13: Based on the target task and physical rules, set physical functions for multiple intelligent agents and physical attributes for multiple obstacles;

[0076] S14: Based on physical functions, establish a multi-agent decision-making model for multiple agents to execute the target task;

[0077] S15: Based on multiple agents, multiple obstacles, target tasks, and multi-agent decision-making models, a scenario environment is formed for multiple agents to cooperate in transporting objects.

[0078] Specifically, in step S1, the multi-agent decision-making method is modeled as a Dec-POMDP (Distributed Partially Observable Markov Decision Process) model, which is defined as a tuple.<I,S,A,T,Z,O,R> Where I = {1, ..., n} represents the combination of finite agents; n is the number of agents; S = {s1, ..., s...} n} is the set of finite states of the agent; Z = {z1, ..., z2} n} is the set of finite observations of the agent; A = {a1, ..., a n Let} represent the joint action of the agents; T be the state transition function, representing the probability that the agent cluster will execute the joint action in the current state to reach the next state; O be the observation function, representing the probability that the agent cluster will obtain a certain observation by executing the joint action in the current state; and R be the reward function, representing the reward obtained by the agent cluster by executing the joint action in the current state. In the simulation environment, a scene environment with random walls and other obstacles is established. A certain number of boxes and robots are randomly placed in the scene environment, and a fixed target task is set, namely, the robot controls the boxes to reach a certain position, such as... Figure 3 As shown, the white squares on the map represent passable roads, gray squares represent walls, black dots represent robot positions, and 2x2 grid shapes represent box locations. Figure 4 The target point for the box is defined by the surrounding area. Based on this, physical properties are assigned to obstacles according to physics rules, such as setting different weights for the boxes, i.e., how many robots are needed to push them. Physical functions are also assigned to the intelligent agent based on these physics rules, such as allowing the robot to observe its own position, the surrounding environment, and the positions of the boxes.

[0079] See Figure 4 In the object transportation method of this embodiment 1, step S2 specifically includes:

[0080] S21: Establish a single-agent decision network to be trained to control individual intelligent agents;

[0081] S22: Randomly initialize the agent's initial position and the target point position to be reached;

[0082] S23: Calculate the optimal path between the initial position and the target position;

[0083] S24: Collect demonstration actions of the agent based on the optimal path;

[0084] S25: Use the optimal path and the corresponding demonstration action as training data for training the individual decision network to be trained;

[0085] S26: Pre-train the individual decision network to be trained based on the training data;

[0086] S27: Based on pre-training, use reinforcement learning algorithms to perform reinforcement learning on the individual decision network to be trained;

[0087] S28: After multiple reinforcement learning iterations, a trained individual decision network is obtained, which is used to control a single agent to perform tasks.

[0088] Specifically, step S2 involves establishing a single-unit decision network to control the movement of a single robot. To ensure the model's generalization and robustness, in complex environments, the robot's initial position and the target point are randomly initialized. The optimal path is calculated using a traditional path planning algorithm, and demonstration actions for the robot are collected based on this optimal path as training data for imitation learning. The single-unit decision network is pre-trained using this training data through imitation learning, and subsequently, reinforcement learning is performed using the PPO algorithm (Proximal Policy Optimization Algorithms, deep reinforcement learning). After a period of training, the model can robustly control the robot to reach the designated position in unstructured complex environments.

[0089] See Figure 5 In the object transportation method of this embodiment 1, step S3 specifically includes:

[0090] S31: Establish a cooperative decision-making network to be trained to control multiple intelligent agents;

[0091] S32: Based on the cooperative decision-making network to be trained, make corresponding final action decisions for each agent;

[0092] S33: In different scenario environments, based on the final action decisions of each agent, the cooperative decision network to be trained is repeatedly trained until the requirements are met, and then the trained cooperative decision network is obtained.

[0093] See Figure 6 In the object transportation method of this embodiment 1, step S31 specifically includes:

[0094] S311: Based on the individual decision network, establish corresponding action decision networks for each agent and initialize the first network parameters, while establishing a unified evaluation network and initializing the second network parameters.

[0095] S312: The agent's actions are planned through the action decision network;

[0096] S313: Evaluate the value of each agent's actions by evaluating the network.

[0097] Specifically, in step S31, a centralized training and distributed execution approach is adopted to establish an independent action decision network π for each robot. i and initialize the first network parameter θ i Establish a unified evaluation network C and initialize the second network parameter σ, as follows: Figure 7 As shown, each action decision network plans the robot's actions based on the robot's current observations, and the evaluation network evaluates the value of the actions based on the actions of all robots.

[0098] See Figure 8 In the object transportation method of this embodiment 1, step S32 specifically includes:

[0099] S321: Collect comprehensive information observed by each agent and use it as input to the action decision network;

[0100] S322: By communicating with the nearest n agents, the hidden variables in the action decision network of the n agents are extracted as the action intentions of the n agents respectively.

[0101] S323: Integrates multiple action intentions and makes the final action decision for the corresponding agent based on the action decision network;

[0102] S324: Control the movement of the agent based on the final action decision through the individual decision network corresponding to the agent.

[0103] Specifically, in step S32, each robot collects its current observations, including its own position, the surrounding environment, and the position of the box. Since actions based solely on the current observation may not be correct, a recurrent neural network (RNN) is used to aggregate all observations from the start frame to the current frame, and this combined information is used as the action decision network π. iThe input is... To better coordinate cooperation among agents, communication between agents is necessary. Each robot communicates with the n nearest agents. Latent variables are extracted from the action decision network and considered as the action intentions of each robot. Through a multi-head attention mechanism, the action intentions of different robots are integrated to make the final action decision.

[0104] See Figure 9 In the object transportation method of this embodiment 1, step S33 specifically includes:

[0105] S331: In a single scenario environment, control multiple intelligent agents to execute the final action decision and form a cluster action;

[0106] S332: As multiple intelligent agents act in clusters, the scene environment is updated accordingly, and the current scene information is put into the memory pool;

[0107] S333: Update the parameters of the first and second networks using gradient descent.

[0108] S334: Repeatedly train the cooperative decision network to be trained in different scenario environments until the requirements are met, and then obtain the trained cooperative decision network.

[0109] Specifically, within a scene (episode), the robot, based on the action decision network θ i Action decisions are made, and the movement of each robot is controlled through the S2 individual decision network. The scene environment is updated based on the collective actions of multiple robot clusters, and the reward for the current step is generated. The observations, states, actions, and rewards for each step are placed into a memory buffer. When the number of trajectories in the memory buffer reaches a certain amount, gradient descent is used to adjust the network parameters θ. i Update σ. Repeat step S33 until the object transport model performs as required.

[0110] Example 2

[0111] Embodiment 2 of the present invention discloses a multi-agent-based object transportation device. By executing the multi-agent-based object transportation method of Embodiment 1 of the present invention, a multi-robot system can learn to cooperate in unstructured scenarios to complete object transportation work.

[0112] See Figure 10The multi-agent object transportation device of this embodiment 2 includes an interconnected establishment module 100, a training module 200, and a learning module 300. The establishment module 100 is used to establish a scene environment in a simulation environment based on physical rules. The training module 200 is used to train the individual decision network of a single agent based on path planning in the scene environment, targeting the agent's ability to reach a specified location in the scene environment. The learning module 300 is used to perform reinforcement learning training on the cooperative decision network of multiple agents in the scene environment, targeting the cooperative object transportation of multiple agents.

[0113] Specifically, the scenario environment is randomly established within a simulation environment. Each time the cooperative decision-making network is trained through reinforcement learning, the scenario environment is different; that is, the number, location, tasks performed, and movement trajectories of the agents in the scenario environment will not be exactly the same. This ensures that the training and reinforcement learning process closely approximates the various environments and variables encountered in real-world applications. An agent is a subject capable of making action decisions independently of human intervention, based on constantly changing information collected in an independent scenario environment; this is typically a robot. The individual decision-making network is the function that determines the action decisions of a robot acting as an independent agent. The cooperative decision-making network is the function that integrates the actions of a robot cluster of all agents in the scenario environment, determining the action decisions for collective action. Using the object transportation method in Embodiment 1, by combining action decisions made for individual robots and action decisions made for cooperation among multiple robots in multiple randomly established scenario environments, and through repeated training and reinforcement learning, a mature and complete model for controlling multi-agent cooperative object transportation is obtained. This endows the multi-robot system with the ability to cooperate to complete object transportation tasks in unstructured scenarios.

[0114] Example 3

[0115] Embodiment 3 of the present invention discloses a computer-readable storage medium storing a computer program that, when run on a computer, causes the computer to execute the multi-agent-based object transportation method disclosed in Embodiment 1 of the present invention.

[0116] Example 4

[0117] Embodiment 4 of the present invention discloses a computer program product, which includes a computer program that, when run on a computer, causes the computer to execute the multi-agent-based object transportation method disclosed in Embodiment 1 of the present invention.

[0118] The multi-agent object transport method and apparatus disclosed in this invention accelerates the training speed of reinforcement learning by separating motion control from top-level planning, enabling multi-agent tasks in complex, unstructured large-scale scenes. By integrating historical observations through recurrent neural networks and collecting the expected actions of neighboring agents through multi-head attention mechanisms, cooperative transport tasks can be completed even under partially observable conditions. This gives multi-robot systems the ability to cooperatively transport objects in complex, unstructured scenes.

[0119] It should be understood that the above description of specific embodiments of the present invention is only for illustrating the technical approach and features of the present invention, and is intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. However, the present invention is not limited to the specific embodiments described above. All changes or modifications made within the scope of the claims of the present invention should be covered within the protection scope of the present invention.

Claims

1. A multi-agent-based object transportation method, characterized in that, The method includes: S1: Establishing a scene environment in a simulation environment based on physical rules; S2: Based on path planning in the scene environment, train the individual decision network of a single agent to assess the agent's ability to reach a specified location in the scene environment; S3: In the scenario environment, for the collaborative object transportation of multiple agents, reinforcement learning training is performed on the cooperative decision-making network of multiple agents; Step S3 specifically includes: S31: Establish a cooperative decision-making network to be trained to control multiple intelligent agents; S32: Based on the cooperative decision-making network to be trained, make corresponding final action decisions for each agent; S33: In different scenario environments, based on the final action decisions of each intelligent agent, the cooperative decision network to be trained is repeatedly trained until the requirements are met and the trained cooperative decision network is obtained. Step S31 specifically includes: S311: Based on the single-agent decision network, establish corresponding action decision networks for each agent and initialize the first network parameters, while establishing a unified evaluation network and initializing the second network parameters. S312: The action decision network is used to plan the actions of the agent, thus forming the agent's actions; S313: The value of each agent's actions is evaluated through the evaluation network; Step S32 specifically includes: S321: Collect comprehensive information observed by each agent and use it as input to the action decision network; S322: By communicating with the nearest n agents, the hidden variables in the action decision networks corresponding to the n agents are extracted as the action intentions of the n agents respectively. S323: Integrate the multiple action intentions and make a final action decision for the corresponding agent based on the action decision network; S324: Control the movement of the intelligent agent based on the final action decision through the individual decision network corresponding to the intelligent agent; Step S33 specifically includes: S331: In a single scenario environment, control the multiple intelligent agents to execute the final action decision and form a cluster action; S332: As the multiple intelligent agents act in clusters, the scene environment is updated accordingly, and the current scene information is placed into the memory pool; S333: Update the first network parameters and the second network parameters using gradient descent. S334: Repeatedly train the cooperative decision network to be trained in different scenario environments until the requirements are met, and then obtain the trained cooperative decision network.

2. The multi-agent-based object transportation method according to claim 1, characterized in that, Step S1 specifically includes: S11: In a simulation environment, multiple intelligent agents and multiple obstacles are randomly set up; S12: Set the target task associated with the plurality of obstacles; S13: Based on the target task and physical rules, set physical functions for the multiple intelligent agents and set physical attributes for the multiple obstacles; S14: Based on the physical function, establish a multi-agent decision-making model for the multiple agents to execute the target task; S15: Based on the multiple intelligent agents, the multiple obstacles, the target task, and the multi-agent decision-making model, a scene environment is formed for the multiple intelligent agents to carry out collaborative object transportation.

3. The multi-agent-based object transportation method according to claim 1, characterized in that, Step S2 specifically includes: S21: Establish a single-agent decision network to be trained to control individual intelligent agents; S22: Randomly initialize the initial position of the agent and the position of the target point to be reached; S23: Calculate the optimal path between the initial position and the target point position; S24: Collect the agent's demonstration actions based on the optimal path; S25: Use the optimal path and the corresponding demonstration action as training data for training the single-unit decision network to be trained; S26: Pre-train the individual decision network to be trained based on the training data; S27: Based on the pre-training, reinforcement learning algorithm is used to perform reinforcement learning on the individual decision network to be trained; S28: After multiple rounds of reinforcement learning, a trained individual decision network is obtained, and the individual decision network is used to control the single agent to perform tasks.

4. A multi-agent-based object transportation device, characterized in that, The device includes an interconnected establishment module, a training module, and a learning module; it executes the multi-agent-based object transportation method as described in any one of claims 1 to 3. The establishment module is used to establish a scene environment in a simulation environment based on physical rules; The training module is used to train a single agent's decision network based on path planning in the scene environment, targeting the agent's ability to reach a specified location in the scene environment. The learning module is used to perform reinforcement learning training on the cooperative decision-making network of multiple agents for collaborative object transportation in the scenario environment.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when run on a computer, causes the computer to perform the multi-agent-based object transportation method as described in any one of claims 1 to 3.

6. A computer program product, characterized in that, The computer program product includes a computer program that, when run on a computer, causes the computer to perform the multi-agent-based object transport method as described in any one of claims 1 to 3.

Images