Method and apparatus for multi-agent path seeking within diagnostic laboratory systems

Abstract

In some embodiments, a method for path finding of sample containers in a diagnostic laboratory system includes receiving a batch of sample containers in a diagnostic laboratory system having a diagnostic laboratory device; obtaining a grid for the diagnostic laboratory system, the grid having units comprising diagnostic laboratory devices and one or more tracks connecting the diagnostic laboratory devices; distributing an intelligent agent to each sample carrier; determining a target for each agent assigned to the sample carrier holding the sample container, the target including an earliest arrival time constraint, a deadline constraint, and a target processing time constraint; and employing a multi-agent path finding (MAPF) trained neural network to determine the action of each agent, the neural network being trained with arrival, deadline and priority constraints, and non-instantaneous target processing times. Numerous other aspects are provided.

Description

[0001] Cross-reference to related applications This application claims the benefit of U.S. Provisional Application No. 63 / 518,565, filed August 10, 2023, pursuant to 35 USC § 119(e). The entire contents of the patent application(s) cited above are expressly incorporated herein by reference. Technical Field

[0002] This disclosure relates to diagnostic laboratories, and more specifically, to methods and apparatus for multi-agent pathfinding within diagnostic laboratory systems. Background Technology

[0003] Diagnostic laboratory systems perform tests or examinations to identify analytes or other components in biological samples such as serum, plasma, urine, interstitial fluid, and cerebrospinal fluid. Medical technicians collect biological samples in sample containers such as test tubes and then bring the biological sample to the diagnostic laboratory system for testing. Each biological sample may have multiple testing requirements, each of which requires different tests to be performed by the diagnostic laboratory system.

[0004] A diagnostic laboratory system may include multiple instruments, each configured to perform one or more tests. When a biological sample with a single test requirement is received in the diagnostic laboratory system, the biological sample is transferred to an instrument configured to perform the test. Similarly, when a biological sample with multiple test requirements is received in the diagnostic laboratory system, the biological sample is directed to one or more instruments collectively configured to perform multiple tests.

[0005] The scheduler creates a timetable for guiding sample containers to specific instruments for testing the samples contained within. However, determining the optimal routes and timings for transporting and handling multiple sample containers while avoiding collisions and unnecessary delays is a challenge. Therefore, there is a need for improved methods and apparatus for sample container routing within diagnostic laboratory systems. Summary of the Invention

[0006] In some embodiments, a method for pathfinding of sample containers in a diagnostic laboratory system is provided, the method comprising: receiving a batch of sample containers in a diagnostic laboratory system having diagnostic laboratory equipment, one or more tracks connecting the diagnostic laboratory equipment, and a plurality of sample carriers configured to transport the sample containers within the diagnostic laboratory system, each sample container containing a sample to be processed; obtaining a grid of the diagnostic laboratory system having cells including the diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment; assigning agents to each of the sample carriers; determining a target for each agent assigned to a sample carrier holding a sample container, the target including an earliest arrival time constraint, a deadline time constraint, and a target processing time constraint; and employing a neural network trained with Multi-Agent Pathfinding (MAPF) to determine the action of each agent, the neural network being trained using arrival, deadline, and priority constraints, as well as a non-instantaneous target processing time.

[0007] In some embodiments, a diagnostic laboratory system is provided, the system including a diagnostic laboratory device; one or more tracks connected to the device; a processor; and a memory coupled to the processor. The memory includes computer-executable instructions that, when executed by the processor, cause the processor to: obtain a grid of the diagnostic laboratory system having cells including the device and one or more tracks connected to it; assign agents to each sample carrier within the system; determine an objective for each agent assigned to a sample carrier holding a sample container, the objective including an earliest arrival time constraint, a deadline time constraint, and a target processing time constraint; and determine the action of each agent using a MAPF-trained neural network trained with arrival, deadline, and priority constraints, as well as a non-instantaneous target processing time.

[0008] In some embodiments, a method for pathfinding of sample containers in a diagnostic laboratory system is provided, the method comprising: creating a grid for the diagnostic laboratory system, the grid having cells including diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment; creating a plurality of agents, each agent representing a sample carrier within the diagnostic laboratory system; determining a target for each agent assigned to a sample carrier holding a sample container, the target including an earliest arrival time constraint, a deadline constraint, and a target processing time constraint; creating a neural network that takes the earliest arrival time constraint, the deadline constraint, and the target processing time constraint of the agent as input, and takes an action or probability of an action of the agent as output; training the neural network using arrival, deadline, and priority constraints and a non-instantaneous target processing time; and creating a pathfinding program that uses the trained neural network to generate actions for each agent within the diagnostic laboratory system.

[0009] Other aspects, features, and advantages of this disclosure may become apparent from the following description and illustration of exemplary embodiments, including the best mode contemplated for carrying out this disclosure. This disclosure may also be capable of having other embodiments and different embodiments, and certain details thereof may be modified in various ways without departing from the scope of this disclosure. Attached Figure Description

[0010] The accompanying drawings described below are provided for illustrative purposes and are not necessarily drawn to scale. Therefore, the drawings and description are intended to be illustrative in nature and not restrictive. The drawings are not intended to limit the scope of this disclosure in any way.

[0011] Figure 1 An embodiment of a diagnostic laboratory system according to the embodiments provided herein is illustrated.

[0012] Figure 2A The illustrations depict some embodiments provided herein. Figure 1 Example embodiments of computers.

[0013] Figure 2B An example flowchart according to some embodiments provided herein is illustrated, which depicts the information flow from the scheduler program to the pathfinding program, from the pathfinding program to the neural network, from the neural network to the pathfinding program, and from the pathfinding program to the sample carrier controller.

[0014] Figure 3A An example embodiment of another diagnostic laboratory system according to the embodiments provided herein is illustrated.

[0015] Figure 3BThe illustration shows a first agent and a second agent according to embodiments provided herein. Figure 3A It is part of the diagnostic laboratory system.

[0016] Figure 3C The illustration shows an embodiment according to the present document. Figure 3B This is part of a diagnostic laboratory system, which shows the vertices of the cells that a sample carrier can travel to, connected by edges.

[0017] Figure 3D The diagram shows... Figure 3B For clarity, the diagnostic laboratory section is illustrated with vertices and edges, but not cells.

[0018] Figure 3E The illustration shows a representation according to the embodiments provided herein. Figure 3A Multiple example masks of the state of units within a diagnostic laboratory system.

[0019] Figure 4 This is a flowchart of an example method for path finding of sample containers in a diagnostic laboratory system, based on some embodiments provided herein.

[0020] Figure 5 This is a flowchart of another method for finding paths for sample containers in a diagnostic laboratory system, based on some embodiments provided herein. Detailed Implementation

[0021] Regardless of grammatical usage, individuals with male, female, or other gender identities are included in the term.

[0022] Scheduling and / or routing sample containers to specific instruments within a diagnostic laboratory system is complex because such systems have many different types of components and may receive many different types of samples. For example, a diagnostic laboratory system may include numerous instruments, each of which can be configured with different test menus, test procedures, and test durations. Furthermore, samples to be analyzed within a diagnostic laboratory system may have priorities, such as routine or stat samples requiring immediate processing.

[0023] The embodiments presented herein employ multi-agent pathfinding techniques to determine actions for transporting and handling sample containers within a diagnostic laboratory system. Pathfinding plays a crucial role in many areas of automation. Many automation setups involve multiple interacting components that require coordinated action. The Multi-Agent Pathfinding Problem (MAPF) elucidates important aspects of the target domain, such as collision avoidance and minimization of travel time. However, a key assumption in current MAPF is that task processing is instantaneous. This discrepancy is amplified in areas of laboratory or factory automation where tasks, as part of an automation pipeline, require non-trivial processing. Machines may spend varying amounts of time processing different tasks. Without temporal relaxation corresponding to machine availability, agent congestion can occur because such methods fail to capture priority constraints and queuing effects caused by non-instantaneous task processing times.

[0024] According to one or more embodiments provided herein, the aforementioned MAPF problem is solved by implementing schedules and priority constraints. For example, schedules can be enforced through automated pipeline capabilities such as machine operating characteristics. In some embodiments, a time-constrained MAPF (MAPF-TC) approach is provided, which can be modeled as a sequential decision problem.

[0025] In one or more embodiments, a multi-agent reinforcement learning (MARL) model is employed to solve a time-constrained multi-agent pathfinding problem. In some embodiments, a time-constrained version of the MAPF problem is provided, which introduces arrival, deadline, and priority constraints with non-instantaneous target processing times. Furthermore, in some embodiments, novel reward and observation spaces are provided for modeling the time aspects and priority constraints.

[0026] The following is for reference. Figure 1-5 These and other embodiments are described herein.

[0027] right Figure 1 For reference, Figure 1 An embodiment of a diagnostic laboratory system 100 according to embodiments provided herein is illustrated. The diagnostic laboratory system 100 may include a plurality of instruments 102 (e.g., diagnostic laboratory devices) configured to process samples stored within sample containers 104 (some labeled) and to test the samples (e.g., trials or other tests). Performing a test may include performing one or more operations on the sample. For example, each operation may include one or more measurements. One or more of the instruments 102 may include a plurality of different modules configured to perform operations.

[0028] Samples can be various biological samples collected from individuals, such as patients being evaluated by medical professionals. Samples can be collected in sample container 104 and delivered to diagnostic laboratory system 100, and subsequently transported throughout the diagnostic laboratory system 100 via track 108, such as to instrument 102. For example, sample container 104 can be transported by sample carriers 110 (some of which are labeled). Figure 1 In one embodiment, the diagnostic laboratory system 100 has three instruments 102, including a sample handler 114, a first analyzer 116, and a second analyzer 118. The diagnostic laboratory system 100 may include a... Figure 1 The number of instruments shown is less or more.

[0029] In some embodiments, track 108 may be close to or extend around instrument 102, such as Figure 1 As shown. Parts or modules of instrument 102 may have devices for transferring sample container 104 to sample carrier 110 and transferring sample container 104 from sample carrier 110, such as automated mechanical devices (robots). Figure 1 (Not shown in the diagram). Track 108 may include multiple interconnected segments 120 (some labeled). Sample carrier 110 may be movable as shown by dashed lines 126 in the segments 120. In some embodiments, some segments 120 may be integrated with one or more of the instruments 102.

[0030] Diagnostic laboratory systems, such as laboratory system 100, can have numerous instruments and may have tracks that link to other laboratory systems. Such laboratory systems (including diagnostic laboratory system 100) can simultaneously move and process multiple sample carriers 110 and their respective sample containers 104. In some embodiments, diagnostic laboratory system 100 can simultaneously move and process hundreds or thousands of sample carriers 110 and their respective sample containers 104.

[0031] Diagnostic laboratory system 100 may include or be coupled to computer 130, which is configured to execute one or more programs and control the operation of diagnostic laboratory system 100. Computer 130 may be configured to communicate with instrument 102 and other components of diagnostic laboratory system 100, such as components in a transport system (e.g., track 108 and components controlling its operation). The transport system may include some or all of the components (e.g., motors, sensors, power supplies, etc.) configured to transport samples throughout diagnostic laboratory system 100. Computer 130 may include processor 132 configured to execute programs, including programs other than those described herein. Programs may be implemented as computer code, computer-executable instructions, etc.

[0032] Computer 130 may include or access memory 134, which may store one or more programs 136 and / or data. Memory 134 may be any suitable type of memory, such as, but not limited to, one or more of volatile memory and / or non-volatile memory. In one or more embodiments, memory 134 may be non-transitory memory (e.g., hard disk drive, solid-state drive, flash drive, another non-transitory computer-readable medium, etc.). Program 136 may be computer code and / or instructions executable on or by processor 132. As described below, such a program may control all or part of the operation of the diagnostic laboratory system 100.

[0033] Computer 130 can be coupled to workstation 138, which is configured to allow a user to interface with diagnostic laboratory system 100. Workstation 138 may include display 140, keyboard 142, and other peripheral devices. Programs in memory 134 can cause display 140 to show information about the processing of samples within diagnostic laboratory system 100.

[0034] When a sample is received in the diagnostic laboratory system 100 for testing, the sample undergoes a complex sequence of operations or procedures within a specific workflow defined by the particular test. Each type of test can have a unique sequence of operations. The workflow sequence may begin with a sample handling or sample container handling operation, followed by the aspiration and dispersion of the sample and / or reagents into cuvettes. The mixture in the cuvettes may undergo other operations required for the test. The workflow sequence may end with a measurement operation, such as a photometric measurement, to determine the chemical properties of the sample. Each operation in the workflow sequence can be performed using one or more of the instruments 102.

[0035] Figure 2A The illustrations depict some embodiments. Figure 1 Example embodiment of computer 130. In Figure 2AIn this embodiment, the memory 134 of the computer 130 includes a scheduler program 202 coupled to the pathfinding program 204. As described below, the pathfinding program 204 may receive schedule information from the scheduler program 202 and employ a neural network 206 trained with Multi-Agent Pathfinding (MAPF) to determine the actions (such as carrier movement of each sample carrier 110) within the diagnostic laboratory system 100. These actions may be communicated from the pathfinding program 204 to a sample carrier controller 208 for execution. For example, the sample carrier controller 208 may control the operation of motors, sensors, power supplies, etc., used to transport the sample carriers 110 on the track 108. Although shown within the memory 134 of the computer 130, it will be understood that the scheduler program 202 and / or the sample carrier controller 208 may reside in different memories and / or be executed by different processors.

[0036] Figure 2B An example flowchart according to some embodiments is illustrated, which depicts the information flow from scheduler program 202 to path finder program 204, from path finder program 204 to neural network 206 and from neural network 206 to path finder program 204, and from path finder program 204 to carrier controller 208.

[0037] Scheduler program 202 can be configured to determine a schedule or workflow for processing all sample containers 104 within the diagnostic laboratory system 100 (e.g., where sample carrier 110 should travel to pick up and transport sample containers 104 for processing). For example, scheduler program 202 can determine that sample containers 104 within sample disposal unit 114 should be retrieved and delivered to one or more of first analyzer 116 and second analyzer 118 for processing. In some embodiments, scheduler program 202 can generate a workflow indicating the diagnostic laboratory equipment to be accessed by each sample container and their order (e.g., based on information from scanning or imaging tags attached to the sample containers regarding one or more analyses to be performed on the samples in the sample containers). In one or more embodiments, processor 132 (or another processor) can simulate the workflow for each sample container to estimate workflow completion time information for each sample container.

[0038] Schedule information can be provided to pathfinding program 204, which in turn can employ a MAPF-trained neural network 206 to determine the action of each sample carrier 110 within the diagnostic laboratory system 100, so that the schedule determined by scheduler program 202 is effectively implemented (e.g., collision-free delivery of sample containers for processing within any required deadlines). These actions can be fed into sample carrier controller 208 for execution. Therefore, scheduler program 202, pathfinding program 204, MAPF-trained neural network 206, and sample carrier controller 208 can be deployed within the diagnostic laboratory system for use to facilitate sample processing.

[0039] In one or more embodiments, the time-constrained MAPF problem can be denoted by an undirected graph G=(V,E) and a set of m agents. The algorithm is defined by a timetable S. The vertex set V corresponds to the positions on the graph, and the edge set E represents the motion constraints for each vertex. At each step, the agent can move or wait at its current position. Each agent is assigned a series of target positions according to the timetable S. The timetable S represents time windows, which include the earliest arrival time (A) for agent i and target j. t i(j) ), deadline (D t i(j) ) and target processing time (P t i(j) For example, the timetable S, along with the earliest arrival time, deadline, and target processing time constraints, can be provided by the scheduler program 202.

[0040] A t i(j) Constrain the agent on when it can begin processing its objective and where it might arrive before starting, but the objective will not be processed before the scheduled time. D t i(j) Constrain the latest time that the agent is allowed to reach its goal for processing. P t i(j) This schedule represents (e.g., and constrains) non-transient goal processing that agents are not allowed to move from. The schedule also enforces priority constraints on goal locations shared among multiple agents (e.g., factory machines such as sample analyzers). While soft violations with penalized deadlines are permissible, priority constraints are still enforced, effectively inducing a queue of agents sharing goal locations. In some embodiments, agents are always present on the graph from the beginning and do not disappear after completing their goals. The persistence of all agents on spatially constrained trajectories is a key challenge.

[0041] In one or more embodiments, the interaction between the agent and the environment can be modeled as a partially observable Markov decision process (POMDP) ​​(S, O, A, P, R, ...). In this context, S is the set of environmental states, O is the set of partial observations, A is the set of actions, P is a function representing the transition probability, and R is the reward function. This is a discount factor. In the embodiments described herein, the environment is constrained to a 2D mesh, where each agent is limited to a local field of view. In other embodiments, a 3D mesh may be used. Homogeneous policies can be learned, which can be deployed to different numbers of agents and executed in a distributed manner.

[0042] In some embodiments, the observation space may include two components. First, the local view may represent the state of neighboring vertices relative to the agent's current position. Second, information about the next target position can be directly encoded using a vector that represents the target's spatial and temporal information in three dimensions, as described below.

[0043] Figure 3A An example embodiment of another diagnostic laboratory system 300 according to the embodiments provided herein is illustrated. References Figure 3A The diagnostic laboratory system 300 is represented as a grid 302, which defines multiple cells 304 within the diagnostic laboratory system 300 (only some are labeled). For example... Figure 3A As indicated by the key in the diagram, unshaded cells 304 represent sample carrier movement locations, such as track segments, track connections, and machine padding locations. Cells 304 with medium shading represent diagnostic laboratory equipment, such as sample analyzers and sample handlers. Cells 304 with dark shading represent empty or unusable areas of the diagnostic laboratory system 300. Figure 3A In some embodiments, carrier movement includes movement in the north (N), south (S), east (E), and west (W) directions. In some embodiments, the size of the unit 304 may be determined based on one or more of the size of the sample carrier 110 (e.g., large enough to accommodate the sample carrier), the movement resolution of the track segment used, etc. For example, in some embodiments, the size of each unit 304 may be determined to hold a single sample carrier 110. Other unit sizes may be used.

[0044] Figure 3B The illustration shows an embodiment according to the present document. Figure 3AA portion 300a of a diagnostic laboratory system 300 includes a first agent 308a and a second agent 308b. A first local field of view (FOV) 310a is shown for the first agent 308a, and a second local field of view (FOV) 310b is shown for the second agent 308b, each FOV including the nearest neighbor unit 304. Other FOV sizes may be used. The first agent 308a is located on a track segment close to a first diagnostic laboratory device 312a (e.g., a first sample analyzer). The second agent 308b is located within a second diagnostic laboratory device 312b (e.g., a second sample analyzer).

[0045] Figure 3C A portion 300a of a diagnostic laboratory system 300 according to an embodiment provided herein is illustrated, showing vertices 314 (connected by edges 316) of cells 304 that can be accessed for a sample carrier. Figure 3D Part 300a is shown in the figure. For clarity, this part is shown with vertex 314 and edge 316 but without element 304.

[0046] like Figure 3D As shown, the local field of view 310a can represent the state of neighboring vertices 314 within a radius R relative to the current position of the first agent 308a. For example, the local field of view (FOV) can be transformed into a square of length 2*R+1 around the agent.

[0047] In some embodiments, the state of each unit 304 within the field of view can be represented by multiple masks 318a-d, such as Figure 3E As shown. Although in Figure 3E Four masks are shown, but it should be understood that fewer or more masks may be used, as well as larger masks (e.g., for a larger field of view). For example, each cell state may be represented by one or more of the following: (1) a binary mask of obstacles; (2) a binary mask of the positions of all agents within the FOV; (3) a binary mask of the target positions of other agents within the FOV; and (4) a binary mask of the target position of an agent within the cell, projected onto the boundary if it is outside the FOV. Furthermore, in some embodiments, if other agents share the same target position with an agent, each cell state of the cell having that agent may also be represented by the normalized distance from each vertex in the FOV to the agent's target and / or the order of other agents in the queue.

[0048] As stated above, in some embodiments, a vector can be used to directly encode information about the agent's next target location, representing the spatial and temporal information of the target in three dimensions. For example, the spatial vector Gs can have a first vector component Gs1 representing the difference between the agent's current x-coordinate and the target x-coordinate. A second vector component Gs2 can represent the difference between the agent's current y-coordinate and the target y-coordinate. A third vector component Gs3 can represent the magnitude of the vector from the agent's current position to its target position, which is limited to an absolute value (e.g., 60 in some embodiments).

[0049] The time vector Gt can represent the agent's target time information using three dimensions. For example, if the current time is less than At, the first vector component Gt1 can represent the remaining time until the earliest arrival time At; otherwise, Gt1 can represent 0.0. If the current time is between At and Dt, the second vector Gt2 can represent the remaining time until the deadline Dt; otherwise, Gt2 can represent 0.0. If the deadline Dt has passed, indicating that the agent is late, the third vector Gt3 can be set to 1; otherwise, Gt3 can be set to 0.0. Other spatial and / or time vectors can also be used.

[0050] Back Figure 2A In some embodiments, the pathfinding procedure 204 may include computer-executable instructions (stored in memory 134) that, when executed by processor 132, cause processor 132 to obtain a grid for the diagnostic laboratory system (e.g., for...). Figure 3A The grid 302 of the diagnostic laboratory system 300. The grid defines multiple cells, which include cells within the diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment (e.g., such as...). Figure 3A and Figure 3B (As shown).

[0051] The pathfinding procedure 204 can also assign an agent to each sample carrier within the diagnostic laboratory system. For example, a first agent 308a and a second agent 308b are assigned to sample carriers 320a and 320b, respectively. Figure 3B ).

[0052] The pathfinding procedure 204 can further determine a target for each agent assigned to the sample carrier holding the sample container, the target including an earliest arrival time constraint, a deadline constraint, and a target processing time constraint. As stated, in some embodiments, such information can be obtained from the scheduler procedure 202 ( Figure 2A )supply.

[0053] The pathfinding procedure 204 can employ a MAPF-trained neural network (e.g., MAPF-trained neural network 206) to determine the actions of each agent, which is trained using arrival, deadline, and priority constraints, as well as non-instantaneous target processing time. Figure 3A In the example, the actions of the agent (e.g., the sample carrier) could include waiting or moving north, south, east, or west. Other actions may be provided.

[0054] In some embodiments, the sample carrier controller 208 may be employed to perform one or more of the defined actions (via computer program instructions in memory 134). For example, the sample carrier controller 208 may cause the transfer of a sample container for processing within the diagnostic laboratory system 300.

[0055] Training of neural networks The MAPF neural network 206 can be any suitable neural network, such as a convolutional neural network, a transformer neural network, etc. In some embodiments, each agent can receive a positive reward when it reaches its goal within the scheduled time window. A negative penalty can be imposed for each time step in which an agent occupies its goal before its earliest arrival time. This approach aims to reduce congestion, as agents waiting for their goals may block the path of other agents. A negative penalty can be imposed for each time step in which any agent has not reached its goal after the deadline.

[0056] In the reinforcement learning (RL) exposition of MAPF, penalties are typically imposed when an agent collides with static obstacles or other agents in the environment. The embodiments provided herein aim to reduce unnecessary subproblems for more efficient learning. One subproblem is static obstacle avoidance. Accordingly, invalid actions can be masked (e.g., ignored) when an agent is about to collide with an obstacle. Another subproblem is collision avoidance with other agents. The inventors have observed that, without collision penalties, neural network policies tend to converge faster on goal completion behaviors while implicitly avoiding collisions. Explicit collision penalties can slow down training because agents are more likely to collide with each other early in the learning process. Collision avoidance is still enforced through the environment. Furthermore, in some embodiments, no penalties are imposed on agent movements. Additionally, in one or more embodiments, an agent may need to wait or reroute its path before its deadline.

[0057] Unlike some existing methods, a global positive reward can be distributed among all agents when any agent completes the goal. Similarly, a global negative reward can be distributed among all agents when any agent is late for their deadline. The idea is to incentivize cooperation among agents, especially those who are relaxed until their deadline and can afford to take a longer path.

[0058] Table 1 provides examples of reward values ​​for different events. Other reward values ​​may be used.

[0059] Table 1 event award Achieve the goal within the time window +1.0 Achieve the goal before the time window –0.1 Arriving late (behind the time window) and failing to achieve the goal –0.1 Achieve the objective during processing. 0.0 Environmental collision Hidden Agent collision 0.0 Agent actions (move / wait) 0.0 Shared Target Rewards +0.2 Shared Delay Rewards –0.04

[0060] In some embodiments, the local receptive field is processed by a convolutional neural network (CNN) (such as a 3-layer CNN), where each layer except the last is followed by a max-pooling operation, and the last layer is followed by a global averaging to produce a receptive field representation vector. Then, positional and temporal window vectors are fed through fully connected layers to produce a combined target representation vector. Both the receptive field and target representation vectors are then passed through a recursive module to incorporate information about past states as a mitigation of the partial observability problem. The model is optimized using proximal policy optimization (PPO) as described in Schulman, John, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. 2017. “Proximal Policy Optimization Algorithms” arXiv. http: / / arxiv.org / abs / 1707.06347.

[0061] Selective scenarios can be used to study the impact of different components on model performance. In some embodiments, the time window can be parameterized to change the resulting scheduling relaxation.

[0062] Unlike warehouses, automated factories are constrained by available space, machine dimensions, and safety constraints such as power supply and ventilation vents. To capture these diverse constraints, a parametric layout planning generator can be employed to generate parametric layouts for diagnostic laboratory systems. Training on various such layouts can generate a model capable of satisfying nearly 100% of the objectives within their respective time windows. In such layouts, agents may be able to learn traffic-like behaviors, such as parallel lanes / corridors for reverse flow. While scaling to a larger number of agents within the same fixed layout is an important aspect of MAPF, it may not maintain the same value under time constraints where the overall system throughput is limited by (automated) scheduling capabilities. Adding more agents may simply result in agents idling and having ample time to navigate tracks. As the number of agents in MAPF increases under time constraints, the bottleneck becomes scheduling rather than pathfinding.

[0063] In some embodiments, a more restrictive grid layout (e.g., a single channel) may still be able to achieve nearly 100% of the agent's objective. In a well-connected layout where the tracks form continuous paths, the agent can learn to rotate around the tracks, thus maintaining a flow state and not blocking other agents (who might deviate from their objectives) until their deadline expires. In a more restrictive layout with forked dead ends, the agent can learn to reverse through filled sections around the decision point (e.g., the location of a sample carrier adjacent to diagnostic laboratory equipment).

[0064] In some embodiments, a layout design can be created to identify the impact of different factors on performance and to establish limitations of the pathfinding procedure 204. For example, the layout can be modified across three design dimensions: redundancy, size, and padding. Redundancy can be modified by connecting tracks to remove forked dead ends. Size can be tested by changing the number of targets / channels. In some embodiments, this can include producing two variations: a small variation with 3 targets and a large variation with 6 targets / channels, respectively. This factor can affect the number of agents (e.g., the number of agents can be set to equal the number of targets plus 2). Finally, different padding can be added around the decision nodes.

[0065] The scheduling distribution can be modified across multiple design dimensions, such as the single-agent shortest distance (A* factor), which is the minimum distance an agent travels assuming no other agents are present in the objective, and the congestion estimate for multi-agents (where the average of all single-agent shortest distances is added as a congestion estimate). These represent scheduling constraints affecting the earliest arrival time of each agent, thus providing the minimum necessary conditions for scheduling feasibility. Furthermore, the size of the time window from the earliest arrival time to the deadline and the task processing runtime before completion can be varied.

[0066] One objective of the MAPF-trained neural network 206 is to maximize the number of objectives completed within its scheduled time window. The relative throughput of objectives completed within the time window can be measured. This allows for standardized comparisons to understand the impact of different time window distributions on the reinforcement learning model. Example metrics to consider include the percentage of objectives completed within the lateness tolerance, thus highlighting the model's suboptimal performance (if any), and the time taken to complete a predetermined number of objectives (e.g., 100), thus standardizing across different time schedules. The coefficient of variation, defined as the ratio between the mean and standard deviation of objective completion times, can also be used to capture the model's reliability when producing similar performance.

[0067] Figure 4 This is a flowchart of an example method 400 for path finding of sample containers in a diagnostic laboratory system, according to some embodiments. In some implementations, Figure 4 One or more process frames can be found by a pathfinding program (such as...) Figure 2A The pathfinding procedure 204 is executed. For example, memory 134 may include computer-executable instructions that, when executed by processor 132, cause processor 132 to execute a reference... Figure 4 One or more of the process steps described. In at least some embodiments, such computer-executable instructions may be stored in non-transitory memory and / or included in a non-transitory computer-readable medium.

[0068] like Figure 4 As shown, method 400 may include receiving a batch of sample containers in a diagnostic laboratory system having diagnostic laboratory equipment, one or more tracks connecting the diagnostic laboratory equipment, and a plurality of sample carriers configured to transport the sample containers within the diagnostic laboratory system, each sample container containing a sample to be processed (box 402). For example, a sample handler (such as...) Figure 1 The sample handler 114) can be used in diagnostic laboratory systems (e.g., Figure 1 Diagnostic laboratory system 100 or Figure 3AA diagnostic laboratory system 300 receives a batch of sample containers (e.g., sample container 104), the diagnostic laboratory system having diagnostic laboratory equipment (e.g., sample handlers, sample analyzers, etc.), one or more tracks connecting the diagnostic laboratory equipment, and multiple sample carriers configured to transport the sample containers within the diagnostic laboratory system, each sample container containing a sample to be processed, as described above (e.g., ...). Figure 1 The sample handler 114, the first sample analyzer 116, the second analyzer 118, the track 108, the sample container 104, and the sample carrier 110. Note that some sample carriers may be empty until they receive the sample container at the input module of the diagnostic laboratory system (e.g., at the sample handler 114 of the diagnostic laboratory system 100).

[0069] For example Figure 4 As shown, method 400 may include obtaining a grid for a diagnostic laboratory system, the grid having cells (box 404) including diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment. For example, Figure 2A The pathfinding procedure 204 can obtain the path for the diagnostic laboratory system 300 ( Figure 3A The grid 302 has units 304 including diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment, as described above. In some embodiments, the grid may be stored in memory 134, or automatically generated by scheduler program 202 and / or pathfinding program 204 and / or another program.

[0070] like Figure 4 As further shown, method 400 may include assigning an agent to each sample carrier (block 406). For example, pathfinding procedure 204 may assign an agent to each sample carrier 110 within the diagnostic laboratory system 100, as described above.

[0071] For example Figure 4 As shown, method 400 may include determining a target for each agent assigned to a sample carrier holding a sample container, the target including an earliest arrival time constraint, a deadline time constraint, and a target processing time constraint (block 408). For example, pathfinding procedure 204 (e.g., via scheduler procedure 202) may determine a target for each agent assigned to a sample carrier 110 holding a sample container 104, the target including an earliest arrival time constraint, a deadline time constraint, and a target processing time constraint, as described above. In some embodiments, this may include obtaining data from a scheduler procedure such as scheduler procedure 202 (… Figure 2AThe program receives the earliest arrival time, the deadline, and the target processing time constraint. Furthermore, in some embodiments, the pathfinding program 204 (and / or the scheduler program 202) may determine a set of targets for each agent, each target including the earliest arrival time constraint, the deadline constraint, and the target processing time constraint.

[0072] like Figure 4 As further shown, method 400 may include employing a MAPF-trained neural network to determine the action of each agent, the neural network being trained using arrival, deadline, and priority constraints, as well as non-instantaneous target processing time (box 410). For example, pathfinding procedure 204 may employ a MAPF-trained neural network 206 to determine the action of each agent, the neural network 206 being trained using arrival, deadline, and priority constraints, as well as non-instantaneous target processing time, as described above.

[0073] In some embodiments, using a MAPF-trained neural network to determine the action of each agent may include inputting local field-of-view information of each agent into the neural network. This local field-of-view information represents the state of neighboring vertices within a predetermined distance from the agent's current position. For example, the predetermined distance may be a predetermined radius R, which defines the local field of view of each agent as a square of length 2*R+1. Figure 3D The local field of view 310a is shown in the middle.

[0074] Each unit can have a unit type, and the agent's local field of vision can include a central unit with the agent and multiple units surrounding that central unit, such as those targeting... Figure 3B-3D The local fields of view 310a and 310b are shown.

[0075] In one or more embodiments, using a MAPF-trained neural network to determine the actions of each agent may include using one or more masks (e.g., ...) for local vision. Figure 3E The masks 318a-d) are input into the neural network, and the one or more masks include at least one of the following: a mask of obstacles, a mask of the positions of all agents within the local field of view, a mask of the target positions of other agents within the local field of view, and a mask of the target positions of agents, as described above. Furthermore, using a MAPF-trained neural network to determine the action of each agent may include at least one of the following: the distance information to the agent's target and the order of other agents if they share the target position with the agent.

[0076] Determining the actions of each agent using a MAPF-trained neural network can also include inputting each agent's next target location information into the neural network. For example, each agent's next target location information may include the difference between the agent's current location and the agent's next target location, the time remaining until the agent's earliest arrival time, the time remaining until the target deadline, and / or information on whether the target deadline has passed.

[0077] In some embodiments, neural network 206 may be trained using at least one of a global positive reward for all agents when any agent completes the objective and a global negative reward for all agents when any agent is late for the deadline.

[0078] although Figure 4 An example block diagram of method 400 is shown, but in some implementations, method 400 may include methods with... Figure 4 The text describes boxes that are compared to additional boxes, fewer boxes, different boxes, or boxes with different arrangements. Alternatively, two or more boxes of method 400 can be executed in parallel.

[0079] In some embodiments, the sample carrier controller 208 of the diagnostic laboratory system 100 or 300 may be used. Figure 2A To perform one or more of the actions determined by the pathfinding procedure 204 and / or the neural network 206, including transferring the sample container for processing (e.g., transferring the sample container 104 to the sample analyzer 116 or 118).

[0080] Figure 5 This is a flowchart of another method 500 for path finding of sample containers in a diagnostic laboratory system, according to some embodiments. In some implementations, Figure 5 One or more process frames can be generated by a processor (such as...) Figure 2A The processor 132 executes the reference. For example, the memory 134 may include computer-executable instructions that, when executed by the processor 132, cause the processor 132 to perform the reference. Figure 5 One or more of the process steps described. In at least some embodiments, such computer-executable instructions may be stored in non-transitory memory and / or included in a non-transitory computer-readable medium.

[0081] like Figure 5As shown, method 500 may include creating a grid for a diagnostic laboratory system having cells (block 502) that include diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment. For example, processor 132 executing one or more programs may create a grid for the diagnostic laboratory system having cells that include diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment, as described above.

[0082] For example Figure 5 As shown, method 500 may include creating multiple agents, each representing a sample carrier within a diagnostic laboratory system (block 504). For example, processor 132 executing one or more programs may create multiple agents, each representing a sample carrier within a diagnostic laboratory system, as described above.

[0083] like Figure 5 As further shown, method 500 may include determining a target for each agent assigned to a sample carrier holding a sample container, the target including an earliest arrival time constraint, a deadline constraint, and a target processing time constraint (block 506). For example, processor 132 executing scheduler program 202 and / or another program (e.g., pathfinding program 204) may determine a target including an earliest arrival time constraint, a deadline constraint, and a target processing time constraint for each agent assigned to a sample carrier holding a sample container, as described above.

[0084] For example Figure 5 As shown, method 500 may include creating a neural network that takes as input an agent's earliest arrival time constraint, a deadline time constraint, and a target processing time constraint, and outputs the agent's action or the probability of the action (box 508). As described above, in some embodiments, neural network 206 may be any suitable neural network, such as a convolutional neural network, a transformer neural network, etc.

[0085] like Figure 5 As further shown, method 500 may include training a neural network using arrival, deadline, and priority constraints, as well as non-instantaneous target processing time (box 510). For example, processor 132 may train neural network 206 using arrival, deadline, and priority constraints, as well as non-instantaneous target processing time, as described above.

[0086] For example Figure 5As shown, method 500 may include creating a pathfinding program that uses a trained neural network to generate actions for each agent within the diagnostic laboratory system (box 212). For example, processor 132 may be used to create pathfinding program 204 that uses a trained neural network 206 to generate actions for each agent within the diagnostic laboratory system, as described above.

[0087] although Figure 5 An example box for method 500 is shown, but in some implementations, method 500 may include... Figure 5 The text describes boxes that are compared to additional boxes, fewer boxes, different boxes, or boxes with different arrangements. Alternatively, two or more boxes of method 500 can be executed in parallel.

[0088] Non-limiting illustrative examples The following is a non-limiting list of illustrative embodiments of this disclosure: An illustrative method for pathfinding of sample containers in a diagnostic laboratory system, the method comprising: receiving a batch of sample containers in a diagnostic laboratory system having diagnostic laboratory equipment, one or more tracks connecting the diagnostic laboratory equipment, and a plurality of sample carriers configured to transport the sample containers within the diagnostic laboratory system, each sample container containing a sample to be processed; obtaining a grid for the diagnostic laboratory system having cells including the diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment; assigning agents to each sample carrier within the diagnostic laboratory system; determining a target for each agent assigned to a sample carrier holding a sample container, including an earliest arrival time constraint, a deadline time constraint, and a target processing time constraint; and employing a neural network trained with Multi-Agent Pathfinding (MAPF) to determine the action of each agent, the neural network being trained using arrival, deadline, and priority constraints, as well as a non-instantaneous target processing time.

[0089] The illustrative method of any of the foregoing illustrative embodiments further includes employing the sample carrier controller of the diagnostic laboratory system to perform one or more of the determined actions, including transferring the sample container for processing.

[0090] The illustrative method of any of the foregoing illustrative embodiments, wherein determining a target for each agent assigned to a sample carrier holding a sample container, including an earliest arrival time constraint, a deadline time constraint, and a target processing time constraint, includes receiving the earliest arrival time constraint, the deadline time constraint, and the target processing time constraint from a scheduler program.

[0091] The illustrative method of any of the foregoing illustrative embodiments further includes determining a set of objectives for each agent, each objective including an earliest arrival time constraint, a deadline constraint, and an objective processing time constraint.

[0092] The illustrative method of any of the foregoing illustrative embodiments, wherein determining the action of each agent using the MAPF-trained neural network includes inputting local field-of-view information of each agent into the neural network, the local field-of-view information representing the state of neighboring vertices within a predetermined distance of the agent's current position.

[0093] The illustrative method of any of the foregoing illustrative embodiments, wherein the predetermined distance is a predetermined radius R, defines the local field of view of each agent as a square of length 2*R+1.

[0094] The illustrative method of any of the foregoing illustrative embodiments, wherein each unit has a unit type, and wherein the agent's local field of view includes a central unit and a plurality of units surrounding the central unit, the central unit including the agent.

[0095] An illustrative method in any of the foregoing illustrative embodiments, wherein determining the action of each agent using the MAPF-trained neural network includes inputting one or more masks for the local field of view into the neural network, the one or more masks including at least one of an obstacle mask, a mask of the positions of all agents within the local field of view, a mask of the target positions of other agents within the local field of view, and a mask of the target positions of the agents.

[0096] The illustrative method of any of the foregoing illustrative embodiments, wherein the determination of the action of each agent using the MAPF-trained neural network includes inputting distance information to the target of the agent and at least one of the following: if other agents to be processed share the target location with the agent, then the order of the other agents.

[0097] The illustrative method of any of the foregoing illustrative embodiments, wherein using the MAPF-trained neural network to determine the action of each agent includes inputting the next target location information of each agent into the neural network.

[0098] The illustrative method in any of the foregoing illustrative embodiments, wherein the next target location information for each agent includes information about the difference between the agent's current location and the agent's next target location.

[0099] The illustrative method of any of the foregoing illustrative embodiments, wherein the next target location information of each agent includes at least one of the time remaining until the earliest arrival time of the agent, the time remaining until the target deadline, and whether the target deadline has passed.

[0100] The illustrative method of any of the foregoing illustrative embodiments, wherein the neural network is trained using at least one of a global positive reward for all agents when any agent completes the objective and a global negative reward for all agents when any agent is late for the deadline.

[0101] An illustrative diagnostic laboratory system includes: a diagnostic laboratory device; one or more tracks connected to the diagnostic laboratory device; a processor; and a memory coupled to the processor, the memory including computer-executable instructions that, when executed by the processor, cause the processor to: obtain a grid for the diagnostic laboratory system, the grid having cells including the diagnostic laboratory device and one or more tracks connected to the diagnostic laboratory device; assign agents to each sample carrier within the diagnostic laboratory system; determine a target for each agent assigned to a sample carrier holding a sample container, including an earliest arrival time constraint, a deadline time constraint, and a target processing time constraint; and determine the action of each agent using a neural network trained with Multi-Agent Pathfinding (MAPF), the neural network being trained using arrival, deadline, and priority constraints, as well as a non-instantaneous target processing time.

[0102] The illustrative diagnostic laboratory system of any of the foregoing illustrative embodiments further includes a sample carrier controller, and wherein the memory includes computer-executable instructions that, when executed by the processor, cause the processor to employ the sample carrier controller to perform one or more determined actions, including transferring a sample container for processing.

[0103] An illustrative diagnostic laboratory system of any of the foregoing illustrative embodiments, wherein the memory includes computer-executable instructions that, when executed by the processor, cause the processor to determine a set of objectives for each agent, each objective including an earliest arrival time constraint, a deadline time constraint, and an objective processing time constraint.

[0104] An illustrative diagnostic laboratory system of any of the foregoing illustrative embodiments, wherein the memory includes computer-executable instructions that, when executed by the processor, cause the processor to input local field-of-view information for each agent into the neural network, the local field-of-view information representing the state of neighboring vertices within a predetermined distance of the agent's current position.

[0105] The illustrative diagnostic laboratory system of any of the foregoing illustrative embodiments, wherein the predetermined distance is a predetermined radius R, which defines the local field of view of each agent as a square of length 2*R+1.

[0106] An illustrative diagnostic laboratory system in any of the foregoing illustrative embodiments, wherein each unit has a unit type, and wherein the agent's local field of view includes a central unit and a plurality of units surrounding the central unit, the central unit including the agent.

[0107] An illustrative diagnostic laboratory system of any of the foregoing illustrative embodiments, wherein the memory includes computer-executable instructions that, when executed by the processor, cause the processor to input one or more masks for the local field of view into the neural network, the one or more masks including at least one of an obstacle mask, a mask of the positions of all agents within the local field of view, a mask of the target positions of other agents within the local field of view, and a mask of the target positions of the agents.

[0108] An illustrative diagnostic laboratory system of any of the foregoing illustrative embodiments, wherein the memory includes computer-executable instructions that, when executed by the processor, cause the processor to: for each agent, input distance information to a target of the agent, and the order of the other agents if they share the target location with the agent.

[0109] An illustrative diagnostic laboratory system of any of the foregoing illustrative embodiments, wherein the memory includes computer-executable instructions that, when executed by the processor, cause the processor to: determine the action of each agent using the MAPF-trained neural network by inputting the next target location information of each agent into the neural network.

[0110] The illustrative diagnostic laboratory system of any of the foregoing illustrative embodiments, wherein the next target location information of each agent includes information about the difference between the agent's current location and the agent's next target location.

[0111] The illustrative diagnostic laboratory system of any of the foregoing illustrative embodiments, wherein the next target location information of each agent includes at least one of the time remaining until the earliest arrival time, the time remaining until the target deadline, and whether the target deadline has passed.

[0112] An illustrative diagnostic laboratory system in any of the foregoing illustrative embodiments, wherein the neural network is trained using at least one of a global positive reward for all agents when any agent completes the objective and a global negative reward for all agents when any agent is late for the deadline.

[0113] An illustrative method for pathfinding of sample containers in a diagnostic laboratory system, the method comprising: creating a grid for the diagnostic laboratory system, the grid having cells including diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment; creating a plurality of agents, each agent representing a sample carrier within the diagnostic laboratory system; determining a target for each agent assigned to a sample carrier holding a sample container, the target including an earliest arrival time constraint, a deadline constraint, and a target processing time constraint; creating a neural network that takes the earliest arrival time constraint, the deadline constraint, and the target processing time constraint of the agent as input, and takes an action or the probability of an action of the agent as output; training the neural network using arrival, deadline, and priority constraints and a non-instantaneous target processing time; and creating a pathfinding program that uses the trained neural network to generate actions for each agent within the diagnostic laboratory system.

[0114] The illustrative method of any of the foregoing illustrative embodiments further includes deploying the trained neural network and the pathfinding procedure for use by the diagnostic laboratory system.

[0115] The illustrative method in any of the foregoing illustrative embodiments further includes employing the pathfinding procedure to determine one or more actions, including transferring the sample container for processing.

[0116] An illustrative method in any of the foregoing illustrative embodiments, wherein the neural network is configured to input local field-of-view information of each agent into the neural network, the local field-of-view information representing the state of neighboring vertices within a predetermined distance of the agent's current position.

[0117] An illustrative method in any of the foregoing illustrative embodiments, wherein the neural network is configured to input one or more masks for the local field of view into the neural network, the one or more masks including at least one of an obstacle mask, a mask of the positions of all agents within the local field of view, a mask of the target positions of other agents within the local field of view, and a mask of the target positions of the agents.

[0118] An illustrative method in any of the foregoing illustrative embodiments, wherein the neural network is configured to: for each agent, input distance information to the target of the agent, and at least one of the following sequences of other agents to be processed if they share the target location with the agent.

[0119] An illustrative method of any of the foregoing illustrative embodiments, wherein the neural network is configured to input the next target location information for each agent.

[0120] The illustrative method in any of the foregoing illustrative embodiments, wherein the next target location information for each agent includes information about the difference between the agent's current location and the agent's next target location.

[0121] The illustrative method of any of the foregoing illustrative embodiments, wherein the next target location information of each agent includes at least one of the time remaining until the earliest arrival time of the agent, the time remaining until the target deadline, and whether the target deadline has passed.

[0122] The illustrative method of any of the foregoing illustrative embodiments further includes training the neural network with at least one of a global positive reward for all agents when any agent completes the objective and a global negative reward for all agents when any agent is late for the deadline.

[0123] While this disclosure is susceptible to various modifications and alternatives, specific methods and apparatus embodiments have been illustrated by way of example in the accompanying drawings and have been described in detail herein. However, it should be understood that the specific methods and apparatus disclosed herein are not intended to limit this disclosure.

Claims

1. A method for pathfinding of sample containers in a diagnostic laboratory system, the method comprising: A batch of sample containers is received in a diagnostic laboratory system, the diagnostic laboratory system having diagnostic laboratory equipment, one or more tracks connected to the diagnostic laboratory equipment, and a plurality of sample carriers configured to transport the sample containers within the diagnostic laboratory system, each of the sample containers containing a sample to be processed. Obtain a grid for the diagnostic laboratory system, the grid having cells including the diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment; The intelligent agent is assigned to each of the sample carriers; For each agent assigned to a sample carrier holding a sample container, a target is determined, including an earliest arrival time constraint, a deadline time constraint, and a target processing time constraint. as well as The action of each agent is determined by a neural network trained with Multi-Agent Pathfinding (MAPF), which is trained using arrival, deadline, and priority constraints as well as non-instantaneous target processing time.

2. The method of claim 1, further comprising using a sample carrier controller of the diagnostic laboratory system to perform one or more of the determined actions, including transferring a sample container for processing.

3. The method according to claim 1, wherein, For each agent assigned to a sample carrier holding a sample container, a target is determined, including an earliest arrival time constraint, a deadline time constraint, and a target processing time constraint, which are received from the scheduler program.

4. The method according to claim 1, further comprising determining a series of objectives for each agent, each objective including an earliest arrival time constraint, a deadline constraint, and an objective processing time constraint.

5. The method according to claim 1, wherein, Determining the action of each agent using the MAPF-trained neural network involves inputting local field-of-view information of each agent into the neural network, wherein the local field-of-view information represents the state of neighboring vertices within a predetermined distance of the agent's current position.

6. The method according to claim 5, wherein, The predetermined distance is a predetermined radius R, which defines the local field of view of each agent as a square with a length of 2*R+1.

7. The method according to claim 5, wherein, Each unit has a unit type, and wherein the agent's local field of vision includes a central unit and a plurality of units surrounding the central unit, the central unit including the agent.

8. The method according to claim 7, wherein, Determining the action of each agent using the MAPF-trained neural network includes inputting one or more masks for the local field of view into the neural network, the one or more masks including at least one of an obstacle mask, a mask of the positions of all agents within the local field of view, a mask of the target position of other agents within the local field of view, and a mask of the target position of the agent.

9. The method according to claim 1, wherein, Determining the action of each agent using the MAPF-trained neural network includes inputting at least one of the following: distance information to the target of the agent, and the order of the other agents if they share the target location with the agent.

10. The method according to claim 1, wherein, Determining the action of each agent using the MAPF-trained neural network involves inputting the next target position information of each agent into the neural network.

11. The method according to claim 10, wherein, The next target location information for each agent includes information about the difference between the agent's current location and the agent's next target location.

12. The method according to claim 10, wherein, The next target location information for each agent includes at least one of the following: the time remaining until the agent's earliest arrival time, the time remaining until the target deadline, and whether the target deadline has passed.

13. The method according to claim 1, wherein, The neural network is trained using at least one of the following: a global positive reward for all agents when any agent completes the objective and a global negative reward for all agents when any agent is late for the deadline.

14. A diagnostic laboratory system, the diagnostic laboratory system comprising: Diagnostic laboratory equipment; One or more tracks connecting the diagnostic laboratory equipment; processor; as well as A memory coupled to the processor, the memory including computer-executable instructions, which, when executed by the processor, cause the processor to: Obtain a grid for the diagnostic laboratory system, the grid having cells including the diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment; Assign an agent to each sample carrier within the diagnostic laboratory system; For each agent assigned to a sample carrier holding a sample container, a target is determined, including an earliest arrival time constraint, a deadline time constraint, and a target processing time constraint. as well as The action of each agent is determined by a neural network trained with Multi-Agent Pathfinding (MAPF), which is trained using arrival, deadline, and priority constraints as well as non-instantaneous target processing time.

15. The diagnostic laboratory system of claim 14, further comprising a sample carrier controller, and wherein, The memory includes computer-executable instructions that, when executed by the processor, cause the processor to employ the sample carrier controller to perform one or more defined actions, including transferring a sample container for processing.

16. The diagnostic laboratory system of claim 14, wherein, The memory includes computer-executable instructions that, when executed by the processor, cause the processor to determine a set of objectives for each agent, each objective including an earliest arrival time constraint, a deadline constraint, and an objective processing time constraint.

17. The diagnostic laboratory system of claim 14, wherein, The memory includes computer-executable instructions that, when executed by the processor, cause the processor to input local field-of-view information for each agent into the neural network, the local field-of-view information representing the state of neighboring vertices within a predetermined distance of the agent's current position.

18. The diagnostic laboratory system of claim 17, wherein, The predetermined distance is a predetermined radius R, which defines the local field of view of each agent as a square with a length of 2*R+1.

19. The diagnostic laboratory system of claim 17, wherein, Each unit has a unit type, and wherein the agent's local field of vision includes a central unit and a plurality of units surrounding the central unit, the central unit including the agent.

20. The diagnostic laboratory system of claim 19, wherein, The memory includes computer-executable instructions that, when executed by the processor, cause the processor to input one or more masks for the local field of view into the neural network, the one or more masks including at least one of an obstacle mask, a mask of the positions of all agents within the local field of view, a mask of the target positions of other agents within the local field of view, and a mask of the target positions of the agents.

21. The diagnostic laboratory system of claim 14, wherein, The memory includes computer-executable instructions that, when executed by the processor, cause the processor to: for each agent, input distance information to a target of the agent, and the order of other agents if they share the target location with the agent.

22. The diagnostic laboratory system of claim 14, wherein, The memory includes computer-executable instructions that, when executed by the processor, cause the processor to: determine the action of each agent using the MAPF-trained neural network by inputting the next target position information of each agent into the neural network.

23. The diagnostic laboratory system according to claim 22, wherein, The next target location information for each agent includes information about the difference between the agent's current location and the agent's next target location.

24. The diagnostic laboratory system of claim 22, wherein, The next target location information for each agent includes at least one of the following: the time remaining until the earliest arrival time, the time remaining until the target deadline, and whether the target deadline has passed.

25. The diagnostic laboratory system according to claim 14, wherein, The neural network is trained using at least one of the following: a global positive reward for all agents when any agent completes the objective and a global negative reward for all agents when any agent is late for the deadline.

26. A method for pathfinding of sample containers in a diagnostic laboratory system, the method comprising: Create a grid for a diagnostic laboratory system, the grid having cells including diagnostic laboratory equipment and one or more tracks connecting the diagnostic laboratory equipment; Create multiple intelligent agents, each representing a sample carrier within the diagnostic laboratory system; For each agent assigned to a sample carrier holding a sample container, a target is determined, the target including an earliest arrival time constraint, a deadline constraint, and a target processing time constraint; Create a neural network that takes the agent's earliest arrival time constraint, deadline constraint, and target processing time constraint as input, and takes the agent's action or the probability of the action as output. The neural network is trained using arrival and deadline times, priority constraints, and non-instantaneous target processing time. as well as A pathfinding program is created, which uses a trained neural network to generate actions for each agent within the diagnostic laboratory system.

27. The method of claim 26, further comprising deploying the trained neural network and the pathfinding procedure for use by the diagnostic laboratory system.

28. The method of claim 27, further comprising employing the pathfinding procedure to determine one or more actions, including transferring the sample container for processing.

29. The method according to claim 26, wherein, The neural network is configured to input local field-of-view information for each agent into the neural network, the local field-of-view information representing the state of neighboring vertices within a predetermined distance of the agent's current position.

30. The method according to claim 29, wherein, The neural network is configured to input one or more masks for the local field of view into the neural network, the one or more masks including at least one of an obstacle mask, a mask of the positions of all agents within the local field of view, a mask of the target positions of other agents within the local field of view, and a mask of the target positions of the agents.

31. The method according to claim 26, wherein, The neural network is configured such that, for each agent, at least one of the following is input: distance information to the target of the agent, and the order of the other agents if they share the target location with the agent.

32. The method according to claim 26, wherein, The neural network is configured to take into account the next target location information for each agent.

33. The method according to claim 32, wherein, The next target location information for each agent includes information about the difference between the agent's current location and the agent's next target location.

34. The method according to claim 32, wherein, The next target location information for each agent includes at least one of the following: the time remaining until the agent's earliest arrival time, the time remaining until the target deadline, and whether the target deadline has passed.

35. The method of claim 26, further comprising: The neural network is trained using at least one of a global positive reward for all agents when any agent completes the objective and a global negative reward for all agents when any agent is late for the deadline.

Images