Automatically generating commissioning exemplars for lighting configuration plan generation

Abstract

A method for commissioning a lighting system comprising a plurality of luminaires corresponding to a premises is provided. The method includes retrieving a floor plan and one or more boundary conditions. The floor plan and the boundary conditions correspond to the premises. The method further includes selecting one or more commissioning exemplars from a demonstration pool comprising a plurality of commissioning exemplars. The commissioning exemplars are selected based on a similarity metric regarding the commissioning exemplars and the floor plan and / or the one or more boundary conditions. Each of the commissioning exemplars comprises a portion of chain-of- thought reasoning steps leading from an example floor plan to a ground truth lighting configuration plan. The method further includes generating, via an LLM, a lighting configuration plan based on the floor plan, the boundary conditions, and the commissioning exemplars.

Description

[0001] AUTOMATICALLY GENERATING COMMISSIONING EXEMPLARS FOR LIGHTING

[0002] CONFIGURATION PLAN GENERATION

[0003] FIELD OF THE DISCLOSURE

[0004] The present disclosure is generally directed to generating lighting configuration plans using large language models (LLMs), and, more particularly, to automatically generating commissioning exemplars to augment lighting configuration plan generation.

[0005] BACKGROUND

[0006] The accuracy of Large Language models (LLMs) depends heavily on the selection and permutation of exemplars submitted to the LLM alongside a user’s prompt. Accordingly, significant efforts have been invested in developing approaches to select appropriate exemplars. Further, implementing randomly selected exemplars from a database is generally not beneficial, even if the exemplar is known to be correctly answered. Previous work in this field has identified a number of methods for choosing exemplars.

[0007] Additionally, LLMs have been previously used for lighting tasks. In these examples, the LLMs were provided with chain-of-thought (COT) prompting and manually constructed exemplars. However, the inherent subjectivity of human annotations makes manually constructing exemplars costly, sub-optimal, and highly inconsistent, especially for more complex reasoning tasks requiring a multitude of reasoning steps.

[0008] SUMMARY OF THE DISCLOSURE

[0009] The present disclosure is generally directed to commissioning a lighting system of a plurality of luminaires by generating a lighting configuration plan for a premises using a large language model (LLM) and one or more automatically generated commissioning exemplars. The lighting configuration plan includes a plurality of luminaire parameters defining the locations of the luminaires, the configuration (light output, color temperature, lighting controls, etc.) of the luminaires, and the types of luminaires to be implemented in the premises. The commissioning exemplars are automatically generated using the same LLM by implementing chain-of-thought (COT) prompting with an example floor plan associated with a ground truth lighting configuration plan. The example floor plan may be chosen using a helper artificial intelligence (Al) model. When provided with the example floor plan, the LLM generates a COT lighting configuration plan and COT reasoning steps. The COT reasoning steps explain the reasoning implemented by the LLM to generate the COT lighting configuration plan. The COT lighting configuration plan is then evaluated for accuracy against the corresponding ground truth lighting configuration plan. If the COT lighting configuration plan is determined to be accurate, a commissioning exemplar is generated based on the accurate COT lighting configuration plan and the example floor plan. If the COT lighting configuration plan is determined to be inaccurate, please-revise prompting is used (without additional guidance or information) to generate a new COT lighting configuration plan and new COT reasoning steps. In please-revise prompting, the LLM is informed that the COT lighting configuration plan is inaccurate, and is asked to try again to generate an accurate COT lighting configuration plan without further guidance. Repeatedly implementing please-revise prompting causes the LLM to eventually develop an accurate COT lighting configuration plan. Accordingly, please-revise prompting is used until the COT lighting configuration plan is determined to be accurate, and the corresponding COT reasoning steps are used to generate a commissioning exemplar. The commissioning exemplar is then added to a demonstration pool of a plurality of commissioning exemplars to be used to generate the new lighting configuration plan. One or more commissioning exemplars (such as five to eight) are then selected from the demonstration pool based on a similarity metric comparing the commissioning exemplars to the floor plan and boundary conditions. The selected commissioning exemplars are then provided to the LLM with the floor plan and the boundary conditions to influence the generation of the lighting configuration plan.

[0010] The LLM used in both the demonstration pool construction and lighting configuration plan inference may, in some examples, be a Generative Pre-trained Transformer 4 (GPT-4) model. The COT prompting used to generate the COT lighting configuration plans and the COT reasoning steps may be zero-shot COT. An LLM temperature parameter may be used to control the randomness of the generated COT lighting configuration plans and COT reasoning steps. Further, once the COT lighting configuration plan is determined to be accurate, summary-prompting may be used to generate a final COT reasoning chain. The final COT reasoning chain may then be used to generate the commissioning exemplar. While any number of commissioning exemplars may be used generate the new lighting configuration plan, in a preferred example, five to eight commissioning exemplars may be used. In some examples, the quantity of commissioning exemplars used to generate the lighting configuration plan may be determined based on quantity criteria, such as task difficulty, LLM accuracy, extent of human review, etc.

[0011] In some examples, the commissioning exemplars are further chosen based on a difficulty level associated with the commissioning exemplars. The difficulty level of each commissioning exemplar may correspond to (i) a step count of the COT reasoning steps used to generate the commissioning exemplar and / or (ii) a word count of each of the COT reasoning steps. In other examples, the repetitions of the please-revise prompting to generate an accurate COT lighting configuration plan may be tracked by an attempt count. The attempt count may be associated with the corresponding commissioning exemplar and subsequently used to select one or more commissioning exemplars from the demonstration pool. Further, the total amount of attempts to generate the accurate COT lighting configuration plan may be limited by an attempt threshold. In some examples, the attempt count and / or the attempt threshold may be weighted based on computational resources used by the LLM for each attempt. Once the attempt count exceeds the attempt threshold, the COT lighting configuration plan may be (1) discarded or (2) used to generate a commissioning exemplar depending on the COT accuracy of the lighting configuration plan.

[0012] The accuracy of the COT lighting configuration plan may be evaluated in different ways. In some examples, the ground truth data may also include one or more gold labels. The gold labels are human-generated labels regarding the human-based reasoning used to generate the corresponding ground truth lighting configuration plan. In other examples, a deviation metric is determined by comparing the COT lighting configuration plan and the ground truth lighting configuration plan. The deviation metric is determined by comparing the COT luminaire parameters to ground truth luminaire parameters associated with the ground truth lighting configuration plan. The deviation metric is then compared to a deviation threshold.

[0013] While the commissioning exemplars described above are generated based on accurate COT reasoning steps, in some cases, “known-to-be-wrong” commissioning exemplars may also be generated based on faulty reasoning. In this example, the commissioning LLM may generate a self-reflection statement based on the inaccurate COT reasoning steps. The self-reflection statement may then be used with the inaccurate COT reasoning steps to generate a “how not to do it” commissioning exemplar to be added to the demonstration pool.

[0014] In other examples, a second, more robust LLM may be used to generate all possible ground truth lighting configuration plans. The accuracy of the COT lighting configuration plan is then determined by comparing the COT lighting configuration plan to all of the ground truth lighting configuration plans.

[0015] In some examples, a dataset including a large array of ground truth data may be used. The ground truth data may be clustered into different types of premises, such as food retail, big box store, fashion retail, e-commerce warehouse, etc. Accordingly, a set of exemplars may then be derived from this ground truth data to more accurately generate a lighting configuration plan for a similar type of premises.

[0016] Generally, in one aspect, a method for commissioning a lighting system comprising a plurality of luminaires corresponding to a premises. The method includes retrieving a floor plan and one or more boundary conditions. The floor plan and the one or more boundary conditions correspond to the premises.

[0017] The method further includes selecting one or more commissioning exemplars from a demonstration pool comprising a plurality of commissioning exemplars. The one or more commissioning exemplars are selected based on a similarity metric regarding the one or more commissioning exemplars and the floor plan and / or the one or more boundary conditions. Each of the plurality of commissioning exemplars comprises a portion of COT reasoning steps leading from an example floor plan to a ground truth lighting configuration plan.

[0018] The method further includes generating, via an LLM, a lighting configuration plan based on the floor plan, the one or more boundary conditions, and the one or more commissioning exemplars.

[0019] According to an example, each of plurality of commissioning exemplars of the demonstrations pool are generated by: (1) querying the LLM using prompting with the example floor plan and / or the one or more boundary conditions to generate the COT reasoning steps and a COT lighting configuration plan; (2) determining a COT accuracy of the COT lighting configuration plan by, at least in part, comparing the COT lighting configuration plan to the ground truth lighting configuration plan corresponding to the example floor plan and / or the one or more boundary conditions; (3) if the COT accuracy is above an accuracy threshold, generating a commissioning exemplar comprising at least a portion of the COT reasoning steps; and (4) if the COT accuracy is below the accuracy threshold, repeatedly querying the LLM to update the COT reasoning steps and the COT lighting configuration plan until the COT accuracy is above the accuracy threshold.

[0020] According to an example, following the determination that the COT accuracy is above the accuracy threshold, the LLM is queried to produce a final COT reasoning chain based on the COT reasoning steps.

[0021] According to an example, the commissioning exemplar is further generated based at least in part on the final COT reasoning chain and the example floor plan.

[0022] According to an example, the prompting is zero-shot COT prompting.

[0023] According to an example, the COT accuracy of the COT lighting configuration plan is determined further based on one or more gold labels.

[0024] According to an example, the COT lighting configuration plan comprises one or more COT luminaire parameters generated by the LLM via COT reasoning. The one or more COT luminaire parameters comprise position data, configuration data, type data, and / or installation data.

[0025] According to an example, comparing the COT lighting configuration plan to the ground truth lighting configuration plan comprises: (1) determining a deviation metric between one or more must have parameters of the one or more COT luminaire parameters and one or more corresponding ground truth luminaire parameters of the ground truth lighting configuration plan; and (2) comparing the deviation metric to a deviation threshold to determine the COT accuracy of the COT lighting configuration plan.

[0026] According to an example, an attempt count is iterated for each repetition of querying the LLM to update the COT reasoning steps and the COT lighting configuration plan. The attempt count is assigned to the commissioning exemplar generated based on the COT lighting configuration plan. A first attempt count associated with a first commissioning exemplar differs from a second attempt count associated with a second commissioning exemplar.

[0027] According to an example, repeatedly querying of the LLM to update the COT reasoning steps and the COT lighting configuration plan ceases if the attempt count exceeds an attempt threshold.

[0028] According to an example, the COT lighting configuration plan is discarded when the attempt count exceeds the attempt threshold.

[0029] According to an example, a commissioning exemplar comprising at least a portion of the COT reasoning steps is generated when the attempt count exceeds the attempt threshold. According to an example, the attempt threshold is generated by a helper artificial intelligence (Al) model trained on one or more of the plurality of commissioning exemplars of the demonstration pool.

[0030] According to an example, the attempt count and / or the attempt threshold is weighted according to computational resources required by the LLM for each repeating querying of the LLM.

[0031] According to an example, the LLM is queried to generate a self-reflection statement if the COT accuracy is below the accuracy threshold. The self-reflection statement describes faulty reasoning associated with the COT lighting configuration plan. The commissioning exemplar is based at least in part on the COT reasoning steps and the selfreflection statement.

[0032] According to an example, the COT reasoning steps are further generated based on LLM temperature.

[0033] According to an example, a second LLM model generates a plurality of synthetic lighting configuration plans based on the ground truth lighting configuration plan and the example floor plan. The COT accuracy of the COT lighting configuration plan is further determined by comparing the COT lighting configuration plan to each of the plurality of synthetic lighting configuration plans.

[0034] According to an example, if the COT accuracy is below the accuracy threshold, repeatedly querying of the LLM informs the LLM that the COT lighting configuration plan is inaccurate.

[0035] According to an example, the example floor plan is selected by a helper Al model.

[0036] According to an example, the one or more commissioning exemplars are chosen from the demonstration pool based at least in part on a difficulty level assigned to each of the plurality of commissioning exemplars.

[0037] According to an example, the difficulty level assigned to each of the plurality of commissioning exemplars is based on a step count of the COT reasoning steps corresponding to each of the plurality of commissioning exemplars.

[0038] According to an example, the difficulty level assigned to each of the plurality of commissioning exemplars is based on a word count of each of the COT reasoning steps corresponding to each of the plurality of commissioning exemplars.

[0039] According to an example, the plurality of commissioning exemplars are derived from a one or more clusters of ground truth data within a dataset. According to an example, a quantity of commissioning exemplars used to generate the lighting configuration plan is determined based on quantity criteria.

[0040] Generally, in another aspect, a system for commissioning a lighting system comprising a plurality of luminaires corresponding to a premises is provided. The system includes a controller configured to retrieve a floor plan and one or more boundary conditions. The floor plan and the one or more boundary conditions correspond to the premises.

[0041] The controller is further configured to select one or more commissioning exemplars from a demonstration pool comprising a plurality of commissioning exemplars. The one or more commissioning exemplars are selected based on a similarity metric regarding the one or more commissioning exemplars and the floor plan and / or the one or more boundary conditions, wherein each of the plurality of commissioning exemplars comprises a portion of COT reasoning steps leading from an example floor plan to a ground truth lighting configuration plan.

[0042] The controller is further configured to generate, via an LLM, a lighting configuration plan based on the floor plan, the one or more boundary conditions, and the one or more commissioning exemplars.

[0043] According to an example, each of plurality of commissioning exemplars of the demonstrations pool are generated by: (1) querying the LLM using prompting with the example floor plan to generate the COT reasoning steps and a COT lighting configuration plan; (2) determining a COT accuracy of the COT lighting configuration plan by, at least in part, comparing the COT lighting configuration plan to the ground truth lighting configuration plan corresponding to the example floor plan; (3) if the COT accuracy is above an accuracy threshold, generating a commissioning exemplar comprising at least a portion of the COT reasoning steps; and (4) if the COT accuracy is below the accuracy threshold, repeatedly querying the LLM to update the COT reasoning steps and the COT lighting configuration plan until the COT accuracy is above the accuracy threshold.

[0044] It should be appreciated that all combinations of the foregoing concepts and additional concepts discussed in greater detail below (provided such concepts are not mutually inconsistent) are contemplated as being part of the inventive subject matter disclosed herein. In particular, all combinations of claimed subject matter appearing at the end of this disclosure are contemplated as being part of the inventive subject matter disclosed herein. It should also be appreciated that terminology explicitly employed herein that also may appear in any disclosure incorporated by reference should be accorded a meaning most consistent with the particular concepts disclosed herein. In various implementations, a processor or controller may be associated with one or more storage media (generically referred to herein as “memory,” e.g., volatile and non-volatile computer memory such as RAM, PROM, EPROM, EEPROM, floppy disks, compact disks, optical disks, magnetic tape, SSD, etc.). In some implementations, the storage media may be encoded with one or more programs that, when executed on one or more processors and / or controllers, perform at least some of the functions discussed herein. Various storage media may be fixed within a processor or controller or may be transportable, such that the one or more programs stored thereon can be loaded into a processor or controller so as to implement various aspects as discussed herein. The terms “program” or “computer program” are used herein in a generic sense to refer to any type of computer code (e.g., software or microcode) that can be employed to program one or more processors or controllers.

[0045] These and other aspects of the various embodiments will be apparent from and elucidated with reference to the embodiment s) described hereinafter.

[0046] BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In the drawings, like reference characters generally refer to the same parts throughout the different views. Also, the drawings are not necessarily to scale, emphasis instead generally being placed upon illustrating the principles of the various embodiments.

[0048] Fig. l is a functional block diagram of a lighting commissioning system in an inference stage, in accordance with an example.

[0049] Fig. 2 is a functional block diagram of a lighting commissioning system in a commissioning exemplar construction stage, in accordance with an example.

[0050] Fig. 3 is a variation of the functional block diagram of FIG. 2, wherein gold labels are implemented to generate commissioning exemplars, in accordance with an example.

[0051] Fig. 4 is a variation of the functional block diagram of FIG. 2, wherein commissioning exemplars are assigned a difficulty level, in accordance with an example.

[0052] Fig. 5 is a variation of the functional block diagram of FIG. 2, wherein a deviation metric is implemented to generate commissioning exemplars, in accordance with an example.

[0053] Fig. 6 is a variation of the functional block diagram of FIG. 2, wherein an attempt count is implemented to generate commissioning exemplars, in accordance with an example. Fig. 7 is a variation of the functional block diagram of FIG. 2, wherein a selfreflection statement is incorporated into commissioning exemplars, in accordance with an example.

[0054] Fig. 8 is a variation of the functional block diagram of FIG. 2, wherein a large- language model (LLM) temperature is used to generate commissioning exemplars, in accordance with an example.

[0055] Fig. 9 is a variation of the functional block diagram of FIG. 2, wherein synthetic lighting configuration plans are used to generate commissioning exemplars, in accordance with an example.

[0056] Fig. 10 is a variation of the functional block diagram of FIG. 2, wherein ground truth data is organized into clusters to generate commissioning exemplars, in accordance with an example.

[0057] Fig. 11 is a variation of the functional block diagram of FIG. 2, wherein the example floor plan is chosen by a helper artificial intelligence (Al) model, in accordance with an example.

[0058] Fig. 12 is a schematic view of a memory of a controller for commissioning a lighting system, in accordance with an example.

[0059] Fig. 13 is a schematic view of a processor of the controller for commissioning a lighting system, in accordance with an example.

[0060] Fig. 14 is a flowchart of a method for commissioning a lighting system, in accordance with an example.

[0061] Fig. 15 is a further flowchart of a method for commissioning a lighting system, in accordance with an example.

[0062] DETAILED DESCRIPTION OF EMBODIMENTS

[0063] The present disclosure is generally directed to commissioning a lighting system of a plurality of luminaires by generating a lighting configuration plan for a premises using a large language model (LLM) and one or more automatically generated commissioning exemplars. The lighting configuration plan includes a plurality of luminaire parameters defining the locations of the luminaires, the configuration of the luminaires, and the type of luminaires. The commissioning exemplars are automatically generated using the same LLM by implementing chain-of-thought (COT) prompting with an example floor plan associated with a ground truth lighting configuration plan. When provided with the example floor plan, the LLM generates a COT lighting configuration plan and COT reasoning steps. The COT reasoning steps explain the reasoning implemented by the LLM to generate the COT lighting configuration plan. The COT lighting configuration plan is then evaluated for accuracy against the corresponding ground truth lighting configuration plan. If the COT lighting configuration plan is determined to be accurate, a commissioning exemplar is generated based on the accurate COT lighting configuration plan and the example floor plan. If the COT lighting configuration plan is determined to be inaccurate, please-revise prompting is used (without additional guidance or information) to generate a new COT lighting configuration plan and new COT reasoning steps. Accordingly, please revising prompting is used until the COT lighting configuration plan is determined to be accurate, and the corresponding COT reasoning steps are used to generate a commissioning exemplar. The commissioning exemplar is then added to a demonstration pool of a plurality of commissioning exemplars to be used to generate the new lighting configuration plan. One or more commissioning exemplars are then selected from the demonstration pool based on a similarity metric. The selected commissioning exemplars are then provided to the LLM with the floor plan and the boundary conditions to influence the generation of the lighting configuration plan.

[0064] Turning now to the figures, FIG. l is a functional block diagram showing aspects of a controller 100 for commissioning a lighting system. Aspects of the controller 100 are shown in more detail in FIG. 12. Broadly, the controller 100 includes a memory 125 for storing data and a processor 175 for processing data. The controller 100 may also include a receiver, transmitter, and / or a transceiver for receiving and / or transmitting data. Further, while the controller 100 is depicted in FIG. 12 as a single component, in some examples, the various functions of the controller 100 may be distributed across more than one component. The lighting system includes a plurality of luminaires corresponding to a premises. The luminaires may be any type of lighting device such as lamps, light emitting diode (LED) strips, lasers, etc. The premises may be any possible type of location or area, such as an office space, a grocery store, a retail shop, a big box store, etc. The premises may be an indoor space, an outdoor space, or a combination of indoor and outdoor spaces. The outdoor spaces could include a parking lot, a sports field, or a street which has artificial lighting. In particular, the commissioning process involves generating a lighting configuration plan 116. The lighting configuration plan 116 defines a plurality of luminaire parameters 166 defining various properties of the individual luminaires. The luminaire parameters 166 may include position data 156. The position data 156 defines the location of each of the plurality of luminaires within the premises. The position data 156 may represent where a previously installed luminaire is currently located and / or where a yet-to-be-installed luminaire should be located upon installation. In some examples, the position data 156 may be embodied as a series of two-dimensional or three-dimensional coordinates. The position data 156 may also define the orientation of the luminaire at its installed location.

[0065] The luminaire parameters 166 may also include configuration data 158. The configuration data 158 may control the various lighting output properties of each of the luminaires, such as intensity, on / off schedule, color temperature, etc. For example, the configuration data 158 could define the initial color temperature of a luminaire upon installation, as well as a range of color temperature settings which may be adjusted via a user input. The user input may be a switch, such as a dual in-line package (DIP) switch. The configuration data 158 may also define one or more modes comprising a variety of lighting settings. For example, the configuration data 158 may define one or more energy efficiency modes, which define various lighting settings (such as lighting intensity and / or on / off schedule) based on the desired level of efficiency.

[0066] The luminaire parameters 166 may also include type data 186, indicative of the variety of luminaire(s) to be used at the locations defined by the position data 156. The type data 186 may broadly define a variety of luminaire or light fixture (such as downlight, wall light, ceiling light, recessed light, etc.). The type data 186 may also more specifically define a manufacturer and / or a model number of the luminaire to be used.

[0067] The luminaire parameters 166 may also include installation data 190 regarding parameters or instructions for wiring, arranging, and / or orienting the luminaires. For example, the installation data 188 may define wiring plans for Digital Addressable Lighting Interface (DALI) wiring. The installation data 188 may also describe how down from a ceiling a suspended luminaire should be installed.

[0068] In some examples, the luminaires have yet to be installed in the premises. Accordingly, in these examples, the commissioning process provide instructions regarding creating the lighting system from nothing, including indications of where to install the luminaires within the premises according to the position data 156. In other examples, the luminaires may have already been installed within the premises, and the commissioning process reconfigures the luminaires, such as to operate according to different settings. According to this example, the commissioning process could be used to reconfigure (but not necessarily relocate or reposition) luminaires when a first tenant moves out of the premises, and a second tenant moves in. As shown in FIG. 1, the controller 100 generates the lighting configuration plan 116 for a specific premises using a LLM 101. Using the LLM 101 to generate the lighting configuration plan 116 may be considered to be an inference stage. As will be demonstrated below, the commissioning LLM 101 is also used in a construction stage to generate data subsequently used by the commissioning LLM 101 to generate the lighting configuration plan 116. In some examples, the LLM 101 may be a multimodal LLM. A multimodal LLM may be able to process and understand data from several different types of input sources, such as text data, image data, audio data, etc. In some examples, the LLM 101 is a Generative Pre-trained Transformer 4 (GPT-4) model.

[0069] Three types of data are used by the LLM to generate the lighting configuration plan: a floor plan 102, one or more boundary conditions 104, and one or more commissioning exemplars 106. The floor plan 102 is a two- or three-dimensional representation of the premises (office space, big box store, apartment building, etc.) within which the lighting system will be installed. Accordingly, the floor plan 102 provides physical limitations and constraints for the design of the lighting configuration plan 116 by defining the physical space in which the luminaires of the lighting system must be positioned. In some examples, the three-dimensional aspects of the floor plans 102 may be generated using Light Detection and Ranging (LiDAR) data or Building Information Modelling (BIM). The boundary conditions 104 define additional limitations for the lighting configuration plan 116 beyond the floor plan 102. For instance, the boundary conditions 104 could define certain areas of the floor plans 102 which should be more or less illuminated. Further, the boundary conditions 104 could define certain areas where luminaires cannot be positioned for practical or safety reasons. The floor plan 102 and / or boundary conditions 104 may be retrieved from an external database or storage device.

[0070] Further, one or more commissioning exemplars 106 are retrieved from a demonstration pool 108 to be provided to the LLM 101 along with the floor plan 102 and the boundary conditions 104. As will be described below (such as in relation to FIG. 2), the commissioning exemplars 106 are automatically generated based on COT reasoning steps 118 corresponding to known, ground truth data 110, in particular, example floor plans 112 and their corresponding ground truth lighting configuration plans 114. Accordingly, the commissioning exemplars 106 provide examples to the LLM 101 of the reasoning required to generate accurate ground truth lighting configuration plans 114, so that the LLM 101 may then generate new lighting configuration plans 116 for previously unknown floor plans 102. The demonstration pool 108 may contain a larger number of commissioning exemplars 106 based on a wide range of types of example floor plans 112. Accordingly, in some cases, it may be more efficient to use a small number (such as 5 to 8) of commissioning exemplars 106 associated with example floor plans 112. Therefore, the commissioning exemplars 106 may be selected based on a similarity metric 168. The similarity metric 168 may be determined by comparing the floor plan 102 (and any associated boundary conditions) to the example floor plans 112 associated with each of the commissioning exemplars 106. Generally, the similarity metric 168 evaluates the equivalence of the premises associated with the floor plan 102 and the example floor plans 112. The similarity metric 168 may also evaluate the compatibility of the lighting effects resulting from the commissioning exemplars 106 to the floor plan 102 and / or the boundary conditions 104 provided to the LLM 101. The similarity metric 168 may take a wide range of factors into account, such as type of floor plan (retail store, restaurant, office space, etc.), overall size, overall dimensions, etc. The commissioning exemplars 106 associated with the most similar example floor plans 112 may then be provided to the LLM 101 to help generate appropriate lighting configuration plans 116 for the provided floor plan 102. Further, in some examples, the quantity of commissioning exemplars 106 used to generate the new lighting configuration plan 116 may be determined according to a set of quantity criteria 172. The quantity criteria 172 may include a wide range of factors, such as the difficulty of the task (generating a new lighting configuration plan 116 for the specific floor plan 102) or a known accuracy parameter of the LLM 101. The quantity criteria 172 may also include an extent of human review metric. This metric may quantify the extent (in terms of time or other measures) a human will review the lighting configuration plan 116 output by the LLM 101. For example, if a human will carefully review the lighting configuration plan 116 following generation, less commissioning exemplars 106 may be needed, as the human is likely to detect errors within the lighting configuration plan 116 and re-run the LLM 101. In other examples where a human will not be reviewing the lighting configuration plan 116, more commissioning exemplars 106 should be used to ensure lighting configuration plan 116 is accurate.

[0071] FIG. 2 broadly illustrates how the commissioning exemplars 106 are automatically generated. Notably, the commissioning exemplars 106 are generated using the same LLM 101 as is used to subsequently generate the lighting configuration plan 116 for the new floor plan 102 as shown in FIG. 1. As previously described, the LLM 101 of FIG. 2 may be described as operating in a construction stage. As shown in FIG. 2, the commissioning exemplars 106 are generated based on the ground truth data 110. The ground truth data 110 includes a number of example floor plans 112 associated with known-to-be-accurate ground truth lighting configuration plans 114.

[0072] In the example of FIG. 2, one example floor plan 112 from the ground truth data 110 is provided to LLM 101. Further, in some examples, the one or more boundary conditions 104 used during the inference stage may also be provided to the LLM 101 during the construction stage. In providing the example floor plan 112 to the LLM 101, zero-shot COT prompting is used in a query to the LLM 101. As a result of the query, the LLM 101 generates a COT lighting configuration plan 120 based on the example floor plan 112 and the one or more boundary conditions 104. The LLM 101 also generates a series of COT reasoning steps 118 explaining how the LLM 101 generated the COT lighting configuration plan 120. Like the new lighting configuration plan 116 generated by the LLM 101 during the inference stage, the COT lighting configuration plan 120 is defined by a series of COT luminaire parameters 130, including COT position data 132, COT configuration data 134, COT type data 184, and COT installation data 188. A comparator 103 then determines whether or not the COT lighting configuration plan 120 is accurate or not. The accuracy of the COT lighting configuration plan 120 is determined by comparing the COT lighting configuration plan 120 to the ground truth lighting configuration plan 114 associated with the example floor plan 112. If the COT lighting configuration plan 120 is accurate, the corresponding COT reasoning steps 118 are also presumed to be accurate. Thus, the presumed to be accurate COT reasoning steps 118 may then be used to create a commissioning exemplar 106.

[0073] As shown in FIG. 2, the comparator 103 provides prompting feedback 150 to the LLM 101 based on the comparison of the COT lighting configuration plan 120 to the ground truth lighting configuration plan 114. If the comparator 103 determines that the COT lighting configuration plan 120 is accurate, the comparator 103 provides a summary -prompt 162 to the LLM 101. The summary -prompt 162 causes the LLM 101 to generate a final COT reasoning chain 122 summarizing and / or streamlining the reasoning used by the LLM 101 to generate the accurate COT lighting configuration plan 120 from the example floor plan 112. The example floor plan 112, the COT reasoning steps 118, and the final COT reasoning chain 122 are then combined to form a new commissioning exemplar 106. The new commissioning exemplar 106 is stored in the demonstration pool 108 for retrieval during the inference stage. In some alternative examples, the step of providing the summary -prompt 162 to the LLM 101 may be omitted, and the commissioning exemplar 106 may be solely based on example floor plan 112 and the COT reasoning steps 118. If the comparator 103 determines that the COT lighting configuration plan 120 is inaccurate, the comparator 103 prompts the LLM 101 to try again and generate new COT reasoning steps 118 and a new COT lighting configuration plan 120 by using different reasoning that in previous, inaccurate attempts. In some examples, the comparator 103 may query the LLM 101 using please-revise prompting 160. In please-revise prompting 160, the LLM 101 is informed that it has generated an inaccurate COT lighting configuration plan 120, and asked to try again. However, the LLM 101 is not provided with any guidance regarding which COT reasoning steps 118 may have led to the inaccurate COT lighting configuration plan 120. Rather, the please-revise prompting 160 is used to repeatedly query the LLM 101 to generate new COT reasoning steps 118 and a new COT lighting configuration plan 120 until the COT lighting configuration plan 120 is determined to be accurate.

[0074] Whether or not the COT lighting configuration plan 120 is accurate is determined by calculating a COT accuracy 170 by comparing the COT lighting configuration plan 120 to the ground truth lighting configuration plan 114. The COT accuracy 170 may be determined by any practical comparison means and represented on any practical scale. For example, the COT accuracy 170 may be represented as a percentage, such as 90% accurate. An example of determining COT accuracy 170 is described below with reference to FIG. 5. The COT accuracy 170 is then compared to a COT accuracy threshold 174 to determine if the COT lighting configuration is accurate. For example, if the COT accuracy 170 is 90%, and the COT accuracy threshold 174 is 85%, the COT lighting configuration plan 120 is considered to be accurate, and the associated COT reasoning steps 118 may be used to generate a commissioning exemplar 106. However, if the COT accuracy 170 is 90%, and the COT accuracy threshold 174 is 95%, the COT lighting configuration plan 120 is considered to be inaccurate, and the LLM 101 is prompted to try again until an accurate COT lighting configuration plan 120 is generated.

[0075] FIG. 3 is a variation of the functional block diagram of FIG. 2, wherein gold labels 124 are implemented to generate commissioning exemplars 106. As shown in FIG. 3, the ground truth lighting configuration plans 114 of the ground truth data 110 also include gold labels 124. The gold labels 124 are manually added to the ground truth lighting configuration plans 114 to provide indications regarding various reasoning steps used to determine the ground truth lighting configuration plan 114. Accordingly, in the non-limiting example of FIG. 3, the comparator 103 determines a value of the COT accuracy 170 of the COT lighting configuration plan 120 not only by comparing the COT lighting configuration plan 120 to the ground truth lighting configuration plan 114, the comparator 103 also compares the gold labels 124 associated with the ground truth lighting configuration plan 114 to the COT reasoning steps 118 used to arrive at the COT lighting configuration plan 114. Thus, if the COT reasoning steps 118 are missing crucial reasoning steps as defined by the gold labels 124, the value of the COT accuracy will decrease.

[0076] FIG. 4 is a variation of the functional block diagram of FIG. 2, wherein commissioning exemplars 106 are assigned a difficulty level 126. The difficulty level 126 may be representative of the complexity of the reasoning steps used to generate the COT lighting configuration plan 120. Accordingly, commissioning exemplars 106 having lower difficulty levels 126 are considered to be less complex than other commissioning exemplars 106. As shown in FIG. 4, the difficulty level 126 is associated with the commissioning exemplar 106 generated based on the example floor plan 112, the COT reasoning steps 118, and the final COT reasoning chain 122. The difficulty level 126 may then be used to determine which commissioning exemplars 106 to retrieve from the demonstration pool 108 during the inference stage. For example, it may be advantageous to utilize lower complexity commissioning exemplars 106 during interference to reduce the computing resources required. The difficulty level 126 may be associated with a step count 128 and / or a word count 176 associated with the COT lighting configuration plan 120. The step count 128 corresponds to the number of COT reasoning steps 118 used to arrive at the COT lighting configuration plan 120. Similarly, the word count 176 corresponds to the total number of words in each of the COT reasoning steps 118.

[0077] FIG. 5 is a variation of the functional block diagram of FIG. 2, wherein a deviation metric 136 is implemented to generate commissioning exemplars 106. In particular, the deviation metric 136 is used to determine the COT accuracy 170 of the COT lighting configuration plan 120. The deviation metric 136 quantifies the total difference between the COT luminaire parameters 130 of the COT lighting configuration plan 120 and the ground truth luminaire parameters 164 of the ground truth lighting configuration plan 114. Like the COT luminaire parameters 130, the ground truth luminaire parameters 164 may include position data, configuration data, and type data. The deviation metric 136 is then compared to a deviation threshold 140 to determine a value for the COT accuracy 170; the closer the deviation metric 136 is to the deviation threshold 140, the greater the value of the COT accuracy 170. For example, if the deviation metric 136 is much higher than the deviation threshold 140, the COT accuracy 170 would be very low, such as 5%. However, if the deviation metric 136 is very close to the deviation threshold 140, the COT accuracy 170 would be very high, such as 95%.

[0078] As further shown in FIG. 5, the comparator 103 may also be provided one or more must-have parameters 138. The must-have parameters 138 describe a subset of COT luminaire parameters 130 which must be accurate. For example, the must-have parameters 138 may require the location data 132 of the COT lighting configuration plan 120 must be accurate, while the other COT luminaire parameters 130 are less critical. Thus, in some examples, the deviation metric 136 may only quantify the difference between the location data 132 of the COT lighting configuration 120 and the corresponding location data of the ground truth lighting configuration plan 114. In other examples, difference between the location data 132 of the COT lighting configuration 120 and the corresponding location data of the ground truth lighting configuration plan 114 may be weighted more heavily in quantifying the deviation metric 136 than the difference between other varieties of parameters.

[0079] FIG. 6 is a variation of the functional block diagram of FIG. 2, wherein an attempt count 142 is used in the process to generate the commissioning exemplars 106. As shown in FIG. 6, the LLM 101 iterates an attempt count 142 for each iteration of the COT lighting configuration plan 120. The attempt count 142 may be a simple integer value, such as 3 attempts, 5 attempts, 10 attempts, etc. This attempt count 142 may be incorporated into the commissioning exemplar 106 generated in FIG. 6, such that each of the commissioning exemplars 106 stored in the demonstration pool 108 is associated with an attempt count 142. Accordingly, the value of the attempt counts 142 associated with the commissioning exemplars 106 of the demonstration pool 108 may vary. In some examples, during the inference stage, the selected commissioning exemplars 106 will each be associated with different attempt counts 142 to provide a more robust selection of commissioning exemplars 106. Generally, it would be beneficial to include at least some commissioning exemplars 106 with relatively high attempt counts 142 in the demonstration pool 108 as examples of how to generate ground truth lighting configuration plans 120 for particularly complicated or challenging example floor plans 104.

[0080] In some examples, the comparator 103 is provided with an attempt threshold 144. The attempt threshold 144 may be used to limit the number of attempts executed by the LLM 101 to generate an accurate COT lighting configuration plan 120. If the attempt count 142 breaches the attempt threshold 144, the comparator 103 may stop repeatedly querying the LLM 101 with please-revise prompts to generated new COT lighting configuration plans 120 based on the currently selected example floor plan 112, and a new example floor plan 112 may be selected. In other examples, breaching the attempt threshold 144 may indicate that the COT lighting configuration plan 120 is accurate enough, and the new commissioning exemplar 106 based upon the COT reasoning steps 118 used to arrive at the current COT lighting configuration plan 120 may be generated. The attempt threshold 144 may be generated by a helper artificial intelligence (Al) model 105. The helper Al model 105 may be trained to generate an attempt threshold 144 based on a variety of factors, including the attempt counts 142 of previously generated commissioning exemplars 106. Further, the attempt count 142 and / or the attempt threshold 144 may be weighted based on the computational resources 178 required to generate the iterations of the COT lighting configuration plan 120 corresponding to the currently selected example floor plan 112. The computational resources 178 required may also fluctuate from iteration to iteration, as the LLM 101 may used different aspects of its internal models to generate various COT lighting configuration plans 120. Accordingly, a weighted attempt count 180 may be compared to a weighted attempt threshold 182, and the weighted attempt count 180 may be incorporated into a commissioning exemplar 106.

[0081] FIG. 7 is a variation of the functional block diagram of FIG. 2, wherein a selfreflection statement 146 is incorporated into the commissioning exemplar 106. In the aforementioned examples, the new commissioning exemplar 106 is created when the COT lighting configuration plan 120 is determined to be accurate. However, in some examples, useful commissioning exemplars 106 can be created from inaccurate COT lighting configuration plans 120. In this example, in response to please-revise prompting 160 when the COT lighting configuration plan 120 is determined to be inaccurate, the LLM 101 also generates a self-reflection statement 146. The self-reflection statement 146 describes why the LLM 101 believes the COT lighting configuration plan 120 is inaccurate. A commissioning exemplar 108 may then be generated based on the COT reasoning steps 118 of the inaccurate COT lighting configuration plan 120 and including the self-reflection statement 146. Thus, during the inference stage, this commissioning exemplar 106 may be used to show the LLM 101 how to avoid the same inaccuracies and faulty reasoning of the commissioning exemplar 106.

[0082] FIG. 8 is a variation of the functional block diagram of FIG. 2, wherein an LLM temperature 148 is used to generate commissioning exemplars 106. The LLM temperature 148 is a parameter used to control the randomness or volatility of the LLM 101 generating the COT lighting configuration plans 120 during the construction stage. While a higher LLM temperature 148 may result in more diverse commissioning exemplars 106, the COT lighting configuration plans 120 are more likely to be inaccurate, thus requiring more attempts (and therefore more computational resources) by the LLM 101 to generate accurate COT lighting configuration plans 120. Similarly, a lower LLM temperature 148 may result in less diverse commissioning exemplars 106, but may require less computational resources when generating the COT lighting configuration plans 120. In some examples, the LLM temperature 148 may be purposefully adjusted during the construction stage. For example, LLM temperature 148 may be initially set to a lower temperature during the first few attempts of the construction stage. However, if the LLM 101 fails to generate an accurate COT lighting configuration plan 120, the LLM temperature 148 may be raised to increase the creativity of the LLM 101.

[0083] FIG. 9 is a variation of the functional block diagram of FIG. 2, wherein synthetic lighting configuration plans 152 are used to generate the commissioning exemplars 106. While the stored ground truth data 110 pairs example floor plans 112 with ground truth lighting configuration plans 114, some example floor plans 112 may accommodate more than one lighting configuration plan. Thus, in determining the accuracy of a COT lighting configuration plan 120, the COT lighting configuration plan 120 should be compared against every lighting configuration plan compatible to the example floor plan. In most cases, the ground truth lighting configuration plans 114 are manually created, real world plans. Thus, as shown in FIG. 9, a robust LLM 107 may be used to generate synthetic lighting configuration plans 152 based on the example floor plan 112 and the ground truth lighting configuration plan 114. Accordingly, in the example of FIG. 9, rather than just comparing the COT lighting configuration plan 120 to the ground truth lighting configuration plan 114, the COT lighting configuration plan 120 is compared to each of the synthetic lighting configuration plans 152 as well. The COT reasoning steps 118 associated with the COT lighting configuration plan 120 are used to create the commissioning exemplar 106 if the COT lighting configuration 120 is then determined to be accurate relative to any one of the synthetic lighting configuration plans 152.

[0084] FIG. 10 is a variation of the functional block diagram of FIG. 2, wherein the ground truth data 110 is organized into one or more clusters 154. These clusters 154 may organize the ground truth data 110 into a variety of categories, such as type of premises (office space, big box store, retail store, apartment building, etc.). Accordingly, when constructing the commissioning exemplars, ground truth data 110 from a certain cluster 154 may be used to ensure that the commissioning exemplar 106 is relevant to the category associated with that cluster 154. As further shown in FIG. 11, the example floor plan 112 may be chosen by a helper Al model 105, in accordance with an example.

[0085] FIG. 12 illustrates a schematic diagram of a non-limiting example of a controller 100 used to generate a lighting configuration plan 116. Broadly, the controller 100 includes a memory 125 and a processor 175. In some examples, the controller 100 may also include a receiver, transmitter, and / or a transceiver for receiving and / or transmitting data. Further, while the controller 100 is depicted in FIG. 12 as a single component, in some examples, the various functions of the controller 100 may be performed by more than one component. The memory 125 is configured to store a wide array of data received by, generated by, or otherwise handled by the controller 100, including the floor plans 102, the boundary conditions 104, the commissioning exemplars 106, the demonstration pool 108 (comprising the commissioning exemplars 106), the ground truth data 110 (which may be organized into clusters 154, including example floor plans 112 and ground truth lighting configuration plans 116 of ground truth luminaire parameters 164 and gold labels 124), the lighting configuration plans 116 (including position data 156, configuration data 158, type data 186, and installation data 190), the COT reasoning steps 118, the COT lighting configuration plans 120 (including COT luminaire parameters 130 of position data 132, configuration data 134, type data 184, and installation data 188), the final COT reasoning chain 122, the difficulty level 126, the step count 128, the deviation metric 136, the must- have parameters 138, the deviation threshold 140, the attempt count 142, the attempt threshold 144, the self-reflection statement 146, the LLM temperature 148, the prompting feedback 150 (including the please-revise prompt 160 and the summary -prompt 162), the synthetic lighting configuration plans 152, the similarity metric 168, the COT accuracy 170, the quantity criteria 172, the COT accuracy threshold 174, the word count 176, the computational resources 178 required to generate an COT lighting configuration plan 120, the weighted attempt count 180, and the weighted attempt threshold 182.

[0086] FIG. 13 illustrates the processor 175 of the controller 100 in more detail. As shown in FIG. 13, the processor 175 is configured to execute the LLM 101, the comparator 103, the helper Al model 105, and the robust LLM 107.

[0087] FIG. 14 illustrates a method 900 for commissioning a lighting system comprising a plurality of luminaires corresponding to a premises. With respect to FIGS. 1-15, the method 900 includes, in step 902, retrieving a floor plan 102 and one or more boundary conditions 104. The floor plan 102 and the one or more boundary conditions 104 correspond to the premises. The method 900 further includes, in step 904, selecting one or more commissioning exemplars 106 from a demonstration pool 108 comprising a plurality of commissioning exemplars 106. The one or more commissioning exemplars 106 are selected based on a similarity 168 between the one or more commissioning exemplars 106 and the floor plan 102 and / or the one or more boundary conditions 104. Each of the plurality of commissioning exemplars 106 comprises a portion of COT reasoning steps 118 leading from an example floor plan 112 to a ground truth lighting configuration plan 114.

[0088] The method 900 further includes, in step 906, generating, via an LLM 101, a lighting configuration plan 116 based on the floor plan 102, the one or more boundary conditions 104, and the one or more commissioning exemplars 106.

[0089] FIG. 15 illustrates further steps of the method 900 for commissioning a lighting system comprising a plurality of luminaires corresponding to a premises. In particular, the steps of FIG. 15 illustrate a sub-routine for the construction stage of generating the commissioning exemplars 106. The method 900 further includes, in step 908, querying the LLM 101 using prompting with the example floor plan 112 to generate the COT reasoning steps 118 and a COT lighting configuration plan 120.

[0090] The method 900 further includes, in step 910, determining a COT accuracy 170 of the COT lighting configuration plan 120 by, at least in part, comparing the COT lighting configuration plan 120 to the ground truth lighting configuration plan 114 corresponding to the example floor plan 112.

[0091] The method 900 further includes, in step 912, if the COT accuracy 170 is above an accuracy threshold 174, generating a commissioning exemplar 106 comprising at least a portion of the COT reasoning steps 118.

[0092] The method 900 further includes, in step 914, if the COT accuracy 170 is below the accuracy threshold 174, repeatedly querying the LLM 101 to update the COT reasoning steps 118 and the COT lighting configuration plan 120 until the COT accuracy 170 is above the accuracy threshold 174.

[0093] All definitions, as defined and used herein, should be understood to control over dictionary definitions, definitions in documents incorporated by reference, and / or ordinary meanings of the defined terms.

[0094] The indefinite articles “a” and “an,” as used herein in the specification and in the claims, unless clearly indicated to the contrary, should be understood to mean “at least one.” The phrase “and / or,” as used herein in the specification and in the claims, should be understood to mean “either or both” of the elements so conjoined, i.e., elements that are conjunctively present in some cases and disjunctively present in other cases. Multiple elements listed with “and / or” should be construed in the same fashion, i.e., “one or more” of the elements so conjoined. Other elements may optionally be present other than the elements specifically identified by the “and / or” clause, whether related or unrelated to those elements specifically identified.

[0095] As used herein in the specification and in the claims, “or” should be understood to have the same meaning as “and / or” as defined above. For example, when separating items in a list, “or” or “and / or” shall be interpreted as being inclusive, i.e., the inclusion of at least one, but also including more than one, of a number or list of elements, and, optionally, additional unlisted items. Only terms clearly indicated to the contrary, such as “only one of’ or “exactly one of,” or, when used in the claims, “consisting of,” will refer to the inclusion of exactly one element of a number or list of elements. In general, the term “or” as used herein shall only be interpreted as indicating exclusive alternatives (i.e. “one or the other but not both”) when preceded by terms of exclusivity, such as “either,” “one of,” “only one of,” or “exactly one of.”

[0096] As used herein in the specification and in the claims, the phrase “at least one,” in reference to a list of one or more elements, should be understood to mean at least one element selected from any one or more of the elements in the list of elements, but not necessarily including at least one of each and every element specifically listed within the list of elements and not excluding any combinations of elements in the list of elements. This definition also allows that elements may optionally be present other than the elements specifically identified within the list of elements to which the phrase “at least one” refers, whether related or unrelated to those elements specifically identified.

[0097] It should also be understood that, unless clearly indicated to the contrary, in any methods claimed herein that include more than one step or act, the order of the steps or acts of the method is not necessarily limited to the order in which the steps or acts of the method are recited.

[0098] In the claims, as well as in the specification above, all transitional phrases such as “comprising,” “including,” “carrying,” “having,” “containing,” “involving,” “holding,” “composed of,” and the like are to be understood to be open-ended, i.e., to mean including but not limited to. Only the transitional phrases “consisting of’ and “consisting essentially of’ shall be closed or semi-closed transitional phrases, respectively. The above-described examples of the described subject matter can be implemented in any of numerous ways. For example, some aspects may be implemented using hardware, software, or a combination thereof. When any aspect is implemented at least in part in software, the software code can be executed on any suitable processor or collection of processors, whether provided in a single device or computer or distributed among multiple devices / computers.

[0099] The present disclosure may be implemented as a system, a method, and / or a computer program product at any possible technical detail level of integration. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present disclosure.

[0100] The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non- exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

[0101] Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.

[0102] Computer readable program instructions for carrying out operations of the present disclosure may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the “C” programming language or similar programming languages. The computer readable program instructions may execute entirely on the user’s computer, partly on the user's computer, as a stand-alone software package, partly on the user’s computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some examples, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) may execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present disclosure.

[0103] Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to examples of the disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer readable program instructions.

[0104] The computer readable program instructions may be provided to a processor of a, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram or blocks.

[0105] The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0106] The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various examples of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.

[0107] Other implementations are within the scope of the following claims and other claims to which the applicant may be entitled.

[0108] While various examples have been described and illustrated herein, those of ordinary skill in the art will readily envision a variety of other means and / or structures for performing the function and / or obtaining the results and / or one or more of the advantages described herein, and each of such variations and / or modifications is deemed to be within the scope of the examples described herein. More generally, those skilled in the art will readily appreciate that all parameters, dimensions, materials, and configurations described herein are meant to be exemplary and that the actual parameters, dimensions, materials, and / or configurations will depend upon the specific application or applications for which the teachings is / are used. Those skilled in the art will recognize, or be able to ascertain using no more than routine experimentation, many equivalents to the specific examples described herein. It is, therefore, to be understood that the foregoing examples are presented by way of example only and that, within the scope of the appended claims and equivalents thereto, examples may be practiced otherwise than as specifically described and claimed. Examples of the present disclosure are directed to each individual feature, system, article, material, kit, and / or method described herein. In addition, any combination of two or more such features, systems, articles, materials, kits, and / or methods, if such features, systems, articles, materials, kits, and / or methods are not mutually inconsistent, is included within the scope of the present disclosure.

Claims

CLAIMS:

1. A method (900) for commissioning a lighting system comprising a plurality of luminaires corresponding to a premises, the method comprising: retrieving (902) a floor plan and one or more boundary conditions, wherein the floor plan and the one or more boundary conditions correspond to the premises; selecting (904) one or more commissioning exemplars from a demonstration pool comprising a plurality of commissioning exemplars, wherein the one or more commissioning exemplars are selected based on a similarity metric regarding the one or more commissioning exemplars and the floor plan and / or the one or more boundary conditions, wherein each of the plurality of commissioning exemplars comprises a portion of chain-of- thought, COT, reasoning steps leading from an example floor plan to a ground truth lighting configuration plan; and generating (906), via a large language model, LLM, a lighting configuration plan based on the floor plan, the one or more boundary conditions, and the one or more commissioning exemplars; wherein each of plurality of commissioning exemplars of the demonstrations pool are generated by: querying (908) the LLM using prompting with the example floor plan and / or the one or more boundary conditions to generate the COT reasoning steps and a COT lighting configuration plan; determining (910) a COT accuracy of the COT lighting configuration plan by, at least in part, comparing the COT lighting configuration plan to the ground truth lighting configuration plan corresponding to the example floor plan and / or the one or more boundary conditions; if the COT accuracy is above an accuracy threshold, generating (912) a commissioning exemplar comprising at least a portion of the COT reasoning steps; and if the COT accuracy is below the accuracy threshold, repeatedly querying (914) the LLM to update the COT reasoning steps and the COT lighting configuration plan until the COT accuracy is above the accuracy threshold.

2. The method (900) of claim 1, wherein, following the determination that the COT accuracy is above the accuracy threshold, the LLM is queried to produce a final COT reasoning chain based on the COT reasoning steps, and wherein the commissioning exemplar is further generated based at least in part on the final COT reasoning chain and the example floor plan.

3. The method (900) of claim 1, wherein the COT lighting configuration plan comprises one or more COT luminaire parameters generated by the LLM via COT reasoning, and wherein the one or more COT luminaire parameters comprise position data, configuration data, type data, and / or installation data.

4. The method (900) of claim 3, wherein comparing the COT lighting configuration plan to the ground truth lighting configuration plan comprises: determining a deviation metric between one or more must have parameters of the one or more COT luminaire parameters and one or more corresponding ground truth luminaire parameters of the ground truth lighting configuration plan; and comparing the deviation metric to a deviation threshold to determine the COT accuracy of the COT lighting configuration plan.

5. The method (900) of claim 1, wherein an attempt count is iterated for each repetition of querying the LLM to update the COT reasoning steps and the COT lighting configuration plan, wherein the attempt count is assigned to the commissioning exemplar generated based on the COT lighting configuration plan.

6. The method (900) of claim 5, wherein repeatedly querying of the LLM to update the COT reasoning steps and the COT lighting configuration plan ceases if the attempt count exceeds an attempt threshold.

7. The method (900) of claim 6, wherein a commissioning exemplar comprising at least a portion of the COT reasoning steps is generated when the attempt count exceeds the attempt threshold.

8. The method (900) of claim 6, wherein the attempt threshold is generated by a helper artificial intelligence (Al) model trained on one or more of the plurality of commissioning exemplars of the demonstration pool.

9. The method (900) of claim 6, wherein the attempt count and / or the attempt threshold is weighted according to computational resources required by the LLM for each repeating querying of the LLM.

10. The method (900) of claim 1, wherein the LLM is queried to generate a selfreflection statement if the COT accuracy is below the accuracy threshold, wherein the selfreflection statement describes faulty reasoning associated with the COT lighting configuration plan, and wherein the commissioning exemplar is based at least in part on the COT reasoning steps and the self-reflection statement.

11. The method (900) of claim 1, wherein the example floor plan is selected by a helper Al model.

12. The method (900) of claim 1, wherein the one or more commissioning exemplars are chosen from the demonstration pool based at least in part on a difficulty level assigned to each of the plurality of commissioning exemplars.

13. A system for commissioning a lighting system comprising a plurality of luminaires corresponding to a premises, the system comprising a controller (100) configured to: retrieve a floor plan (102) and one or more boundary conditions (104), wherein the floor plan (102) and the one or more boundary conditions (104) correspond to the premises; select one or more commissioning exemplars (106) from a demonstration pool (108) comprising a plurality of commissioning exemplars (106), wherein the one or more commissioning exemplars (106) are selected based on a similarity metric (168) regarding the one or more commissioning exemplars (106) and the floor plan (102) and / or the one or more boundary conditions (104), wherein each of the plurality of commissioning exemplars (106) comprises a portion of chain-of-thought, COT, reasoning steps (118) leading from an example floor plan (112) to a ground truth lighting configuration plan (114); andgenerate, via a large language model, LLM, (101), a lighting configuration plan (116) based on the floor plan (102), the one or more boundary conditions (104), and the one or more commissioning exemplars (106). wherein each of plurality of commissioning exemplars of the demonstrations pool are generated by: querying the LLM (101) using prompting with the example floor plan (112) to generate the COT reasoning steps (118) and a COT lighting configuration plan (120); determining a COT accuracy (170) of the COT lighting configuration plan (120) by, at least in part, comparing the COT lighting configuration plan (120) to the ground truth lighting configuration plan (114) corresponding to the example floor plan (112); if the COT accuracy (170) is above an accuracy threshold (174), generating a commissioning exemplar (106) comprising at least a portion of the COT reasoning steps (118); and if the COT accuracy (170) is below the accuracy threshold (174), repeatedly querying the LLM (101) to update the COT reasoning steps (118) and the COT lighting configuration plan (120) until the COT accuracy (170) is above the accuracy threshold (174).

Images