A geospatial data system

Abstract

A geospatial data system comprises (i) an Al-based data processing sub-system that includes geospatial data and that has been trained to recognise and classify features in an environment; and (ii) a number of devices, such as cars, or augmented reality glasses, each including a computer vision sub-system configured to provide spatial awareness to enable the device to navigate through the environment or to provide augmented reality for a user in that environment, and in which the geospatial data system re-purposes the computer vision sub- systems to create crowdsourced survey data of the environment and to provide that crowdsourced survey data back to the Al-based data processing sub-system. The Al-based data processing sub-system is configured to use the crowdsourced survey data as training data.

Description

[0001] A GEOSPATIAL DATA SYSTEM

[0002] FIELD OF THE INVENTION

[0003] This invention relates to a geospatial data system; in particular it relates to a geospatial data system that comprises an Al-based data processing sub-system that has been trained to recognise and classify features in an environment.

[0004] BACKGROUND

[0005] Traditional methods of image data capture for geospatial mapping applications, like street-level views, rely on manually driven specialist vehicles equipped with 3D cameras and Lidar sensors driving tens of thousands of miles to comprehensively create accurate, geo-located images. This results in limitations in scalability, efficiency, and data quality. These existing methods are used to provide not just the street-level images that can be seen by consumers using the map, but also to generate the training data used to train Al models - e g. to train the models to recognise and accurately classify features like road signs, kerbs, traffic lights and to develop a spatial model of roads, pavements, kerbs, traffic lights, signs etc.

[0006] Using specialist, human-driven vehicles means that significant resources are needed to manage the large vehicle fleets that are needed, and to capture the data they generate, all working across different timeframes and geographies. Additionally, human drivers introduce inconsistencies and inefficiencies that impact the accuracy and effectiveness of training data capture.

[0007] SUMMARY OF THE INVENTION

[0008] A first aspect of the invention is a geospatial data system comprising:

[0009] (i) an Al-based data processing sub-system that includes geospatial data and that has been trained to recognise and classify features in an environment;

[0010] (ii) a number of devices, each including a computer vision sub-system configured to provide spatial awareness to enable the device to navigate through the environment or to provide augmented reality for a user in that environment, and in which the geospatial data system re-purposes the computer vision sub-systems to create crowdsourced survey data of the environment and to provide that crowdsourced survey data back to the Al-based data processing sub-system; and in which the Al-based data processing sub-system is configured to use the crowdsourced survey data as training data.

[0011] A second aspect is a wearable device, such as a pair of smart glasses, the device including a computer vision sub-system configured to provide spatial awareness to enable the device to map its environment and / or to provide augmented reality for a user in that environment; and in which device is configured so that the computer vision sub-system is re-purposed to create crowdsourced survey data of the environment and to provide that crowdsourced survey data back to the Al-based data processing sub-system for use as training data.

[0012] A third aspect is a method of generating geospatial data, the method including:

[0013] (i) operating an Al-based data processing sub-system that is trained to recognise and classify features in an environment;

[0014] (ii) deploying a number of devices, each including a computer vision sub-system configured to provide spatial awareness to enable the device to navigate through the environment or to provide augmented reality for a user in that environment, and in which the geospatial data system re-purposes the computer vision sub-systems to create crowdsourced survey data of the environment and to provide that crowdsourced survey data back to the AI- based data processing sub-system; and in which the Al-based data processing sub-system uses the crowdsourced survey data as training data.

[0015] Any one or more of the following optional features can be applied to any of the above aspects and can be combined with any one or more of these optional features.

[0016] The geospatial data

[0017] • the geospatial data includes a master version of a geospatial map.

[0018] • the geospatial data defines a master, navigable map of at least parts of the real, physical world. • the geospatial data defines a master, navigable map of at least parts of a virtual or simulated world, such as the metaverse or omniverse.

[0019] The Al-based data processing sub-system

[0020] • the Al-based data processing sub-system uses the crowdsourced survey data to augment or update a master, navigable map of at least parts of the real world.

[0021] • the Al-based data processing sub-system uses the crowdsourced survey data to augment or update a master, navigable map of at least parts of a virtual world or simulated, such as the metaverse or omniverse.

[0022] • the Al-based data processing sub-system is trained using the training data to recognise and classify features in the environment.

[0023] • the Al-based data processing sub-system is trained using the training data to develop a spatial model of the real-world.

[0024] The training data

[0025] • the training data includes data that is provided in real-time, continuously or at or in excess of a pre-set frequency.

[0026] • the training data enables the Al-based data processing sub-system to improve its ability to recognise and classify features in the environment.

[0027] • the training data enables the Al-based data processing sub-system to improve the quality of the geospatial data.

[0028] • the training data is used to train Al models.

[0029] • the training data is used to update, enhance or improve the geospatial data.

[0030] • the training data enables the Al-based data processing sub-system to improve its route planning capability.

[0031] The feedback loop

[0032] • the Al-based data processing sub-system compares the stored geospatial data with the crowdsourced survey data and corrects or updates the stored geospatial data dependent on the crowdsourced survey data, iteratively repeating this process as more crowdsourced survey data is received and processed in a regular or continuous feedback loop. The environment features

[0033] • the features in the environment include any of the following: road markings, traffic signs, traffic signals, roadworks, diversions, traffic data, traffic jams.

[0034] • the features in the environment include any of the following: weather data; event data, such as sporting events, or anything that affects route planning.

[0035] The devices

[0036] • the devices are configured to move through the environment as part of their normal operations and are not dedicated surveying or mapping devices

[0037] • the devices include vehicles, such as autonomous vehicles or drones

[0038] • the devices include wearable devices, such as smart glasses.

[0039] • the devices include robo-taxis or robo-delivery cars, vans, bikes, robots or drones, configured to traverse the environment as part of their normal robo-taxi or robo- delivery operations.

[0040] Dynamic re-routing

[0041] • the Al-based data processing sub-system uses the crowdsourced survey data to update in real-time, continuously or at or in excess of a pre-set frequency, a master map of at least parts of the real world with one or more of the following that affects route planning: roadworks, diversions, traffic data, traffic jams, weather, events and the system includes a dynamic route planning sub-system configured to generate revised routes for vehicles that take into account the updated master map.

[0042] Route planning to ensure comprehensive data capture

[0043] • the system includes a route planning sub-system that is configured with a route planning algorithm that automatically takes into account one or more of the following when determining whether to include a route section into a route provided to a device: (i) how recently that that route section has been surveyed or for how long that route section has not been surveyed; and (ii) how many devices have surveyed that route section within a pre-set earlier period.

[0044] • the route planning algorithm is configured to provide route planning for devices traversing a region in the environment in a manner that provides comprehensive survey data across that region. • the route planning algorithm is configured to balance a requirement to provide the fastest or shortest route with a requirement to generate crowdsourced survey data across all navigable sections across the entire region.

[0045] • the route planning algorithm is configured to balance a requirement to provide a route through a region that optimises a parameter, such as minimising time, distance or energy efficiency, picking up / delivery of packages or people to a schedule, with a requirement to generate crowdsourced survey data across all navigable sections across the entire region.

[0046] • the route planning algorithm is configured to provide route planning for devices traversing a region in the environment in a manner that takes into account one or more of the following, as determined by the Al-based data processing sub-system using machine learning, when determining whether to include a route section into a route provided to a device: (i) whether that route section includes a new or altered feature in the environment, such as a map update or new infrastructure, like a new EV charging station; (ii) deviations in actual transit times over that route section compared with predicted times; (iii) a safety parameter associated with that route section, generated by the Al-based data processing sub-system; (iv) a fuel or power efficiency parameter associated with that route section.

[0047] Generative Al

[0048] • the system includes a generative Al system configured to generate alternative, synthetic versions of features in the geospatial data.

[0049] BRIEF DESCRIPTION OF THE FIGURES

[0050] An implementation of the invention is shown schematically in Figures 1 and 2.

[0051] Figure 3 shows schematically a further feature, using generative Al, to generate alternative, synthetic versions of features in the geospatial data. This feature is the subject of Appendix 1.

[0052] DETAILED DESCRIPTION

[0053] In this section, we will describe an implementation of the invention. This implementation provides comprehensive, persistent and up to date training data for the geospatial Al system, since it is not reliant on (although can include) conventional, dedicated human-driven vehicles whose sole purpose is to generate geospatial data; these conventional vehicles might pass and image a specific location just once in a year, or less. Instead, by re-purposing the imaging and sensor systems of vehicles that are traversing a region anyway (e.g. they are delivery vehicles, or robo-taxis, or ordinary vehicles being driven for ordinary purposes, like commuting, shopping trips etc) and specifically using this to provide training data for the geospatial Al system, Al training data can be provided at a much greater scale, which is updated regularly or frequently, or in real-time or on demand: this training data greatly improves the accuracy and stability of the geospatial Al system, and its ability to handle rare or unknown edge cases. So a specific location, that might be passed just once a year by a dedicated vehicle imaging and scanning that location for a mapping company, can now be imaged and scanned hundreds of times a day, providing a far richer source of spatial training data for the geospatial Al system. The geospatial Al system can much more rapidly and accurately develop its spatial model of the world, including the more unusual edge cases that normal training data would typically lack.

[0054] Where the Al system determines that there is uncertainty or ambiguity in its training data for a specific item (e g. a thing or feature in the real world, or a spatial relationship between things or features in the real world) at a specific location, and hence how it classifies or otherwise uses that item in the trained model, then with this system, additional training data can be sourced far more rapidly from the vehicles that are passing that item, either as part of their normal operations, or because the Al system has suggested to those vehicles a route that goes past that item, as part of a routing algorithm used by those vehicles.

[0055] An implementation of the present invention comprises a real-time Al management system with a feedback loop that optimises the operation of a fleet of robo taxi vehicles for capturing image data at a city or country scale. The system continuously trains the Al model using a feedback loop from frequent or real-time data received, ensuring superior performance in route planning and data capture. By leveraging the inherent Al capabilities of the robo taxi vehicles, the system eliminates the need for human intervention in driving, route planning, and capture fleet management. The system's approach to dynamic route planning is based on an Area Of Interest rather than the quickest route, which differentiates it from standard techniques.

[0056] An implementation of the invention relates to a superior method for utilising artificial intelligence (Al) with a feedback loop to manage a fleet of robo taxi vehicles, such as electric vehicles (EVs) or autonomous vehicles (AVs), for efficient image data capture at a city or country scale. The system enables frequent or real-time monitoring and control of the fleet, continuously training the Al model from the received data for improved performance. By removing the need for human personnel to drive vehicles, define route planning, or manage the capture fleet data across different timeframes and locations. The efficiency of capture is enhanced by leveraging the Al capabilities already present in the Robo Taxi vehicles to navigate the road networks safely. The vehicles continuously refresh their dynamic route plans based on a feedback loop from other vehicles capturing in the same city, as well as other potential data feeds from traffic cameras, traffic data, weather data and analysis. Additionally, the system introduces a novel approach to route planning, prioritising the effective collection of a defined Area Of Interest rather than the quickest route from A to B, setting it apart from standard route planning software and techniques.

[0057] The real-time Al management system comprises a central computer implemented control unit, a communication network, and a fleet of robo taxi vehicles equipped with image capture devices and various sensors. The control unit utilises Al algorithms and machine learning techniques with a feedback loop to process data and make intelligent decisions regarding fleet operations.

[0058] The system's operation can be summarised as follows: 1. Fleet Monitoring and Optimisation:

[0059] - Each robo taxi vehicle continuously transmits telemetry data, including location, battery / fuel levels, and vehicle health, to the control unit via the communication network.

[0060] - The control unit analyses the received data, identifies optimisation opportunities, and updates the Al model with the feedback loop.

[0061] - The Al model, continuously trained with frequent or real-time data, improves its decisionmaking capabilities and captures efficiency over time.

[0062] - By continuously learning from captured data, the Al model adapts to unknown edge cases, improving performance over time.

[0063] 2. Route Planning and Optimisation based on Area of Interest:

[0064] - The system introduces a novel approach to route planning, focusing on capturing all routes within a defined Area of Interest rather than simply finding the quickest route from A to B.

[0065] - This unique route planning methodology ensures effective collection of image data within the specified area, optimising data capture efficiency.

[0066] - The control unit uses the continuously updated Al model to calculate the most efficient routes for each robo taxi vehicle.

[0067] - The routes are dynamically adjusted based on frequent or real-time data, including road conditions, traffic congestion, accidents, weather conditions, and the need to avoid duplicate data capture.

[0068] - The feedback loop ensures that the Al model incorporates learnings from the fleet's collective data capture experience.

[0069] 3. Sensor Integration and Data Capture:

[0070] - The robo taxi vehicles are equipped with image capture devices (e.g. cameras, 360 degree cameras) and various sensors (such as air quality, GNSS, LiDAR, thermal, sound, electromagnetic radiation signals), which collect data during fleet operations.

[0071] - The captured data is processed frequently or in real-time or stored for later analysis, further enhancing the Al model's training and data quality.

[0072] 4. Data Synchronisation and Analysis:

[0073] - The control unit synchronises and analyses the captured data to identify duplicate miles, overlapped routes, or areas with incomplete coverage. - The Al model's continuous training enables it to learn from previous data capture instances, improving the accuracy and efficiency of future captures.

[0074] 5. Leveraging Robo Taxi Al Capabilities:

[0075] - The system takes advantage of the Al capabilities already present in the robo taxi vehicles, enabling safe navigation through road networks.

[0076] - The efficiency of data capture benefits from the integrated Al, as the vehicles autonomously navigate the routes specified by the dynamic route planning.

[0077] 6. Elimination of Human Personnel:

[0078] - The system removes the need for human personnel to drive the vehicles, define route planning, or manage the capture fleet data.

[0079] - This eliminates inconsistencies and inefficiencies introduced by human drivers, enhancing data capture accuracy and effectiveness.

[0080] Figures 1 and 2 schematically shows this geospatial data system (Figure 1 forms the left side of the system chart and Figure 2 the right side; note that the position of the arrow heads do not imply that data flows in the arrow direction; in most cases, data is bidirectional). Looking at Figure 1, at the apex is the Central Al 3D Model that has been trained to recognise and classify features in an environment (this includes being trained on and to further develop a spatial model of the environment). There are a number of different devices that each include a computer vision sub-system configured to provide spatial awareness to enable the device to navigate through the environment or to provide augmented reality for a user in that environment: Figure 1 shows the Central Al 3D Model connected to a Communication Network, which in turn is connected to the devices, namely a fleet of robo taxi vehicles as one set, and smart glass and wearables as another, that provide the crowd-sourced training data to the Central Al 3D Model. Figure 1 also shows the geospatial data system including or exchanging data with a Data Synchronisation and Analysis sub-system; one function is to Identify Duplicates (e.g. image duplicates) and Incomplete Coverage; data derived from identifying areas of incomplete coverage is used by a sub-system to Improve Future Captures.

[0081] Figure 2 is the right side of the system diagram; at the apex is the Al Algorithms and Machine Learning Techniques subsystem, which is connected to the Central Al 3D Model, which is being continuously (e g. regularly) trained using the crowdsourced training data from the vehicles and wearables. The Al Algorithms and Machine Learning Techniques sub-system connects to the Fleet Monitoring and Optimisation sub-system, which in turn connects to the Analyse Data process, the Update Al Model process, the Improve Decision-Making process and the Adapt to Unknown Edge Cases process. The overall effect is for the geospatial Al 3D model to be updated and also trained using vehicles under the control of the Fleet Monitoring and Optimisation sub-system; this leads to improved decision making and rapid learning and hence adaptation to rare or unknown edge cases. Figure 2 also shows the Dynamic Route Planning sub-system: this includes a Calculate Efficient Routes process, a Dynamically Adjust Routes process and an Incorporate Learnings process. The system enables comprehensive data capture by vehicles through a Prioritise Area of Interest process (e.g. to route vehicles to sections of a route that are rarely imaged, or for which there is some ambiguity in the data, or for which that section provides exemplary training data for an unusual, rare edge case). The system includes a Feedback Loop sub-system, connected to various data feeds (Traffic Cameras, Weather Data, Other Input Sources, Traffic Data).

[0082] Advantages

[0083] The present system offers several significant advantages over conventional approaches:

[0084] - The feedback loop with continuous training enhances the Al model's performance, allowing it to outperform manually pre-trained models and handle unknown edge cases effectively.

[0085] - By leveraging Al capabilities, the system removes the need for human personnel in driving vehicles, defining route planning, or managing capture fleet data across various timeframes and geographies.

[0086] - The system benefits from the inherent efficiency of robo taxi vehicles' Al capabilities, enabling safer navigation of road networks while optimising capture efficiency.

[0087] - The dynamic route planning based on the feedback loop from other vehicles capturing in the same city ensures the fleet captures data most efficiently, avoiding redundant coverage.

[0088] The real-time Al management system with a feedback loop for fleet capture of image data described herein represents a significant advancement in the field of data capture. By continuously training the Al model with frequent or real-time data and leveraging the Al capabilities of robo taxi vehicles, the system ensures superior efficiency and accuracy in capturing image data at a city scale. The scope of this system encompasses the broad application of this technology, independent of specific camera or sensor technologies and not limited to road-going vehicles and not limited to robo taxi vehicles and can incorporate human driven vehicles also.

[0089] Appendix 1

[0090] Generative AI-Based Design Modification in High-Resolution 3D Geospatial Data

[0091] The feature described in this Appendix 1 relates to a system and method for utilising generative artificial intelligence (Al) techniques on a large-scale, high-resolution 3D geospatial data model. It can be used in the geospatial data system described earlier, but can also be used entirely independently of that system. Figure 3 is a schematic view of this computer implemented system.

[0092] The system allows users to select specific areas or buildings within the dataset and generate simulations of alternate designs, including modifications to road access. The generative Al enables interactive design exploration, measurement, and modification, ultimately facilitating the creation of exportable architectural drawings and measurements for quotation purposes. Additionally, the system offers options for generating marketing imagery, video fly-through content, and animations, further expanding its applications in promotional and visualisation endeavours. This feature opens new possibilities for efficient design iterations and accurate representation in the construction industry and other applications relying on 3D geospatial data. The system can pull upon historical data to accurately represent a location, place, building, or structure. Leveraging generative Al, the system can interpolate missing information, enabling a time machine-like functionality that allows users to explore the past and future appearances of a given location, including topological changes. This feature expands the capabilities of 3D geospatial data analysis and visualisation, providing valuable insights and enhanced historical understanding.

[0093] This feature pertains to the field of generative artificial intelligence applied to large-scale 3D geospatial data, specifically targeting the modification of buildings, road access, and related infrastructure. Furthermore, the feature encompasses the generation of marketing imagery, video fly-through content, and animations based on the generative Al techniques applied to the geospatial data. The system may incorporate historical or other data feeds and enable a time machine-like functionality to visualise the past and future appearances of a location, place, building, or structure. The availability of high-resolution 360-degree digital cameras, accurate GNSS positioning and tracking systems, and lidar point cloud data has revolutionised the creation of detailed 3D models of outdoor environments. However, traditional design modification processes for buildings and infrastructure often involve time-consuming manual work and lack accurate visual representations. Additionally, generating captivating marketing content based on these 3D models requires significant effort and resources. There is a need for an improved system and method that leverages generative Al to enable efficient design modification, visualisation, and measurement within such high-resolution 3D geospatial data models.

[0094] Also, accurately representing the historical aspects of a location, including its appearance at different points in time and topological changes, has been challenging. There is a need for an improved system and method that leverages generative Al to interpolate missing historical data, offering a time machine-like functionality to explore the evolution of a location, place, building, or structure.

[0095] This feature introduces a novel system and method for utilising generative Al on a large-scale 3D geospatial data model. The system allows users to interactively select specific areas or buildings within the model and generate simulations of alternate designs, including changes to road access and other relevant infrastructure. The generative Al facilitates the exploration and modification of these designs, enabling manual adjustments or guided prompts to achieve desired outcomes. The resulting modified model can be exported with accurate architecture drawings and measurements, streamlining the quotation process with contractors and ensuring a comprehensive understanding of the proposed design changes. Additionally, the system offers options for generating marketing imagery, video fly-through content, and animations based on the generative Al techniques applied to the geospatial data. This feature enables users to create compelling visual materials for promotional, presentation, or visualisation purposes, enhancing their ability to showcase design concepts and attract stakeholders.

[0096] The present feature also introduces a novel system and method that utilises generative Al on a large-scale 3D geospatial data model to provide time machine-like functionality. By incorporating historical data and leveraging generative Al techniques, the system accurately represents the past and future appearances of a given location, including topological changes. This expands the capabilities of 3D geospatial data analysis and visualisation, offering valuable insights and enhanced historical and future understanding. The system comprises a computer-implemented process that takes as input a high-resolution 3D geospatial data model incorporating panoramic images, accurate GNSS positioning and tracking data, and lidar point cloud information. Additionally, historical data related to the location, place, building, or structure could be collated and incorporated into the dataset. The generative Al algorithms embedded in the system utilise deep learning techniques to understand the existing model and simulate the appearance and feasibility of alternate design proposals.

[0097] Upon user input, specifying the target area or building for modification, the system generates a simulation of the proposed changes. This includes rendering a 3D model of the new building or modified structure, as well as visual representations of associated road access modifications. The generative Al leverages its learned understanding of the dataset to ensure accurate and contextually appropriate design suggestions.

[0098] The user can then interact with the generated simulation to measure and modify the design manually or through guided prompts provided by the system. Measurements, annotations, and modifications made within the system are updated in real-time, allowing for rapid design iterations and precise adjustments.

[0099] Once the design modification process is complete, the system can export the final modified model along with comprehensive architecture drawings and precise measurements. This facilitates the quotation process with contractors, as they can accurately evaluate the proposed changes and provide informed estimates.

[0100] The generative Al algorithms embedded in the system utilise deep learning techniques to understand the available historical data and the existing 3D geospatial model. By analysing patterns, structures, and contextual information, the generative Al can intelligently interpolate missing data points, allowing for an accurate representation of the location's appearance at various points in time.

[0101] Through user input, specifying a desired time frame, the system generates a simulation of the location's appearance during that period. This includes rendering the buildings, structures, and topological features as they would have existed at the chosen time, based on the generative ATs interpolation capabilities. Users can navigate and explore the 3D geospatial model, experiencing a time machine-like functionality that provides a unique historical perspective.

[0102] Figure 3 is a schematic representation of this system. It includes, at the apex, a 3D Geospatial Data Model; the right hand limb shows the various Data Sources for the Model (360 Imagery, Environmental Planning Data, Weather, GNSS, Legal, Historical Data Records, Image Data Classification, City Planning Data and Legislation, Lidar, and other data sources). The system includes a Generate Simulation of Alternative Designs sub-system, itself powered by a Generative Al sub-system. The Generative Al sub-system enables or is takes as input a number of processes: Interpolate Missing Information, Modify Road Access, Consider Local Planning Regulations, Time-Machine-Like Functionality, Explore Past and Future Appearances, Enhanced Historical Understanding, Topological Changes. The system enables Interactive Design Exploration, including quantified Measurement and Modification. The system outputs Metaver se / Omni verse content, Video Fly-Through Content, Animations, Marketing Imagery, Architectural Drawings and Measurements.

[0103] Applications:

[0104] The described system and method have broad applicability in the field of urban planning, architecture, civil engineering and marketing visualisation. Some potential use cases include:

[0105] 1. Urban redevelopment: Users can simulate and visualise the impact over time of replacing existing buildings with new designs, assess the feasibility of road network modifications, and generate exportable models, drawings and marketing materials for further analysis and stakeholder communication. This facilitates informed decision-making and long-term planning.

[0106] 2. Infrastructure improvement: The system enables the evaluation of potential modifications to road layouts, pedestrian walkways, and transportation systems within a 3D geospatial context, leading to optimised designs and enhanced accessibility. Users can generate video fly- through content and animations to showcase the proposed improvements and their benefits.

[0107] 3. Environmental impact assessment: Generative Al can be employed to simulate the effects of new construction projects on the surrounding environment, including sunlight exposure, wind flow, and noise propagation. The resulting visual materials can effectively communicate the envisioned design changes and their impact on the environment. The time machine functionality aids in evaluating the impact of environmental changes and climate-related factors on the evolution of a location. This enables better-informed decision-making regarding environmental preservation and mitigation efforts.

[0108] 4. Historical preservation and restoration: By manipulating the geospatial data model, users can virtually modify historical structures, assess the impact of restoration efforts, and generate accurate documentation for preservation purposes. Researchers can utilise the time machine functionality to explore and understand the historical evolution of a location, providing valuable insights for historical research and preservation efforts.

[0109] In conclusion, the present feature enables the utilisation of generative Al techniques on a large- scale 3D geospatial data model to facilitate interactive design modification, visualisation, and measurement. By offering accurate representations and exportable architectural drawings, the system enhances the efficiency and effectiveness of design iterations and the quotation process, revolutionising the use of 3D geospatial data in various industries. By incorporating historical data to provide a time machine-like functionality. By accurately representing the past and future appearances of a location, including topological changes, the system enhances the analysis, visualisation, and understanding of 3D geospatial data. This system opens up new possibilities for historical research, urban planning and environmental impact assessment, facilitating informed decision-making and providing a unique historical perspective.

[0110] We can generalise this feature to the following:

[0111] A geospatial data system including

[0112] (i) a datastore including geospatial data that includes features representing actual real- world features derived from surveying or mapping the real-world;

[0113] (ii) a generative Al sub-system configured to generate alternative, synthetic versions of features in the geospatial data.

[0114] A method of generating geospatial data, the method including the steps of:

[0115] (i) providing a datastore including geospatial data that includes features representing actual real-world features derived from surveying or mapping the real-world; (ii) operating a generative Al sub-system configured to generate alternative, synthetic versions of features in the geospatial data.

[0116] Any one or more of the following optional features can be applied to any of the above aspects and can be combined with any one or more of these optional features.

[0117] The geospatial data

[0118] • the geospatial data includes any of the following: panoramic images, accurate GNSS positioning and tracking data, lidar point cloud information, .

[0119] • the geospatial data includes a master version of a geospatial map

[0120] • the geospatial data includes a master, navigable map of at least parts of the real, physical world

[0121] • the geospatial data includes a master, navigable map of at least parts of a virtual or simulated world, such as the metaverse or omniverse

[0122] • the geospatial data includes features in the environment that include any of the following: road markings, traffic signs, traffic signals, roadworks, diversions, traffic data, traffic jams.

[0123] • the geospatial data includes features in the environment that include any of the following: weather data; event data, such as sporting events, or anything that affects route planning.

[0124] • the geospatial data includes natural features, such as rivers and flood plains, and projections for how those features will evolve in time, for example considering climate change.

[0125] •

[0126] Generative Al sub-system

[0127] • the generative Al sub-system includes generative Al algorithms that utilise deep learning techniques to understand or model the features in the geospatial data.

[0128] • the generative Al sub-system includes generative Al algorithms that utilise deep learning techniques to understand or model historical data, such as previously existing features, like existing buildings and their location, appearance and function. • the generative Al sub-system includes a LLM sub-system configured to process text based prompts and generate alternative, synthetic versions of features in the geospatial data that are Al-based continuations of the text based prompts

[0129] • the generative Al sub-system is configured to enable a user to select specific features, such as buildings, within the geospatial data to generate simulations of alternate designs for those features

[0130] • the generative Al sub-system is configured to enable prompt-based adjustments of the alternate designs

[0131] • the generative Al sub-system is configured to enable manual adjustments of the alternate designs generated by the generative Al sub-system

[0132] • the generative Al sub-system is configured to interpolate missing data points.

[0133] Generative Al sub-system outputs

[0134] • the generative Al sub-system is configured to output measurement of alternate designs of real-world features

[0135] • the generative Al sub-system is configured to output transport, e.g. road access, modifications for new or modified buildings.

[0136] • the generative Al sub-system is configured to output architectural drawings and measurements for quotation purposes

[0137] • the generative Al sub-system is configured to output marketing imagery, video fly- through content, and animations

[0138] Time machine

[0139] • the geospatial data includes data on historic, or past features that no longer exist and the generative Al algorithms utilise deep learning techniques to understand or model these historic or past features in the geospatial data

[0140] • the generative Al sub-system is configured to allow users to explore the past and future appearances of a given location, including historic, or past features and historic or past topological changes, as well as projected topological changes

[0141] • the generative Al sub-system is configured with a time machine-like functionality to show the evolution of a location, place, building, or structure through time. • the generative Al sub-system is configured to enable a user to specify a desired time frame for a location and the system then generates a simulation of the location's appearance, including changing appearance, during that time frame.

[0142] • the generative Al sub-system is configured to render buildings, structures, and topological features as they would have existed at the chosen time frame, based on the generative Al' sub-system's interpolation capabilities.

[0143] Applications

[0144] • the generative Al system is configured to enable the simulation and visualisation over time of replacing existing buildings with new designs, assess the feasibility of road network modifications, and generate exportable models, drawings and marketing materials, for urban development applications.

[0145] • the generative Al system is configured to enable the evaluation of potential modifications to road layouts, pedestrian walkways, and transportation systems, for Infrastructure improvement.

[0146] • the generative Al system is configured to enable the simulation of the effects of new construction projects on the surrounding environment, including sunlight exposure, wind flow, and noise propagation, for environmental impact assessment.

[0147] • the generative Al system is configured to enable the modification of historical structures, in light of proposed restoration efforts, for historical preservation and restoration.

Claims

CLAIMS1 . A geospatial data system comprising:(i) an Al-based data processing sub-system that includes geospatial data and that has been trained to recognise and classify features in an environment;(ii) a number of devices, each including a computer vision sub-system configured to provide spatial awareness to enable the device to navigate through the environment or to provide augmented reality for a user in that environment, and in which the geospatial data system re-purposes the computer vision sub-systems to create crowdsourced survey data of the environment and to provide that crowdsourced survey data back to the Al-based data processing sub-system; and in which the Al-based data processing sub-system is configured to use the crowdsourced survey data as training data.The geospatial data2. The geospatial data system of Claim 1 in which the geospatial data includes a master version of a geospatial map.

3. The geospatial data system of Claim 1 or 2 in which the geospatial data defines a master, navigable map of at least parts of the real, physical world.

4. The geospatial data system of any preceding Claim in which the geospatial data defines a master, navigable map of at least parts of a virtual or simulated world, such as the metaverse or omniverseThe Al-based data processing sub-system5. The geospatial data system of any preceding Claim in which the Al-based data processing sub-system uses the crowdsourced survey data to augment or update a master, navigable map of at least parts of the real world.

6. The geospatial data system of any preceding Claim in which the Al-based data processing sub-system uses the crowdsourced survey data to augment or update a master, navigable map of at least parts of a virtual world or simulated, such as the metaverse or omniverse.

7. The geospatial data system of any preceding Claim in which the Al-based data processing sub-system is trained using the training data to recognise and classify features in the environment.

8. The geospatial data system of any preceding Claim in which the Al-based data processing sub-system is trained using the training data to develop a spatial model of the real- world.The training data9. The geospatial data system of any preceding Claim in which the training data includes data that is provided in real-time, continuously or at or in excess of a pre-set frequency.

10. The geospatial data system of any preceding Claim in which the training data enables the Al-based data processing sub-system to improve its ability to recognise and classify features in the environment.11 . The geospatial data system of any preceding Claim in which the training data enables the Al-based data processing sub-system to improve the quality of the geospatial data.

12. The geospatial data system of any preceding Claim in which the training data is used to train Al models.

13. The geospatial data system of any preceding Claim in which the training data is used to update, enhance or improve the geospatial data.

14. The geospatial data system of any preceding Claim in which the training data enables the Al-based data processing sub-system to improve its route planning capability.The feedback loop15. The geospatial data system of any preceding Claim in which the Al-based data processing sub-system compares the stored geospatial data with the crowdsourced survey data and corrects or updates the stored geospatial data dependent on the crowdsourced survey data, iteratively repeating this process as more crowdsourced survey data is received and processed in a regular or continuous feedback loop.The environment features16. The geospatial data system of any preceding Claim in which the features in the environment include any of the following: road markings, traffic signs, traffic signals, roadworks, diversions, traffic data, traffic jams.

17. The geospatial data system of any preceding Claim in which the features in the environment include any of the following: weather data; event data, such as sporting events, or anything that affects route planning.The devices18. The geospatial data system of any preceding Claim in which the devices are configured to move through the environment as part of their normal operations and are not dedicated surveying or mapping devices.

19. The geospatial data system of any preceding Claim in which the devices include vehicles, such as autonomous vehicles or drones.

20. The geospatial data system of any preceding Claim in which the devices include wearable devices, such as smart glasses.

21. The geospatial data system of any preceding Claim in which the devices include robotaxis or robo-delivery cars, vans, bikes, robots or drones, configured to traverse the environment as part of their normal robo-taxi or robo-delivery operationsDynamic re-routing22. The geospatial data system of any preceding Claim in which the Al-based data processing sub-system uses the crowdsourced survey data to update in real-time, continuously or at or in excess of a pre-set frequency, a master map of at least parts of the real world with one or more of the following that affects route planning: roadworks, diversions, traffic data, traffic jams, weather, events and the system includes a dynamic route planning sub-system configured to generate revised routes for vehicles that take into account the updated master map.Route planning to ensure comprehensive data capture23. The geospatial data system of any preceding Claim in which the system includes a route planning sub-system that is configured with a route planning algorithm that automatically takes into account one or more of the following when determining whether to include a route section into a route provided to a device: (i) how recently that that route section has been surveyed or for how long that route section has not been surveyed; and (ii) how many devices have surveyed that route section within a pre-set earlier period.

24. The geospatial data system of any preceding Claim in which the route planning algorithm is configured to provide route planning for devices traversing a region in the environment in a manner that provides comprehensive survey data across that region.

25. The geospatial data system of any preceding Claim in which the route planning algorithm is configured to balance a requirement to provide the fastest or shortest route with a requirement to generate crowdsourced survey data across all navigable sections across the entire region.

26. The geospatial data system of any preceding Claim in which the route planning algorithm is configured to balance a requirement to provide a route through a region that optimises a parameter, such as minimising time, distance or energy efficiency, picking up / delivery of packages or people to a schedule, with a requirement to generate crowdsourced survey data across all navigable sections across the entire region.

21. The geospatial data system of any preceding Claim in which the route planning algorithm is configured to provide route planning for devices traversing a region in the environment in a manner that takes into account one or more of the following, as determined by the Al-based data processing sub-system using machine learning, when determining whether to include a route section into a route provided to a device: (i) whether that route section includes a new or altered feature in the environment, such as a map update or new infrastructure, like a new EV charging station; (ii) deviations in actual transit times over that route section compared with predicted times; (iii) a safety parameter associated with that route section, generated by the Al-based data processing sub-system; (iv) a fuel or power efficiency parameter associated with that route section.Generative Al28. The geospatial data system of any preceding Claim in which the system includes a generative Al system configured to generate alternative, synthetic versions of features in the geospatial data.

29. A wearable device, such as a pair of smart glasses, the device including a computer vision sub-system configured to provide spatial awareness to enable the device to map its environment and / or to provide augmented reality for a user in that environment; and in which device is configured so that the computer vision sub-system is re-purposed to create crowdsourced survey data of the environment and to provide that crowdsourced survey data back to the Al-based data processing sub-system for use as training data.

30. A method of generating geospatial data, the method including:(i) operating an Al-based data processing sub-system that is trained to recognise and classify features in an environment;(ii) deploying a number of devices, each including a computer vision sub-system configured to provide spatial awareness to enable the device to navigate through the environment or to provide augmented reality for a user in that environment, and in which the geospatial data system re-purposes the computer vision sub-systems to create crowdsourced survey data of the environment and to provide that crowdsourced survey data back to the AI- based data processing sub-system;and in which the Al-based data processing sub-system uses the crowdsourced survey data as training data.

Images