Navigation method for an autonomous mobile robot, robot and related system

Abstract

The present invention relates to a navigation system for autonomous mobile robot(s) (1), comprising at least one robot and a processing unit for determining a high-definition map (HDMAP) of the robot's movement space. This map comprises: a metric data level mapping the robot's movement space, a level of traffic and operating rules associated with said metric data level, and a semantic mapping level enabling the semantic categorization of areas and objects present in the robot's environment. The high-definition map (HDMAP) is updated by the processing unit using data from objects detected in the robot's environment by sensors.The robot(s) adapt their movement and tasks in real time based on the updated mapping data, the traffic and operating rules linked to the objects and zones on the map, and a semantic hierarchy between these zones and / or objects. Figure for the abbreviation: Fig. 1.

Description

Title of the invention: Navigation method for an autonomous mobile robot, robot and related system. General technical field and prior art.

[0001] The present invention relates to the field of navigation for autonomous mobile robots (AMRs). More specifically, it concerns an efficient navigation method for AMRs operating in complex indoor and outdoor environments.

[0002] In the current context of increasing automation and artificial intelligence, AMRs are playing an increasingly crucial role in logistics distribution, industrial manufacturing, the service sector, and many other fields. Particularly in scenarios requiring precise navigation and high adaptability to the environment, the navigation systems of autonomous mobile robots face demanding requirements.

[0003] Currently, the navigation of automated guided vehicles (AGVs) is at the heart of research advances in intelligent automation. The increasing integration of these technologies in environments of unprecedented complexity—ranging from industrial logistics to automated warehouses and service robots—raises significant challenges. These applications require not only that AGVs navigate with infallible precision, but also that they dynamically adjust their trajectories while rigorously adhering to specific operational guidelines. However, current navigation technologies face significant limitations, particularly when it comes to efficiently processing high-precision map data or implementing optimized route planning strategies.

[0004] Conventional approaches to autonomous mobile robot navigation rely primarily on simplified environmental representations and predefined navigation algorithms, thus limiting their adaptability and effectiveness in complex environments. These limitations are particularly evident in contexts requiring in-depth interaction and the ability to adapt to dynamic environmental changes, such as in factory logistics circuits. To meet such challenges, AMRs must be able to analyze and process large amounts of accurate mapping data in real time to ensure precise environmental perception and facilitate rapid trajectory optimization.

[0005] Furthermore, the evolution of application needs highlights the inherent shortcomings of traditional navigation systems, particularly in terms of Flexibility and extensibility are needed for updating maps and improving route planning algorithms. Faced with complex layouts and volatile operating conditions in industrial environments, these traditional systems frequently require manual intervention to update map data or readjust navigation strategies. This leads to a significant increase in maintenance costs and compromises the agility and responsiveness of AMR systems.

[0006] In this context, the development of an AMR navigation technology capable of efficiently exploiting data from high-precision mapping, promoting efficient route planning in complex environments and flexibly adapting to operational directives, presents itself as a key technical challenge. General presentation of the invention

[0007] The invention aims to overcome the navigation challenges encountered by autonomous mobile robots (AMRs) in complex indoor and outdoor environments, by proposing an innovative navigation technique.

[0008] To this end, it proposes a navigation method for an autonomous mobile robot (AMR), the robot comprising sensors to detect objects in its environment, in which the trajectory of the robot is determined according to a map of the environment transmitted to the robot, characterized in that the map is a high-definition map (HDMAP) which includes: • a level of metric data mapping the robot's movement space, • a level of rules governing traffic and operation within the travel space, • a level of semantic mapping allowing the semantic categorization of the zones and objects present in the robot's environment,

[0009] the high-definition mapping (HDMAP) being updated with the data of the objects detected in the robot's environment by the sensors, the robot adapting in real time its movement and its tasks according to the mapping data thus updated, the rules of circulation and operation linked to the objects and zones of the mapping, and a semantic hierarchy between these zones and / or objects.

[0010] The proposed navigation method is thus based on a multi-level high-definition map (HD Map), significantly improving the reliability, accuracy and efficiency of navigation of autonomous mobile robots in various environments.

[0011] The construction of a multi-level mapping structure provides autonomous mobile robots with a complete dataset that integrates not only detailed geometric information of the environment, as well as rules and semantic information crucial for navigation, thus ensuring complete assistance to AMRs when performing their tasks.

[0012] The method incorporates sophisticated processing of map data, including the fusion of data from multiple sensors, dynamic updating of maps based on environmental changes, and the extraction of semantic information relevant to navigation. Furthermore, it implements advanced optimization algorithms for trajectory planning, dynamic obstacle avoidance, and improved energy efficiency of travel, without being limited to specific methods.

[0013] Thus, the accuracy and adaptability of autonomous mobile robot navigation in these environments are improved. Route planning and the decision-making process are further optimized, ensuring safe, reliable, and cost-effective operation.

[0014] In particular, the proposed method improves the navigation capability of autonomous mobile robots in complex environments and allows the navigation strategy to be adjusted flexibly according to real-time environmental changes, thus considerably improving their performance in dynamic environments.

[0015] In addition, this invention takes into account practical needs in terms of cost and ease of maintenance, reducing the operating costs of the system and simplifying subsequent maintenance work through intelligent processing of mapping data and efficient planning algorithms.

[0016] The proposed method is further advantageously complemented by the following various features, taken alone or in combination: - High-definition mapping (HDMAP) is determined and updated by the implementation of a simultaneous localization and mapping (SLAM) algorithm based on data collected by the sensors of the robot(s) moving in the environment; - the robot's movement is adapted in real time according to traversability data and / or environmental constraints included in the high-definition mapping (HDMAP); - a global route plan is determined remotely from the robot based on high-definition mapping (HDMAP), this route plan being transmitted to the robot; - the robot implements a local plan based on the global plan transmitted to it, the updated high-definition map and the robot's location within this map; - the speed and orientation of the robot are dynamically adjusted by a processing unit; - the raw data collected by the sensors is processed by subsampling and / or filtering processes.

[0017] The invention further relates to an autonomous mobile robot (AMR) comprising#: • sensors to detect objects in its environment, • a control controller, • an onboard computer that transmits control parameters to the controller, • means of transmitting to a processing unit the data from objects detected in the robot's environment by the sensors,

[0018] characterized in that said embedded computer adapts in real time the movement and tasks of the mobile robot according to#: • data from a map updated by said processing unit and transmitted to the on-board computer, • traffic and operating rules related to the objects and areas on the map, and • of a semantic hierarchy between these zones and / or objects,

[0019] the mapping transmitted to the embedded computer being a high-definition mapping (HDMAP) which includes#: • a level of metric data mapping the robot's movement space, • a level of rules governing traffic and operation within the travel space, • a level of semantic mapping allowing the semantic categorization of areas and objects present in the robot's environment.

[0020] It also relates to a navigation system for autonomous mobile robot(s) (AMR), comprising at least one such robot according to claim 8,

[0021] the system comprising a processing unit which determines a high-definition map (HDMAP) of the robot's movement space, said high-definition map (HDMAP) comprising#: • a level of metric data mapping the robot's movement space, • a level of traffic and operating rules associated with this level of metric data, • a level of semantic mapping allowing for the semantic categorization of areas and objects present in the robot's environment,

[0022] the system also comprising access to at least one library of general rules and at least one library of semantic data, the processing unit defining the level of the rules of circulation and operation as well as the level of semantic mapping by referring to the content of these libraries,

[0023] the high-definition mapping (HDMAP) being updated by the processing unit with data from objects detected in the robot's environment by the sensors, the robot(s) adapting in real time their movement and their tasks according to the mapping data thus updated, the traffic and operating rules linked to the objects and areas of the mapping, and a semantic hierarchy between these areas and / or objects.

[0024] The invention also relates to a computer program comprising code instructions for implementing the method when executed on the processor of an autonomous mobile robot (that of the robot's onboard computer). Brief description of the drawings

[0025] Other features and advantages of the invention will become apparent from the following description, given by way of non-limiting example, with reference to the accompanying figures in which: • [Fig.1]: Schematic representation of the architecture of a navigation system according to one embodiment of the invention#; • [Fig.2]: Schematic representation illustrating an example of metric layer data from an HDMAP# map; • [Fig.3]: Schematic representation illustrating an example of metric layer data and rule layer of an HDMAP# map; • [Fig.4]: Schematic representation illustrating an example of data from the semantic layer of an HDMAP# map; • [Fig.5]: illustration of the three data levels for the construction of the three layers of the HDMAP map.

[0026] Description of one or more implementation and realization methods General architecture

[0027] The navigation system illustrated in [Fig.1] comprises one or more autonomous mobile robots 1 equipped with various sensors 2a, 2b, etc., such as LiDAR, IMU, cameras or any other sensor useful for navigation.

[0028] Each robot 1 is equipped with communication means enabling it to exchange information with a remote processing unit 3. This processing unit 3 receives information from sensors 2a, 2b, etc., and implements, from this data, a simultaneous localization and mapping algorithm (SLAM, for Simultaneous Localization and Mapping) 4. It thus generates the different layers of a high-precision map (HDMAP) which will be described below.

[0029] The processing unit 3 uses in particular one or more libraries 5 of semantic and regulatory data to enrich the geometric mapping information established from the measurements of the sensors 2a, 2b, etc., of the robots 1.

[0030] The high-precision mapping is stored by unit 3 in a memory 6 of the system and comprises, as detailed below, a metric layer 6a, a rules layer 6b and a semantic layer 6c.

[0031] Each robot 1 includes a controller 7 which provides commands to the robot's motors to generate its controlled movement. These commands include navigation instructions (speed of movement, orientation of the robot in its plane of movement), as well as commands assigning the robot tasks to perform (movement of the robot arm, actuation of its gripper to grasp an object, etc.).

[0032] These different commands are determined according to a local plan established by an on-board computer 8 on the robot 1, incorporating a planner module 8a. This local plan determines in real time the movement parameters and the tasks that must be executed by the robot 1 under the control of the controller 7.

[0033] To do this, the embedded computer 8 uses the high-precision mapping (HDMAP) established by the remote processing unit 3. This mapping is continuously updated as the robots 1 move and as readings are taken by their sensors 2a, 2b, etc. It is transmitted in real time to all the robots.

[0034] Local planning also depends on the location of robot 1 in this mapping, as well as on a global plan previously determined (offline) by a remote global planner 9 integrated into the system.

[0035] The location of robot 1 is that which is assigned to it in the HDMAP map transmitted by the processing unit 3.

[0036] The system's global planner 9 is synchronized with enterprise resource planning (ERP) and warehouse management systems (WMS), enabling real-time updates of routes based on production priorities and logistical demands. This integration ensures that the robots 1 work in harmony with all industrial operations, thereby improving the consistency and efficiency of workflows.

[0037] The integrated route planning approach (described later) leads to a significant reduction in downtime, increased productivity, and better resource allocation. By providing a navigation method that anticipates and adapts to the specific requirements of the industrial environment, the AMRs are becoming even more strategic tools in optimizing production and logistics processes. Multi-level mapping

[0038] We describe below the three layers of the high-definition mapping (HDMAP) used. Metric layer

[0039] The data at this level are cartographic data representing the 2D or 3D environment in which the robots 1 move. They superimpose geometric and topographic data (plans) as well as data enabling the identification of objects or areas of the mapped environment.

[0040] Fig. 2 illustrates an example of this data level in the case of an industrial environment comprising milling workshops 10 and painting workshops 11.

[0041] The mapping data indicates the position of these workshops 10 and 11 within the work environment. It also provides the position of dedicated racks 12, on which the raw parts are stored, and of separate racks 13 where the validated parts are placed. Furthermore, it includes information identifying each shelf of the racks 12 and 13, as well as their height. The mapping data may also include information on the height of the work surfaces of workshops 10 and 11.

[0042] Generation by the SLAM algorithm and integration of pre-loaded data

[0043] The geometric data of this environment are captured by the robot(s) 1 using a variety of sensors 2a, 2b, etc., such as LiDAR, stereoscopic cameras, infrared sensors, or sonar. These sensors collect environmental information from different angles to ensure complete data coverage and depth.

[0044] However, high-precision mapping can also be enhanced by preloaded data relating to the static environment in which the robots operate. This pre-existing information can come from architectural plans, CAD models, or previous topographic surveys. Integrating this data provides an accurate initial mapping basis, thus reducing the time required for field mapping and improving the overall accuracy of the system.

[0045] The data thus acquired, whether from the robots' sensors or from pre-loaded sources, are integrated into the first information layer of the high-precision mapping. This combined approach makes it possible to produce an accurate and comprehensive representation of the industrial environment.

[0046] The first phase of the navigation process involves, for this purpose, the generation by the processing unit 3 of maps by means of a localization algorithm and Simultaneous mapping (SLAM method, for Simultaneous Localization and Mapping). For a general description of SLAM methods, please refer to the following publication: • Durrant-Whyte, H., & Bailey, T. (2006). "Simultaneous Localization and Mapping: Part I". IEEE Robotics & Automation Magazine, 13(2), 99-110.#

[0047] This step can be achieved by combining pre-loaded data with surveys conducted by robots, thus enabling real-time updates of the mapping based on environmental changes. This integration of pre-existing and real-time data ensures increased flexibility and adaptability to environmental changes.

[0048] During this phase, the AMR can be manually controlled via a remote control, allowing direct human intervention in the mapping process if necessary. Furthermore, it is possible to configure predefined start and finish points, between which intermediate coordinate points are inserted to define a specific route.

[0049] Following this predefined route, the robot 1 implements the SLAM 4 method to generate the map in a manner synchronized with its movement, thus progressively enriching its representation of the navigated space, while integrating the preloaded data. Nature of the maps

[0050] The nature of the maps produced varies depending on the types of sensors used and the pre-loaded data available.

[0051] LiDAR sensors, highly valued for their accuracy, are generally used to produce point clouds providing a detailed three-dimensional representation of the environment. These maps are particularly useful for applications requiring high navigation accuracy, such as in cluttered or unstructured environments.

[0052] Preloaded data, such as architectural plans or CAD models, provides a solid basis for mapping, allowing the system to have an accurate representation of the static environment from the outset. This information can be used to supplement or refine the data collected by the onboard sensors.

[0053] On the other hand, visual sensors, often used in addition to or as an alternative to LiDAR, can generate grid maps that simplify the visualization of space through regular divisions. These grid maps transform the complex environment into an easy-to-manipulate model.

[0054] These grid maps can also include information on the practicability of the corresponding areas, facilitating route planning and obstacle avoidance.

[0055] Visual sensors also make it possible to generate cost maps which assign traversability values ​​to different areas of the map, thus helping the AMR to plan routes by avoiding obstacles or choosing the most efficient paths in terms of energy or time.

[0056] These traversability values ​​may depend on the pre-loaded static environment and the dynamic obstacles detected in real time.

[0057] The determination and use of traverseability values ​​are well known in autonomous navigation. For an example of implementation, reference may be made to the following publication: • Papadakis, P. (2013). "Terrain Traversability Analysis Methods for Unmanned Ground Vehicles: A Survey". Engineering Applications of Artificial Intelligence, 26(4), 1373-1385.#

[0058] In general, map generation is not simply a static reproduction of the environment, but involves a continuous interaction between the robot, the pre-loaded data and its environment.

[0059] The data collected by the sensors 2 are continuously transmitted to the processing unit 3, which updates the metric layer of the map in real time and continues to update this layer and the HDMAP map by continuous localization and SLAM mapping processing, in combination with the pre-loaded data.

[0060] Real-time map updates allow for the incorporation of newly detected features or environmental changes. This dynamic process is particularly well-suited to rapidly evolving environments, such as construction sites or industrial areas where the layout can change frequently. The ability to react to new obstacles or changes is crucial for navigational safety and efficiency.

[0061] This hybrid approach, combining pre-loaded data of the static environment and dynamic information captured in real time by the robot's sensors, offers a robust and flexible solution for AMR navigation. It not only reduces the time and resources required for initial mapping, but also improves the accuracy and reliability of navigation in complex environments.

[0062] By integrating these different sources of information, the navigation system can anticipate foreseeable obstacles while adapting to unforeseen changes, thus ensuring the safe and efficient movement of autonomous mobile robots in their operational environment. Sensor data processing

[0063] The measurements from sensors 2 can undergo various advanced processing techniques to ensure the accuracy and usefulness of the data provided. For example, downsampling techniques can be applied to point clouds generated by LiDAR sensors to effectively reduce the data volume. By selecting one point out of n or by using adaptive decimation methods, the amount of data to be processed is reduced while preserving essential information, thus improving the efficiency of subsequent processing.

[0064] Furthermore, filtering techniques, such as low-pass filters, Kalman filters, or median filters, are used to eliminate noise present in sensor data. For example, by applying a median filter to images captured by a camera, artifacts caused by electronic noise can be reduced, thus improving data quality and object detection accuracy.

[0065] Data fusion techniques are also used to integrate information from different sensors. For example, combining data from a LiDAR sensor with that from a stereoscopic camera produces a richer and more accurate representation of the environment, enhancing the stability and reliability of the environmental model. This fusion compensates for the individual limitations of each sensor and improves the robot's overall perception.

[0066] Finally, feature extraction techniques, such as interest point detection (e.g., corners or edges) and visual descriptor extraction, make it possible to identify key information from large volumes of data. These features serve as the basis for object recognition, precise localization, and autonomous decision-making by robots.1 Period layer

[0067] This advanced level deepens and refines the guiding principles of autonomous mobile robot navigation, ensuring that their decision-making in complex environments is both compliant with pre-established safety standards and carried out with optimal efficiency. The construction of this level is based on a meticulous analysis of environmental information provided by the metric layer, transforming this information into a series of operational rules and concrete trajectory guidelines to guide the navigation of autonomous mobile robots in various scenarios. Definition and application of advanced rules

[0068] Within the rules layer, route definition is not limited to identifying usable areas; it extends to optimizing trajectory choices. This includes, for example, calculating the shortest paths to the destination using efficient planning algorithms, avoiding high-risk areas such as narrow passages prone to congestion or areas with high pedestrian traffic, and taking into account operational constraints such as restricted access times to certain areas.

[0069] Furthermore, for specific operational environments such as industrial sites or public spaces, this layer incorporates traffic rules and operational standards specific to the environment. For example, speed limits can be adjusted according to the zones to ensure safety, right-of-way priorities can be established at intersections to avoid collisions, and traffic restrictions can be applied at certain times of day to coordinate with other operational activities. Adaptability to dynamic environments

[0070] The design of the rules layer fully takes into account the dynamics of the environment, dynamically adjusting the navigation strategy of autonomous mobile robots through a real-time feedback mechanism. For example, by constantly monitoring moving obstacles such as people, other robots, or vehicles, and changes in environmental conditions such as temporarily closed passages or construction zones, the AMRs can make instantaneous decisions regarding trajectory or speed adjustments to ensure continuous and safe navigation. Cooperative Navigation Mechanism

[0071] In scenarios involving the collaboration of multiple AMRs, the rules layer emphasizes the importance of cooperative navigation. By defining communication protocols and cooperation rules between robots—such as allocating right-of-way at certain intersections, coordinating routes to avoid congestion, or developing avoidance strategies in confined spaces—it ensures effective collaboration and minimizes conflicts within a multi-AMR system. Furthermore, this can include optimizing task allocation and execution order to improve the overall operational efficiency of the system. Balancing safety and efficiency

[0072] In the design of the rule layer, the balance between safety and efficiency is considered a fundamental principle. Each rule is established and implemented with the objective of ensuring the safety of mobile robot operations. autonomous while maximizing efficiency in task execution. This includes assessing and avoiding potential risks encountered by AMRs in complex environments, as well as seeking optimizations in routes and operations while ensuring safety.

[0073] This sophisticated rule layer design enables precise and efficient navigation of autonomous mobile robots in various complex environments, significantly improving the intelligence level and operational efficiency of AMR systems. It paves the way for the future development of automation technology. Example

[0074] The high-definition mapping (HDMAP) rule layer is developed by the processing unit 3 in addition to the map generated by the SLAM algorithm 4 and its updates.

[0075] In what follows, we consider the case of a point cloud map created using a LiDAR system.

[0076] This three-dimensional map, rich in geometric details, serves as a basis for the integration of essential structural elements such as roads, traffic lanes, intersections, pedestrian crossings, stop lines and barriers.

[0077] On this point cloud map, each element is meticulously identified and annotated to meet the high standards of HDMAP. Roads and lanes are drawn to accurately reflect their actual geometry, while intersections are detailed, including information on traffic lights and road signs to enable precise and safe navigation. Pedestrian crossings and stop lines are clearly marked to ensure that AMRs comply with priority rules and contribute to the safety of industrial logistics areas. Barriers and other permanent obstacles are also integrated to prevent collisions and optimize routes.

[0078] This first layer of the HDMAP (data layer, also called metric layer in this text), focused on infrastructure and signaling, is essential because it establishes the framework in which AMRs operate.

[0079] Once this first layer is established, the processing unit 3 associates specific operating rules with the different areas of the map, based on the structural elements present in those areas. To do this, the processing unit 3 segments the map into areas defined according to geographical or functional criteria. For each area considered, it consults the rules stored in the library 5 and applies them according to the structural elements identified in the area concerned.

[0080] Thus, the local rules are derived from the general rules of library 5, interpreted and adapted according to the structural characteristics of the local environment.

[0081] The processing unit 3 therefore enriches the first layer of data from the sensors with a second layer, which associates local traffic and operating rules with the different areas of the map.

[0082] These rules are carefully defined to meet the specific mobility and regulatory compliance needs of autonomous mobile robots. They include, but are not limited to, areas where movement is permitted, thus preventing AMRs from entering regions where their presence might be restricted or inappropriate. Speed ​​limits are also specified for each zone, adapted to local conditions and operational requirements, thereby preserving safety while optimizing traffic flow.

[0083] In addition, priority rules are established to regulate the interaction of autonomous mobile robots with other road users, including traditional vehicles, pedestrians, and other robots. This includes detailed instructions on how to navigate intersections, roundabouts, and other convergence points, ensuring that the AMRs operate smoothly and predictably.

[0084] In the example illustrated in [Fig.3], the complementary elements shown correspond to the local rules which are added and superimposed on the data layer according to the structural elements initially detected in the environment.

[0085] Typically, the rules in library 5 include speed limits, one-way traffic patterns, right-of-way rules, and the use of signals such as traffic lights. All of these directives are interpreted locally, and the local traffic and operating rules are recorded in the rules management layer and associated with the different travel zones on the map, ensuring that mobile robots incorporate them into their trajectory planning and navigation.

[0086] Traffic lights, for example, are integrated into the map during the development of the rule layer. They can represent virtual signals defined in the HDMAP, which robots must take into account during their navigation. These lights can also correspond to physical devices present in the environment, detected by the robot's sensors.

[0087] Travel lanes, or traffic lanes, are an integral part of the metric layer as structural elements of the environment. They are determined during map generation by identifying usable areas and delimiting possible traffic lanes for robots. The rules associated with These bands, such as speed limits or traffic directions, are then defined in the rules layer.

[0088] The implementation of the traffic rules that appear on the HD MAP is then carried out on each robot during the determination of the local planning by the on-board computer 8, according to the environment detected by the sensor system 2 of the robot 1 and its location on the HDMAP.

[0089] Advanced algorithms further enable the robot 1 to understand and correctly interpret environmental signals, including physical road signs detected by its sensors, road markings, and other navigation indicators. Thus, the robot adjusts its behavior not only according to the trajectory calculated by the local planner, but also by taking into account the signals detected in real time in its environment.

[0090] This adaptability is crucial, particularly in complex environments where traffic conditions can change rapidly. AMRs equipped with this HDMAP technology can thus adjust their behavior in real time to ensure safe and compliant driving.

[0091] The second layer of the HDMAP offers several significant advantages. It enables precise regulation of the speed and maneuvers of autonomous mobile robots, which is vital for minimizing the risk of accidents and improving operational efficiency. Furthermore, by providing clear guidelines on authorized traffic areas and priority rules, this layer helps prevent regulatory violations that could lead to penalties or disruptions in the operations of autonomous mobile robots.

[0092] By consolidating these rules in the HDMAP, developers of navigation systems for AMRs ensure that these devices can operate autonomously in a variety of environments while complying with the safety and compliance standards required by local authorities and site managers. This makes AMRs not only safer, but also better suited to seamless integration into the overall urban and industrial mobility ecosystem.

[0093] The design rules of the HDMAP are rigorously applied to ensure that each element of the map is both accurate and functional. These rules take into account not only the physical aspects of the environment, but also the legal and regulatory standards that govern traffic and coexistence in shared spaces. By following these guidelines, the HDMAP allows AMRs to navigate complex environments with a high degree of confidence, taking into account potential interactions with human-driven vehicles and pedestrians.

[0094] The use of HDMAP in the autonomous navigation of autonomous mobile robots transcends traditional navigation applications. It enables not Not only does it provide precise localization and navigation, but it also opens the door to more complex operations, such as autonomous driving in dense urban environments, logistics management in vast warehouses, and emergency response where accuracy and speed are crucial. The ability of autonomous mobile robots to interpret and respond appropriately to navigational challenges depends largely on the quality and accuracy of the HDMAP.

[0095] Consolidating a rule layer on the map generated by the SLAM 4 process is a crucial step that enhances the ability of mobile robots to operate autonomously and efficiently in diverse environments. This systematic approach not only standardizes the data used for navigation but also ensures that AMRs can operate safely and in compliance with regulatory and operational requirements. Semantic layer

[0096] The semantic expression level represents the third advanced level of the proposed multi-level navigation system, established on the basis of metric and rule layers.

[0097] It is implemented by the processing unit 3, which assigns detailed semantic categorizations to objects and areas identified in the environment.

[0098] This layer enriches the mapping by associating contextual and functional information with each element. The semantic hierarchy thus defined allows autonomous mobile robots (AMRs) to understand not only where objects are located, but also what they represent and how to interact with them.

[0099] To illustrate semantic mapping technologies in the field of mobile robotics, reference can be made to the following publications: • Nüchter, A., Hertzberg, J., Surmann, H., & Thrun, S. (2008). "Towards Semantic Maps for Mobile Robots". Robotics and Autonomous Systems, 56(11), 915-926.# • Sünderhauf, N., Dayoub, F., McMahon, S., Talbot, B., Schulz, R., Corke, P., Wyeth, G., & Milford, M. (2016). "Meaningful Maps with Object-Oriented Semantic Mapping". Proceedings of the IEEE / RSJ International Conference on Intelligent Robots and Systems (IROS), 5079-5085.#

[0100] In the example illustrated in [Fig.4], the semantic layer associates physical objects with specific terms#(in bold in the following text): • Milling workshops Fl to F3, • PI and P2 painting workshops, • El to E7 storage racks.

[0101] Pedestrian crossings PP1 to PP5, the various speed limits L1 to L3, and traffic lanes VI to V6 are also identified and semantically categorized. People present DI to D4 are detected in real time by the robot sensors, and their position is continuously updated on the HDMAP map.

[0102] All this information, as well as the semantic qualifications of the different zones and objects identified in the mapped environment, are integrated into the third layer of the HDMAP mapping by the processing unit 3. This creates a multi-layered map comprising a geometric layer, a rules layer and a semantic layer, thus offering a complete and detailed representation of the industrial environment enriched with a contextual understanding.

[0103] This level goes beyond simply categorizing areas by associating specific functions, such as "loading zones" or "unloading zones". It also identifies and categorizes various objects present in the environment, such as obstacles, moving objects, and safety zones, providing AMRs with a rich set of environmental information and behavioral guidelines.

[0104] Real-time environmental considerations are achieved through a combination of HDMAP mapping updates and processing carried out on the robots. The sensors on the robots 1, such as cameras, LiDAR, and other devices, detect changes in the environment in real time, such as the appearance of new obstacles or the movement of people (DI to D4). This information is processed by the robot's onboard computer 8, which uses object recognition and semantic classification algorithms. The processed data is used to update the semantic layer of the HDMAP, either locally on the robot or by transmitting it to the processing unit 3 for a global update.

[0105] For example, AMRs can autonomously navigate to the appropriate location for loading goods based on the "loading area" annotation in the semantic layer. They can also avoid areas marked as "restricted areas" to ensure safe and efficient operation. The semantic layer thus provides crucial contextual information for the robots' autonomous decision-making.

[0106] In the example in [Fig.4], the semantic and metric layers allow the robot to be precisely located in its environment, for example: "The robot is located near the milling machines F2." This precise localization is possible thanks to the association of geometric data with semantic information, allowing the robot to understand not only its position, but also the surrounding context.

[0107] These three layers are used together for overall navigation planning. The semantic layer provides a deep understanding of The environment guides task planning. For example, if the robot receives a mission such as "go retrieve parts from rack E6", it uses semantic information to identify the exact location of rack E6. Then, it plans its route using the rules layer to respect speed limits, right-of-way priorities, etc., and the geometry layer to generate a precise trajectory while avoiding obstacles. Global planning

[0108] The information provided by high-definition multi-level mapping (HDMAP) plays a crucial role in route planning and localization processes for autonomous mobile robots (AMR).

[0109] Thanks to HDMAP, the system associates detailed road infrastructure data—specifically adapted to areas such as industrial logistics—with traffic rules and semantic information, thus facilitating precise and safe navigation in various environments.

[0110] The system implements global route planning, which can be carried out offline or online.

[0111] This task consists of defining a route that respects not only traffic rules, but also the specific characteristics of the work environment, from the starting point to the destination. The method is designed to be flexible, supporting both offline planning, ideal for constant and predictable operating environments such as those of industrial facilities, and online planning, necessary in contexts where working conditions are dynamic and likely to change frequently.

[0112] In industrial logistics, where the routes of autonomous mobile robots are generally fixed, offline planning minimizes the use of computing resources, thereby optimizing response times and reducing operational costs. By eliminating the need to constantly recalculate the same routes, this approach improves not only the efficiency but also the reliability of autonomous mobile robot operations.

[0113] For scenarios where operational requirements are subject to frequent changes, online planning becomes essential. This planning method begins with a topological framework based on the semantic data of the third layer of the HDMAP. This speed-oriented process allows AMRs to adapt instantly to new configurations or unforeseen interruptions, thus ensuring operational continuity and security.

[0114] After the topological route has been established, more detailed planning is carried out using the information from the first layer of the HDMAP. This step incorporates the traffic rules defined in the second layer, ensuring that all AMR actions comply with current regulations and internal safety and efficiency protocols.

[0115] Thus, the semantic layer is an integral part of the mapping by providing the contextual information necessary for the robot to correctly interpret its environment and make appropriate decisions.

[0116] The introduction of the semantic expression level allows AMRs not only to follow basic navigation rules, but also to adjust and optimize their navigation decisions based on the actual meanings of the environment. For example, knowing that a particular area is a "risk zone," the robot can decide to adopt a reduced speed or modify its route to avoid that zone, even if this is not explicitly dictated by traffic rules.

[0117] This intelligence and flexibility greatly enhance the ability of autonomous mobile robots to perform complex tasks in dynamic environments, such as the accurate delivery of goods in constantly changing work environments or the provision of services in densely populated public spaces.

[0118] In summary, the design and implementation of the semantic expression level provide AMRs with a new way of perceiving and interacting with their environment. By enabling them to understand and adapt more deeply to their operational environment, they improve the efficiency and quality of task execution. By fully integrating and utilizing semantic information from the environment, this invention significantly increases the intelligence level of autonomous mobile robots, laying a solid foundation for their deployment and application in a wider range of fields.

[0119] The semantic layer therefore plays a crucial role in enriching the map with a deep and contextual understanding of the environment. It focuses on the precise identification and classification of the various objects present in the navigation space, such as speed limit signs, pedestrian crossings, loading zones and other significant markers.

[0120] The elements integrated into the semantic layer are not limited to a simple visual representation; they are associated with enriched metadata that provides added value for autonomous navigation. Each semantic object has specific attributes that make it identifiable and interpretable by AMRs. For example, a speed limit sign is not only represented by its geometric position, but is also associated with information on the maximum permitted speed that the AMR must respect.

[0121] The distinction between the rule layer and the semantic layer lies in the fact that the rule layer defines the general behaviors that AMRs must adopt under certain conditions (for example, yielding at an intersection), while the semantic layer provides contextual information about specific objects and areas of the environment (for example, identifying that an area is a "P2 paint shop").

[0122] The objects in the semantic layer also serve as reference points for localization. Thanks to the precise position of the semantic markers integrated into the map, an AMR that recognizes one of these markers can improve its localization within the environment. This semantic localization capability complements localization methods based on distance measurements, such as those used in traditional LiDAR systems. Indeed, semantic localization makes it possible not only to determine where the AMR is located, but also to understand its surroundings and how it should interact with these elements.

[0123] The expression of semantic information is inherently multidimensional, integrating both the spatial dimension and the functional context of objects. This wealth of information allows AMRs to perceive their environment not as a simple physical space, but as a set of interactive elements with which they must coordinate. This enables them to navigate more naturally and intuitively, reacting appropriately to signals and changes in their environment.

[0124] Furthermore, semantic information helps to build more complex behavioral models for AMRs, enabling them to perform tasks with greater autonomy and precision. For example, when an AMR identifies a "loading zone," it can activate specific protocols for approach and loading, ensuring that operations are carried out safely and efficiently.

[0125] The semantic layer significantly enhances the navigation of autonomous mobile robots by providing them with an operational framework that goes beyond simple locomotion. It involves dynamic interaction with the environment, fostering autonomous decision-making based on a deep and contextualized understanding. This sophistication in navigation opens up new possibilities for the use of autonomous mobile robots in a variety of applications, from urban environments to industrial complexes, where precision and responsiveness are essential for successful operations. Robot location

[0126] The localization of autonomous mobile robots (AMR) based on HD MAP mapping exploits the richness of map data for navigation Precise and efficient, this process begins with topological localization, which uses the HDMAP's 6c semantic layer to determine the AMR's relative position with respect to significant landmarks in the environment. The semantic layer provides contextual information about environmental elements, such as workshops, storage areas, pedestrian crossings, and so on. For example, the AMR can be identified as being near paint shop P2 or opposite storage rack E5. This semantic information allows the robot to determine its initial position more efficiently based on clearly defined reference points.

[0127] This initial localization is then refined by integrating data from the first layer of the HDMAP (layer 6c), which contains detailed information on the physical infrastructure and geometry of the environment. By combining this information with data collected in real time by the AMR's sensors, such as LiDAR or cameras, the system can perform precise localization by correlating predefined structural elements with current observations. This process improves the reliability and accuracy of the AMR's positioning within its working environment.

[0128] This advanced localization method effectively addresses three major challenges encountered in mobile robotics: • Global localization: Identifying the initial position of the AMR within a broad context, often necessary after system startup or a significant interruption. Thanks to the semantic layer, the robot can quickly estimate its position by recognizing key environmental landmarks. • Position tracking: Continuously maintaining the precise location of the AMR as it navigates and interacts with the environment, using the fusion of first-layer metric data and real-time sensor observations to ensure that all actions are based on up-to-date location information. • Kidnapping problem: The ability of the AMR to re-determine its position after being unexpectedly moved to an unmapped location or after a significant localization error. By using semantic information to quickly identify its new approximate position and metric data to refine this estimate, the robot can efficiently reorient itself within its environment.

[0129] Integrating these localization capabilities into AMR systems offers multiple advantages. It not only enables more robust and secure autonomous navigation, but also optimizes logistical operations by reducing route errors and maximizing travel efficiency. Furthermore, accurate location tracking helps minimize the risk of collisions and incidents, thus protecting both infrastructure and equipment investments.

[0130] The use of HDMAP for localization represents a significant technological advancement in the management of autonomous mobile robots, offering a sophisticated solution to meet the complex challenges of navigation and localization in dynamic industrial and commercial environments. This approach not only improves robot autonomy but also their integration into broader operational processes, ensuring optimal performance and increased safety. Local planning

[0131] Once the AMR has determined its exact position using advanced localization systems incorporating HDMAP, it is ready to follow the pre-planned route. However, the environments in which AMRs operate are often subject to dynamic changes requiring continuous real-time route adaptation. To effectively manage these variables, the use of a local planner becomes essential.

[0132] The local planner, integrated into the planner module 8a of the robot's onboard computer, allows the AMR to adjust its trajectory in real time based on information received from its sensors and data from the HDMAP. It continuously analyzes the robot's immediate environment, detects potential obstacles, and evaluates the necessary modifications to the initial trajectory to ensure safe and efficient navigation.

[0133] This local planner plays a crucial role in adapting the AMR's navigation to the fluctuating conditions of its working environment. This adaptation may include reacting to temporary obstacles, such as newly installed equipment or moving objects like other vehicles or operational personnel. Its main function is to enable the AMR to respond appropriately and safely to these obstacles, making decisions such as stopping, going around the obstacle, or adjusting its speed.

[0134] To ensure safe and compliant navigation, the local planner uses the information contained in the second layer of the HDMAP, which details the traffic rules specific to the AMR's environment. These rules include guidelines on speed limits, right-of-way, and avoidance protocols. By adhering to these guidelines, the local planner ensures that all actions taken by the AMR are not only efficient but also safe and compliant with established regulatory standards.

[0135] The effectiveness of the local planner relies on its ability to integrate and process AMR sensor data and HDMAP information in real time. This integration enables accurate environmental assessment and immediate responsiveness to changes. For example, if a new obstacle is detected on the planned route, the planner can instantly recalculate a new trajectory to avoid the obstacle while maintaining route efficiency.

[0136] The operation of the local scheduler generally involves several key steps: 1. Environmental perception: The robot uses its sensors (LiDAR, cameras, etc.) to detect obstacles and changes in its immediate environment. 2. Local modeling: Sensory information is used to construct a local representation of the environment, often in the form of a cost map or an occupancy grid. 3. Trajectory generation: Based on local modeling and HDMAP information, the local planner generates an optimal trajectory that avoids obstacles while respecting task constraints and traffic rules. 4. Control and execution: The corresponding movement commands are sent to the robot's actuators to follow the planned trajectory.

[0137] The application of the local planner is not limited to simple obstacle avoidance; it also encompasses the continuous optimization of AMR trajectories to maximize productivity and minimize delays. This optimization takes into account not only physical obstacles, but also fluctuating traffic conditions, processing times at workstations, and other critical operational factors.

[0138] Local planning enhances the navigation system of autonomous mobile robots by making it more dynamic and adaptable to changing conditions. By leveraging the structures and rules provided by HDMAP and integrating real-time decision-making capabilities, it ensures safe, compliant, and optimized navigation, essential for maintaining continuity and efficiency in complex industrial environments.

[0139] Local route planning, performed in real time, is therefore an essential component of AMR navigation, although it is not used at every moment of the journey. This online navigation strategy is specifically designed to intervene in situations where the conditions of the initial route change unexpectedly, requiring immediate responsiveness to ensure efficient and safe navigation.

[0140] Local planning is primarily activated in two types of critical scenarios: first, when unexpected obstacles block the pre-established path of the AMR#; second, when route modifications are required in response to changes in the work environment or operational directives. These obstacles may include temporary elements such as debris, mobile equipment, or personnel flows in industrial areas.

[0141] When an obstacle is detected on the route, the AMR navigation system immediately performs an analysis to assess whether the obstacle requires a simple pause or a detour. If a detour is necessary, the local planner re-evaluates the available route options based on current data and taking into account safety and productivity parameters. This re-evaluation may include searching for alternative routes, estimating additional travel time, and verifying compliance with operational standards.

[0142] Furthermore, local planning allows for continuous trajectory optimization by adjusting navigation parameters according to dynamic site conditions. This includes adapting the AMR's speed, modifying work sequences, or temporarily rescheduling tasks based on operational priorities. This flexibility is crucial in complex industrial environments where requirements can fluctuate rapidly depending on production processes or emergencies.

[0143] Local route planning is also seamlessly integrated with global management systems, enabling synchronization between local AMR navigation and broader operational objectives. This integration ensures that trajectory adjustments contribute to the overall efficiency of the entire logistics system, minimizing disruptions and maximizing responsiveness to operational demands. Controller

[0144] Once the route has been meticulously planned, whether global or local, the autonomous mobile robot (AMR) begins moving to follow the path, generally represented as a series of coordinate points. The main challenge for motion control lies in the ability to steer the AMR so as to precisely follow these points, which requires fine-tuning and an adaptive response to path data.

[0145] To navigate efficiently along the planned route, the controller 7, integrated into the robot's onboard computer 8, uses advanced motion control systems. Among these systems, predictive modeling methods such as model-based predictive control (MPC) are employed. MPC is particularly effective because it allows the AMR's trajectory to be predicted and optimized. by calculating future actions based on a dynamic model of the robot and its interaction with the environment. This method uses algorithms that take into account movement constraints and operational objectives to adjust the trajectory in real time, thus ensuring precise control and a significant reduction in path errors.

[0146] In parallel, the system also supports methods that do not require complex modeling, such as pure tracking. This technique, also implemented at the embedded computer level, is less computationally intensive and well-suited to environments where high reaction times are required. Pure tracking is based on simple path geometry and aims to guide the AMR towards a target point on the route, continuously adjusting the steering angle to keep the robot on the intended path.

[0147] The application of these motion control techniques not only enables smooth and precise navigation, but also increases operational safety. By anticipating obstacles and dynamically adjusting speed and direction, AMRs can better react to changing environmental conditions and avoid potential collisions, while maintaining optimal efficiency in performing their tasks.

[0148] Integrating these motion control systems with overall operations management systems is essential to align the navigation of autonomous mobile robots with the company's strategic objectives. This synchronization ensures that robot movements not only conform to navigation plans but are also aligned with the real-time needs of production and logistics.

[0149] The advanced motion control technologies integrated into this invention substantially improve the ability of autonomous mobile robots to follow complex routes accurately and efficiently. Whether through predictive control methods or simple geometry-based approaches, the AMR is equipped to respond flexibly and safely to navigation requirements in dynamic and potentially unpredictable environments.

[0150] When the AMR is in motion, it uses a sophisticated array of sensors 2 to perceive its environment in real time, which is essential for precise and efficient navigation. These sensors continuously collect data on surrounding conditions, ranging from obstacle detection to the recognition of traffic signals. This information is used in conjunction with high-definition multi-level mapping (HDMAP), enabling the AMR to calibrate its position and adjust its trajectory according to actual conditions.

[0151] This sequence of actions—perception, localization, planning, and movement control—constitutes a continuous optimization cycle that allows the AMR to Navigate autonomously in complex environments. At each stage, the AMR assesses its environment, adapts its trajectory and updates its navigation parameters to maintain optimal efficiency and meet operational requirements.

[0152] In summary, the constant interaction between the AMR sensors and the HDMAP, combined with advanced motion planning and control capabilities, ensures that the AMR can navigate safely and efficiently, adapting in real time to the multiple challenges of its environment. This integrated approach illustrates the technological sophistication required to operate in modern industrial and logistics environments, where precision and responsiveness are essential.

[0153] Using location and trajectory data, the system can perform local planning to handle dynamic events, such as pedestrians crossing the road or obstacles blocking the path. Local planning allows the robot to bypass these obstacles or wait, depending on the situation, thus avoiding accidents. Finally, the trajectory data is sent to the controller 7, which generates the necessary commands for the robot to move autonomously. This process is repeated in a loop to adjust the robot's path in real time and update the map as needed.

Claims

Demands

1. A navigation method for an autonomous mobile robot (AMR), the robot comprising sensors for detecting objects in its environment, wherein the robot's trajectory is determined based on an environment map transmitted to the robot, characterized in that the map is a high-definition map (HDMAP) comprising: • a metric data level mapping the robot's movement space, • a level of rules for movement and operation within the movement space, • a semantic mapping level enabling the semantic categorization of areas and objects present in the robot's environment, the high-definition map (HDMAP) being updated with data on objects detected in the robot's environment by the sensors, the robot adapting its movement and tasks in real time based on the updated mapping data.rules governing traffic and operation related to the objects and areas of the map, and a semantic hierarchy between these areas and / or objects.

2. A method according to claim 1, wherein the high-definition mapping (HDMAP) is determined and updated by implementing a simultaneous localization and mapping (SLAM) algorithm based on data collected by the sensors of the robot(s) moving in the environment.

3. Method according to claim 1, wherein the robot movement is adapted in real time according to traversability data and / or environmental constraints included in the high-definition mapping (HDMAP).

4. A method according to claim 1, wherein an overall route plan is predetermined remotely from the robot based on high-definition mapping (HDMAP), this route plan being transmitted to the robot.

5. A method according to claim 4, wherein the robot implements local planning based on the global planning which is transmitted, along with updated high-definition mapping and the robot's location within that mapping.

6. A method according to claim 1, wherein the speed and orientation of the robot are dynamically adjusted by a processing unit.

7. A method according to claim 1, wherein the raw data collected by the sensors are processed by subsampling and / or filtering processes.

8. Autonomous Mobile Robot (AMR) comprising: • sensors for detecting objects in its environment, • a control controller, • an on-board computer that transmits control parameters to the controller, • means for transmitting to a processing unit the data of objects detected in the robot's environment by the sensors, characterized in that said on-board computer adapts in real time the movement and tasks of the mobile robot according to: • data from a map updated by said processing unit and transmitted to the on-board computer, • traffic and operating rules related to the objects and areas of the map, and • a semantic hierarchy between these areas and / or objects,The mapping transmitted to the onboard computer is a high-definition map (HDMAP) which includes: • a level of metric data mapping the robot's movement space, • a level of rules for movement and operation within the movement space, • a level of semantic mapping allowing for the semantic categorization of areas and objects present in the robot's environment.

9. Navigation system for autonomous mobile robot(s) (AMR), comprising at least one robot according to claim 8,

10. the system comprising a processing unit which determines a high-definition map (HDMAP) of the robot's movement space, said high-definition map (HDMAP) comprising#: • a level of metric data mapping the robot's movement space, • a level of traffic and operating rules associated with this level of metric data, • a level of semantic mapping allowing for the semantic categorization of areas and objects present in the robot's environment, the system also including access to at least one library of general rules and at least one library of semantic data, the processing unit defining the level of traffic and operating rules as well as the level of semantic mapping by referring to the content of these libraries, The high-definition mapping (HDMAP) is updated by the processing unit with data from objects detected in the robot's environment by the sensors. The robot(s) adapt their movement and tasks in real time based on the updated mapping data, the traffic and operating rules associated with the objects and areas of the map, and a semantic hierarchy between these areas and / or objects. A computer program comprising code instructions for implementing a method according to any one of claims 1 to 7 when executed on the processor of an autonomous mobile robot.

Images