TRANSPORT ROBOT AS WELL AS METHOD AND SYSTEM FOR OPERATING SUCH A TRANSPORT ROBOT IN A WAREHOUSE

Description

[0001] The invention relates to a transport robot as well as a method and system for operating such a transport robot in a warehouse.

[0002] For the transport and storage of products, goods, and materials, load carriers such as wire mesh boxes or pallets, especially Euro pallets, are often used. To handle such load carriers, for example in intralogistics (i.e., the internal flow of materials within a company, such as in a warehouse), industrial trucks such as forklifts, transport robots, and the like are used. A warehouse is generally a very dynamic environment in which the control of mobile industrial trucks is usually either manual, semi-automated, or fully automated.

[0003] In intralogistics, strong growth combined with demographic change is creating a need for automated solutions. However, the introduction of new, complex digital technologies is hampered by the fact that highly qualified personnel are often underrepresented in the field of intralogistics.

[0004] Current automation solutions for transporting unit loads in intralogistics are primarily based on continuous conveyors, stacker cranes, and automated guided vehicles (AGVs). While continuous conveyors and stacker cranes are used wherever rigid process chains and high material throughput are required, AGVs offer more application possibilities in intralogistics.

[0005] The somewhat hesitant adoption of existing automation solutions is due not only to the high investment costs associated with them, but also to their increased space requirements and, often, to the infrastructure built for manual processes. Furthermore, automation solutions like AGVs operate on predefined transport routes. Any changes to the transport layout necessitate a complete redesign, resulting in reduced flexibility. Consequently, for numerous potential applications requiring greater flexibility, either no economically viable solutions are available, or the current technologies, due to their dynamic nature, incur high redesign costs. This leads to existing automation solutions occupying only a niche market compared to manually oriented transport solutions.Autonomous Mobile Robots (AMRs), such as transport robots, are characterized by greater flexibility compared to AGVs (Automated Guided Vehicles). AMRs, which move in a variety of open environments and perform a wide range of tasks and interactions, require explicit capabilities to fulfill their roles. AMRs interact with their environment via sensory-motor interaction and must act consciously to accomplish their tasks. In this context, conscious action means that AMRs perform actions motivated by given goals and justified, and potentially optimal, through environment-model-dependent considerations regarding these goals. These deliberative capabilities distinguish AMRs from AGVs.

[0006] For automated transport by AMRs, especially transport robots, the AMRs must be programmed with the knowledge necessary to create routes, navigation maps, and subsequent actions (such as receiving and delivering goods, waiting, loading, and the like). This programming requires intervention by highly specialized personnel to adapt the AMRs to their intended use. This constitutes the majority of the commissioning and adaptation efforts and, due to the associated costs and personnel resources, limits the available options.

[0007] DE 10 2021 006 476 A1 concerns the navigation of mobile logistics robots using an intelligent topological map without global metric map information. Nodes describe unique landmarks designed via an identification system (ArUco markers); the nodes are connected by edges that are stored with annotations (preferred path, distance, lane width, speed limits). Navigation between nodes is reactive, based on environmental perception.

[0008] Knowledge of the routes and corresponding work locations within a warehouse exists in the form of warehouse and application knowledge. Warehouse employees, in particular, possess a great deal of practical knowledge that could be readily available to the machine as a basis for task execution by the automated guided vehicles (AGVs).

[0009] Against this background, the present invention aims to provide an improved transport robot as well as a method and system for commissioning, adapting and operating such a transport robot in a warehouse.

[0010] This task is solved, according to one aspect, by a transport robot operating in a warehouse. The transport robot can be a material handling vehicle, particularly a forklift. The transport robot, which can be automated, semi-automated, and / or manually operated, is designed to carry out transport orders for moving goods within a warehouse containing multiple transport robots. The transport robot includes a communication interface, for example, a communication interface for wireless communication via a wireless communication network. This communication interface is configured to receive an environmental model of the warehouse and a capability model of the multiple transport robots operating within the warehouse.In one embodiment, the communication interface of the transport robot is configured to receive the environment model of the warehouse and the capability model of the multiple transport robots within the warehouse from a central device, for example, a server for operating the multiple transport robots. Furthermore, the transport robot can also receive transport orders from the central device. In another embodiment, the communication interface of the transport robot is configured to receive the environment model of the warehouse and the capability model of the multiple transport robots from another transport robot within the multiple transport robots.

[0011] According to the invention, the environment model of the warehouse defines a plurality of unique key locations (hereinafter also referred to as semantic key locations or nodes) of the warehouse in the form of a topological map, in which each key location has at least one neighboring key location, i.e., connected via an edge of the topological map. The capability model of the plurality of transport robots defines or comprises, for each of the key locations of the environment model, i.e., the topological map, motion sequence information, in particular at least one trajectory, for an automated movement from the corresponding key location to the at least one neighboring key location. In other words, the capability model with the motion sequence information, in particular trajectories, for each key location of the environment model enables the transport robot to connect neighboring key locations of the environment model with one another.For each key location of the plurality of key locations of the environment model, the motion information, in particular the at least one trajectory, of the capability model (which is linked to the corresponding key location) is based, according to the invention, on at least one movement learned by imitation learning of a transport robot of the plurality of transport robots, which is operated at least temporarily manually, i.e., by an employee, between the key location and the at least one adjacent key location. The transport robot further comprises a control unit, which is configured to carry out a transport order from a starting key location to a destination key location of the plurality of key locations, and to automatically control the movement of the transport robot based on the environment model and the capability model. According to the invention, the expert knowledge required for the automatic operation of the transport robot is thus..."Know-How" is conveyed to the transport robot as imitation knowledge in the form of the capability model.

[0012] According to one embodiment, the control unit for executing the transport order from the start key point to the destination key point of the plurality of key points can be configured to automatically control the movement of the transport robot, in particular a forklift truck, based on the environment model and first motion sequence information, in particular a first trajectory, and at least second motion sequence information, in particular at least one second trajectory, of the capability model, wherein the first motion sequence information, in particular the first trajectory, defines a movement from the start key point to an intermediate key point adjacent to the start key point, and wherein the at least second motion sequence information, in particular the at least one second trajectory,Define a movement from the intermediate key position to the target key position, or a movement from the intermediate key position to another intermediate key position adjacent to the intermediate key position.

[0013] In one embodiment, the control unit is further configured to determine, after the movement of the transport robot from the starting key point to the intermediate key point, which is based on the first motion sequence information, in particular the first trajectory, an accumulated deviation of an actual position or pose from a target position or pose of the transport robot relative to the intermediate key point, and to vary the movement of the transport robot from the intermediate key point to the target key point or the next intermediate key point, which is based on the second motion sequence information, in particular the second trajectory, in order to counteract the deviation.

[0014] According to one embodiment, the control unit can further be configured to determine the movement of the transport robot from the starting key point to the target key point via one or more intermediate key points using a cost function.

[0015] According to one embodiment, each key location of the plurality of key locations can have an artificial marking (for example, a marker), and the transport robot can further have a sensor unit which is configured to detect the marking of a respective key location and to identify the respective key location on the basis of the detected marking.

[0016] In one embodiment, the control unit of the transport robot is designed to determine the position, in particular the pose, of the transport robot relative to the respective key point based on the artificial marking of a respective key point detected by the sensor unit.

[0017] According to one embodiment, the marking of the respective key point can be a visual marking and the sensor unit of the transport robot can be designed as a camera, in particular a depth camera, for capturing the visual marking of the respective key point.

[0018] In one embodiment, the transport robot further comprises a drive unit configured to drive the movement of the transport robot based on motion control signals from the control unit. The drive unit can, for example, include a plurality of drive wheels and an electric drive motor for driving the plurality of drive wheels.

[0019] According to a second aspect, the aforementioned problem is solved by a system for operating a plurality of transport robots in a warehouse. This system comprises a plurality of transport robots according to the first aspect of the invention, as well as a central device for operating the plurality of transport robots in the warehouse, wherein the central device is configured to provide each transport robot of the plurality of transport robots with the environment model of the warehouse and the capability model of the plurality of transport robots.

[0020] According to a third aspect, the aforementioned task is solved by a method for operating a transport robot, in particular a forklift truck, which can be operated automatically, semi-automatically, and / or manually, to carry out transport orders for moving goods in a warehouse with a plurality of transport robots. The method comprises the following steps: obtaining an environment model of the warehouse and a capability model of the plurality of transport robots in the warehouse (optionally from a central device for operating the plurality of transport robots), wherein the environment model of the warehouse defines a plurality of key locations of the warehouse and, for each key location, at least one adjacent key location, and wherein the capability model provides motion sequence information, in particular at least one trajectory, for each of the key locations of the environment model.for an automated movement to the at least one adjacent key point, wherein for each key point of the plurality of key points of the environment model, the movement sequence information, in particular the at least one trajectory, of the capability model is based on at least one movement of a manually operated transport robot of the plurality of transport robots learned by imitation learning between the key point and the at least one adjacent key point; and controlling the movement of the industrial truck on the basis of the environment model and the capability model to carry out a transport order from a starting key point to a destination key point of the plurality of key points.

[0021] Further advantages and details of the invention are explained in more detail by way of example with reference to the embodiments shown in the schematic figures. These show: Figure 1a schematic representation of a transport robot in the form of a forklift truck for transporting goods in a warehouse according to one embodiment; Figure 2 a schematic representation of a system according to the invention with a plurality of transport robots in the form of industrial trucks and a central device for providing an environment model and a capability model for operating the plurality of transport robots; Figure 3a a representation of a transport robot according to an embodiment in an exemplary warehouse with a key point in the form of a branch; Figure 3b a representation of an environment model of the warehouse of Figure 3a ; Figure 3c a representation of a transport robot according to an embodiment in an exemplary warehouse with two adjacent key locations and the automated movement of the transport robot based on the capability model; Figure 3da representation of an environment model of the warehouse of Figure 3c ; Figure 3e a representation of the movement of a transport robot according to an embodiment based on the capability model from a starting key point via an intermediate key point to a destination key point; Figure 3f an automated correction of the movement of Figure 3e by a transport robot according to one embodiment; and Figure 4 A flowchart with steps of a method for operating a transport robot for transporting goods in a warehouse according to one embodiment.

[0022] Figure 1Figure 1 shows a schematic representation of a transport robot 120a designed as a forklift truck 120a according to an embodiment for transporting goods 140 in an industrial environment, in particular a warehouse. The transport robot 120a, designed as a forklift truck 120a, can in particular be a forklift truck 120a that is operated partially autonomously, semi-autonomously, and / or manually, i.e., guided by an operator. The goods 140 can, for example, be objects 143, such as packaging cartons 143, which are arranged on a respective load carrier 141. The load carrier 141 can, for example, be a pallet 141, in particular a Euro pallet 141, or a wire mesh container 141.

[0023] As in Figure 1As shown, the transport robot 120a, designed as a forklift truck, comprises a load-handling device in the form of a pair of load forks 124a,b, which are designed to be inserted into respective recesses, in particular pockets 141a,b, on an end face of the load carrier 141 in order to receive the load carrier 141 and the goods object 143 arranged on it. According to further embodiments, the load-handling device can also be designed as a mandrel, for example for receiving rolls of film or wire coils, as an under-hooking load-handling device (e.g., comparable to garbage trucks for receiving garbage cans), or as bale and roll clamps, for example for receiving paper rolls.

[0024] The in Figure 1The transport robot 120a, designed as a forklift truck 120a, further comprises a drive unit 121, for example at least one motor 121, in particular an electric motor 121, wherein the drive unit 121 is configured to move the forklift truck 120a and the pair of load forks 124a,b relative to the goods 140, for example to change the orientation and / or distance between the forklift truck 120a and the goods 140 and / or to raise or lower the pair of load forks 124a,b. For this purpose, as shown in Figure 1 As indicated, the drive unit 121 may be suitablely connected to wheels 122a-d and / or the pair of load forks 124a,b of the industrial truck 120a.

[0025] The transport robot 120a, configured as a forklift truck 120a, can further comprise a sensor unit 130a, in particular an image acquisition unit 130a, which is configured to capture a multitude of images of the forklift truck 120a's surroundings as the transport robot 120a moves. Preferably, the image acquisition unit 130a comprises a camera 130a, in particular a depth camera 130a. As described in Figure 1As indicated, the image acquisition unit 130a is preferably mounted on the transport robot 120a, which is designed as a forklift truck 120a, such that the field of view of the image acquisition unit 130a lies essentially along a forward direction of movement A of the transport robot 120a. Preferably, the image acquisition unit 130a can be mounted in the plane of symmetry between the two load forks 124a,b. In addition to the image acquisition unit 130a with the field of view along the forward direction of movement A of the transport robot 120a, the transport robot 120a, which is designed as a forklift truck 120a, can also include further image acquisition units or sensor units, for example, an image acquisition unit with a field of view along a backward direction of movement of the transport robot 120a and / or an image acquisition unit with a field of view perpendicular to the forward direction of movement A of the transport robot 120a.

[0026] According to the invention, the transport robot 120a, designed as a forklift truck 120a, further comprises a control unit 123, which may, for example, include one or more processors or microcontrollers with suitable software and is designed to control the transport robot 120a, designed as a forklift truck 120a, at least partially automatically, as described in detail below.

[0027] Figure 2 Figure 1 shows a schematic representation of a system 100 according to the invention for operating the transport robot 120a designed as a forklift truck 120a and at least one further transport robot 120b designed as a forklift truck 120b, which can each be operated temporarily automatically, semi-automatically and / or manually in order to carry out transport orders for the transport of goods 140 in a warehouse.

[0028] System 100 comprises, in addition to the multitude of transport robots 120a,b, a central management device 110, for example a server 110, which is configured to communicate with each of the multiple transport robots 120a,b and to assign transport orders to each of the multiple transport robots 120a,b, for example, for transporting goods 140 in the warehouse. For this purpose, for example, the transport robot 120a, configured as a forklift truck, includes a communication interface 126, which is configured to communicate with a corresponding communication interface 113 of the central management device 110 and with an external sensor unit 130b, such as a sensor unit 130b of the warehouse, for example via a wireless communication network 150, e.g. a Wi-Fi network or a mobile network. In the Figure 2The central management device 110 shown can be, for example, an industrial PC 110 or a cloud server, in particular an edge cloud server 110.

[0029] As in the Figure 2 As shown, the central management device 110 can, in addition to the aforementioned communication interface 113, comprise one or more processors 111 and a memory 115, in particular a non-volatile one. The memory 115 can be configured to store data and executable program code which, when executed by the processor 111 of the central management device 110, causes the processor 111 to perform the functions, operations, and procedures described below.

[0030] As further explained below with reference to the Figures 3a-fAs described in detail, the communication interface 126 of the transport robot 120a, which is designed as a forklift truck, is configured, as is an environment model of the warehouse (example environment models 200 of a warehouse are shown in the Figure 3b and 3d(shown) and to obtain or receive a capability model of the plurality of transport robots 120a,b in the warehouse. In one embodiment, the communication interface 126 of the transport robot 120a is configured to obtain the environment model 200 of the warehouse and the capability model of the plurality of transport robots 120a,b in the warehouse from the central management device 110. In another embodiment, the communication interface 126 of the transport robot 120a is configured to obtain the environment model 200 of the warehouse and the capability model of the plurality of transport robots 120a,b from another transport robot, e.g., the transport robot 120b, or from the plurality of transport robots.As described in detail below, the environment model 200 of the warehouse and, in particular, the capability model of the multitude of transport robots 120a,b are based on imitation knowledge collected in the warehouse, which can be used to advantage, in particular, for setting up a new transport robot in the warehouse.

[0031] In one embodiment, the central management device 110 is configured to generate the environment model 200 of the warehouse and the capability model of the plurality of transport robots 120a,b configured as industrial trucks, and to make these available to the transport robots 120a,b configured as industrial trucks. In another embodiment, the plurality of transport robots can generate the environment model of the warehouse and the capability model in a distributed manner.

[0032] As described in more detail below, the warehouse environment model 200 defines a multitude of unique key locations 210a-c (also referred to herein as semantic key locations 210a-c or nodes 210a-c) of the warehouse in the form of a topological map. Figure 3a and 3c Each shows a section of a warehouse with key locations and the Figure 3b and 3d Each shows the corresponding environment model 200. How this manifests itself in particular, the Figure 3d As can be seen, each key point 210a-h has at least one neighboring key point, i.e., one connected via an edge of the topological map. For example, in the Figure 3d In the exemplary environment model 200 shown, the key point 210a is connected to the key point 210b via an edge 215a.

[0033] As described in more detail below, the capability model of the plurality of transport robots 120a,b defines and includes, for each of the key locations 210a-h of the environment model 200, i.e., the topological map 200, motion sequence information, in particular at least one trajectory, for automated movement from the corresponding key location to the at least one adjacent key location or for handling pallets (automatic depositing or picking up). In other words, the capability model with the motion sequence information, in particular trajectories, for each key location of the environment model 200 enables the transport robot 120a to connect adjacent key locations of the environment model 200.

[0034] At the in Figure 3cIn the illustrated example, the transport robot 120a according to the invention moves automatically along a first trajectory 220a to a first key point provided with a first optical marker 230a. Subsequently, the transport robot 120a according to the invention moves automatically along a second trajectory 220b, making a right turn from the first key point to a second key point provided with a second optical marker 230b, in order to continue its journey automatically along a third trajectory 220b. As the person skilled in the art will recognize, in the Figure 3cIn the example shown, for instance, the second trajectory 220b, which is part of the capability model, is linked to the first key point defined by the environment model, which is marked with the first marker 230a. Of course, the first key point defined by the environment model can be linked to further trajectories or movement sequence information from the capability model, for example, a trajectory which in the example of Figure 3c describes a left turn.

[0035] For each key location of the plurality of key locations 210a-h of the environment model 200, the motion information, in particular the at least one trajectory, of the capability model (which is linked to the corresponding key location) is based, according to the invention, on at least one movement learned by imitation learning of a transport robot of the plurality of transport robots, operated at least temporarily manually, i.e., by an employee, between the key location and the at least one adjacent key location. In other words, during the manual operation of the transport robots in a warehouse, motion sequence information, in particular trajectories, can be recorded and collected in order to generate or extend / update the capability model.

[0036] As a person skilled in the art will recognize, the environment model 200 and the capability model thus define an annotated topological map of the warehouse with a multitude of interconnected key locations 210a-h, wherein the capability model links motion sequence information, for example, at least one trajectory 220a-c, for movement to at least one adjacent key location to each key location 210a-h. This motion sequence information 220a-c of the capability model can, as already described, define or include a trajectory to describe the necessary movement commands for reaching the neighboring key locations, but also other types of motion sequence information that depict more complex movements, for example, for placing or picking up pallets, loading the transport robot, and the like.

[0037] As described above, the transport robot 120a comprises a control unit 123, which is designed to carry out a transport order from a starting key point to a destination key point of the plurality of key points 210a-h, and to automatically control the movement of the transport robot 120a based on the environment model 200 and the capability model. According to the invention, the knowledge or "know-how" required for the automatic operation of the transport robot 120a is thus provided to the transport robot 120a in the form of the environment model 200 and as imitation knowledge in the form of the capability model. The transport robot 120a, designed as a forklift truck 120a according to the invention, and the system 100 according to the invention thus provide a representation of automated goods transport by imitation. The necessary location and action information is obtained by observing previously executed movements of a manually (i.e.,A transport robot 120b (which may be identical or similar in construction to transport robot 120a) controlled by a person with expert knowledge (e.g., warehouse employee) and trained as a forklift truck is generated concurrently so that it can be used to generate or update the capability model. In this context, concurrency describes an exploration phase of the multitude of transport robots 120a,b trained as forklift trucks, which automatically learn all movements performed and their spatial references within the warehouse by observing the application environment during routine work.

[0038] According to the invention, imitation knowledge in the form of environmental knowledge can thus be automatically learned during the exploration phase by observing manual transport operations with the transport robot 120a,b, which is configured as a forklift. During the demonstration, as described above, the environmental model 200 of the warehouse is generated in the form of a topological map 200, which is based on a multitude of semantic key locations 210a-h of the warehouse. According to one embodiment, the semantic key locations 210a-h can be considered as locations for the transport of goods 140 and fundamental decisions within the warehouse in order to achieve the desired goal of automated transport orders. In addition to the information contained in the environmental model 200, the capability model, which serves to link the previously observed semantic key locations 210a-h, can also be learned, as described above.

[0039] The imitation knowledge can be linked to concrete transport orders in a subsequent exploitation phase using algorithm-based action planning, enabling the previously demonstrated manual transports to be executed automatically. Potential disruptions in the execution of the learned movements due to inherent system errors or dynamic obstacles can be compensated for, for example, by an adapted state-lattice path planner. The localization required for executing automatic movements in space can be based on a proximity-based localization approach. This approach utilizes the linking of nodes 210a-h of the environment model 200 (i.e., the key points 210a-h) via the capability model to meet metrically continuous localization requirements, thus eliminating the need for a singular reference point, such as a coordinate origin in metric maps.

[0040] According to one embodiment, each key location of the plurality of key locations 210a-h can have an artificial marking 230a,b, as already mentioned above in connection with the Figure 3a and 3c As described above, each marking 230a,b is designed such that it can be detected by the sensor unit 130a, in particular the camera 130a of the transport robot 120a designed as a forklift truck, in order to identify the respective key point based on the detected marking 230a,b. According to one embodiment, the sensor unit 130a can include, in addition to a camera, in particular a depth camera, an infrared light source for illuminating the field of view of the depth camera.

[0041] In one embodiment, the control unit 123 of the transport robot 120a designed as a forklift truck is configured to determine the position or pose 240a-c of the transport robot 120a designed as a forklift truck relative to the respective key point 210a-c on the basis of the marking 230a,b detected by the sensor unit 130a at a respective key point.

[0042] Figure 3e Figure 1 shows a representation of the movement of the transport robot 120a according to an embodiment based on the capability model from a position at a start key point 210a via a position at an intermediate key point 210b to a position at a target key point 210c of the environment model 200. Figure 3f shows an automated correction of the movement of Figure 3e by a transport robot 120a according to an embodiment as described in more detail below.

[0043] At the in Figure 3eIn the example shown, the control unit 123 of the transport robot 120a, designed as a forklift truck, is designed to carry out the transport order from the start key point 210a to the target key point 210c of the plurality of key points 210a-c, to automatically control the movement of the transport robot 120a, designed as a forklift truck, on the basis of the environment model 200 and first motion sequence information 220a, in particular a first trajectory 220a, and at least second motion sequence information 220b, in particular at least a second trajectory 220b, of the capability model.The first motion sequence information 220a, in particular the first trajectory 220a, defines a movement from the start key point 210a to an intermediate key point 210b adjacent to the start key point 210a, and the at least second motion sequence information 220b, in particular the at least one second trajectory 220b, defines a movement from the intermediate key point 210b to the destination key point 210c. According to one embodiment, the control unit 123 of the transport robot 120 can further be configured to determine, by means of a cost function, the movement of the transport robot 120a, designed as a forklift truck, from the start key point 210a to the destination key point 210c via the one or more intermediate key points 210b.For example, the control unit 123 can determine a path from the start key point 210a to the destination key point 210 via one or more intermediate key points 210b based on the cost function, which minimizes a travel time or a travel distance of the transport robot 120a.

[0044] How this manifests itself in particular in the Figure 3fAs can be seen, in one embodiment the control unit 123 of the transport robot 120a, designed as a forklift truck, is further configured to determine, after the movement of the transport robot 120a from the starting key point 210a to the intermediate key point 210b, which is based on the first motion sequence information 220a, in particular the first trajectory 220a, of the capability model, an accumulated deviation of an actual position or actual pose 240b from a target position or target pose 240b' of the transport robot 120a relative to the intermediate key point 210b. As this is shown in Figure 3fAs indicated and described in more detail below, the control unit 123 of the transport robot 120a is designed according to an embodiment to vary the movement of the transport robot 120a from the intermediate key point 210b to the target key point 210c, which is based on the target trajectory 220b, in order to counteract the deviation accumulated at the intermediate key point 210b, which can be considered as integral error propagation.

[0045] According to one embodiment, the control unit 123 of the transport robot 120a is designed to be located in Figure 3fThe integral error propagation shown during movement to the intermediate key point 210b is counteracted by correcting the localization (proximity-based) at the intermediate key point 210b. For this purpose, in addition to the movement sequence information 220a, in particular the trajectory 220a for the movement to the intermediate key point 210b, the position of the intermediate key point 210b relative to the starting key point 210a can also be stored in the environment model 200 (for example, in the form of metadata) and used for error correction.

[0046] For the realization of mission-specific metric localization, the position of the starting key point 210a is used as a reference point according to one embodiment. As soon as expected nodes of the environment model 200 appear during the further course of the automated movement of the transport robot 120a, for example the intermediate key point 210b, these can be used as a correction parameter due to their known, linked, geometric relationship to the starting key point 210a. According to one embodiment, all relevant geometric relationships of the key points 220a-c can be taken from the topological map 200, i.e., from the environment model 200.

[0047] As described above, the starting point is in Figure 3f The illustrated example shows the movement of the transport robot 120a near the starting key point 210a (in Figure 3f(also referred to as node A), which is used as a reference point, and runs via the intermediate key point 210b (in Figure 3f (also referred to as node D) to the target key point 210c (in Figure 3c(also referred to as node N). Thus, the transport robot 120a starts its planned first trajectory 220a (i.e., the first target trajectory 220a), which is part of the capability model, at reference point A. Assuming that the tracking of the first trajectory 220a, which maps the connection between the starting key point 220a (point A) and the intermediate key point 220b (point D), has no offset at the beginning, the first actual trajectory 220a' starts with the same pose as the planned first trajectory 220a. Since the localization of the semantic key point may be spatially limited, according to one embodiment, the localization in the further tracking of the movement of the transport robot 120a uses odometry signals from the transport robot 120a. The odometry signals can generate a random error during the movement, so that the Figure 3fThe depicted deviation between the target trajectory 220a and the actual trajectory 220a' is shown.

[0048] According to one embodiment, successful localization relies on the fact that the cumulative error (referred to herein as ek) at the end of the movement along a trajectory of the capability model allows sensory detection of the next node. The in Figure 3fThe marked ellipses 225 and 225c define this target area, within which sensor localization is possible, for the intermediate key point 210b and the target key point 210c, respectively. Upon reaching the target area 250b at the end of the movement from the starting key point 210a to the intermediate key point 210b, the intermediate key point 210b is used as the new reference point. The actual error can be calculated by the transport robot 120a detecting the intermediate key point 210b. The error between the planned target trajectory 220a and the actual trajectory 220b is the difference between the expected pose and the measured pose, according to the following equation: D e d = x <mprescripts / > <none / > D d − x <mprescripts / > <none / > D d ′ mit x <mprescripts / > <none / > D d = D A T − 1 ⋅ x <mprescripts / > <none / > A d

[0049] According to one embodiment, the error ed thus determined at the intermediate key point 210b at time d can be compensated for in order to readjust the movement of the transport robot 120a to a planned target movement. Due to the non-holonomic properties of many transport robots, according to one embodiment, the compensation of the determined error is carried out by a compensation trajectory with an individual path length. This path required for this purpose is in Figure 3f The dashed portion of the actual trajectory 220b' is shown. Thus, the error accumulated up to that point can be completely compensated for after it is detected by the control unit 123 of the transport robot 120a.

[0050] During the further movement of the transport robot 120a from the intermediate key point 210b to the target key point 210c, another error can occur (especially due to faulty odometry signals) as soon as the transport robot 120a leaves the sensory node detection area of ​​the intermediate key point 210b. Until the sensory detection area of ​​the target key point 210c is reached, the localization error, as with the movement from the starting key point 210a to the intermediate key point 210b, can now also increase during the current movement to the target key point 210c. As soon as the transport robot 120a enters the sensor detection area 225c of the target key point 210c, the error e can be calculated again.For the overall movement from the starting key point 210a to the target key point 210c, this means that the previously calculated error corrections enable the reaching of the subsequent key point, provided that the transport robot 120a does not exceed the maximum permissible error ellipse 225 (in . Figure 3f (referred to as emax) does not leave.

[0051] The generally derived calculation rule for determining the error ek at the key point c at time k results from the chaining of the nodes in the topological map 200 according to the following equation: c e k = x <mprescripts / > <none / > c k − x <mprescripts / > <none / > c k ′ mit x <mprescripts / > <none / > c k = ∏ i = S c i + 1 i T − 1 ⋅ x <mprescripts / > <none / > S k

[0052] In this context, the index S in the previously presented equation corresponds to the starting point of the transport robot's movement at the starting key point 210a, while c represents the next expected key point. The notation Sx k corresponds to the current odometry-supported, error-corrected pose of the transport robot 120a at the discrete time k in the starting coordinate system.

[0053] As those skilled in the art will recognize, according to the invention, the number of mission nodes, i.e., key points, is irrelevant for successful localization. Only the relative pose between adjacent nodes is necessary for correcting the localization. Furthermore, the nodes must be detectable by the transport robot 120a in order to calculate their relative pose.

[0054] The representation of the environment of a warehouse according to the invention as a topological map or environment model 200 initially has the advantage that the location information can be stored as a discrete planning network. This discrete planning network represents, via its edges, all connection possibilities that the transport robot 120a can execute. Furthermore, the topological map can combine the environment model 200 and the (learned) capabilities for linking adjacent nodes. The topological maps therefore fundamentally extend the geometric-semantic approach of previous map models by integrating a capability model. The capability model is used as a metric connection between the nodes of the environment model 200.The connection of multiple nodes thus creates a kinematic chain that interrupts the accumulating unknown localization error at the node points (semantic key points), since the positions of the sensorily detected nodes are environment-invariant. Two essential deductions initially result from this observation.

[0055] Integral localization errors caused by external influences such as slippage, faulty sensors, digital losses, etc., can be limited in their impact at the sensor-detected nodes. For this reason, the annotated topological maps described here minimize the risk of an invalid community map, provided it is automatically learned and extended by a fleet of transport robots. The map itself can therefore be easily replaced.

[0056] The mappable geometric length of the trajectories of the capability model for connecting the nodes depends only on the sensors used by the transport robot and can also be extended by sensor post-processing steps (e.g. visual odometry).

[0057] Another possible design allows for the limitation of the incremental error, which increases with distance from the traversed intersection, by using relative supporting localization, e.g., by employing visual odometry methods or road markings, curbs, and the like. This allows the distance between the support points in a warehouse to be increased.

[0058] A further advantage arises from mapping logical information of the visually perceived nodes, provided these nodes have a direct mapping relationship to the inventory management system. In this case, the inventory management system can work directly with the node information, thus eliminating the need for the complex conversion of logical information from the inventory management system to geometric information for the transport robots 120a,b.

[0059] Figure 4Figure 400 shows a flowchart with steps of a method 400 for operating a transport robot 120a configured as a forklift truck to carry out transport orders for moving goods 140 in a warehouse containing a plurality of transport robots 120a,b configured as forklift trucks. The method 400 includes a step 401 of receiving an environment model 200 of the warehouse and a capability model of the plurality of forklift trucks 120a,b in the warehouse. According to one embodiment, the transport robot 120a can receive the environment model 200 of the warehouse and / or the capability model of the plurality of forklift trucks 120a,b from the central control device 110 and / or from another forklift truck in the plurality, for example, the transport robot 120b configured as a forklift truck.

[0060] As already described in detail above, the environment model 200 of the warehouse defines a plurality of semantic key locations 210a-h of the warehouse and for each key location of the plurality of key locations 210a-h at least one neighboring key location and the capability model defines for each of the plurality of key locations 210a-h of the environment model movement sequence information, in particular at least one trajectory 220a-c, for an at least semi-automated movement to the at least one neighboring key location.For each key point of the plurality of key points 210a-c of the environment model 200, the motion sequence information 220a-c, in particular the at least one trajectory 220a-c, of the capability model is based according to the invention on at least one movement learned by imitation learning of an at least temporarily manually operated industrial truck 120b of the plurality of transport robots 120a,b designed as industrial trucks between the key point and the at least one adjacent key point of the plurality of key points 210a-c.

[0061] Furthermore, the procedure 400 includes a step 403 of controlling the movement of the transport robot 120a, configured as a forklift truck, based on the environment model 200 and the capability model, in order to carry out a transport order from a starting key point 210a to a destination key point 210c of the plurality of key points 210a-h. As already described above, this movement from the starting key point to the destination key point can take place via one or more intermediate key points.

[0062] The method 400 according to the invention can be carried out using the transport robot 120a according to the invention, which is designed in the form of a forklift truck. Therefore, further embodiments of the method 400 according to the invention result from the embodiments of the transport robot 120a according to the invention described above.

Claims

1. Transport robot (120a) which is operable so as to carry out transport orders for transporting goods (140) in a warehouse with a multiplicity of transport robots (120a,b), wherein the transport robot (120a) comprises: a communication interface (126) which is designed to receive a surroundings model (200) of the warehouse and a capability model of the multiplicity of transport robots (120a,b) in the warehouse, wherein the surroundings model (200) of the warehouse defines a multiplicity of key locations (210a-c) of the warehouse and, for each key location (210a-c), at least one adjacent key location, and wherein, for each of the key locations (210a-c) of the surroundings model (200), the capability model defines items of movement sequence information (220a-c) as at least one trajectory (220a-c) for automated movement to the at least one adjacent key location, wherein, for each key location of the multiplicity of key locations (210a-c) of the surroundings model (200), the at least one trajectory (220a-c) of the capability model is based on at least one movement learned by imitation learning of a manually operated transport robot (120b) of the multiplicity of transport robots (120a,b) between the key location and the at least one adjacent key location, wherein the trajectories (220a-c) are recorded and collected during manual operation of the transport robots (120a,b) in the warehouse in order to generate the capability model, wherein the capability model, by way of the items of movement sequence information (220a-c), in particular trajectories (220a-c), for each key location (210a-c) of the surroundings model (200), enables the transport robot (120a) to connect adjacent key locations (210a-c) of the surroundings model to one another; and a control unit (123) which, for carrying out a transport order from a starting key location (210a) to a target key location (210c) of the multiplicity of key locations (210a-c), is designed to control the movement of the transport robot (120a) on the basis of the surroundings model (200) and the capability model.

2. Transport robot (120a) according to Claim 1, wherein the control unit (123), for carrying out the transport order from the starting key location (210a) to the target key location (210c) of the multiplicity of key locations (210a-c), is designed to control the movement of the transport robot (120a) on the basis of the surroundings model (200) and first items of movement sequence information (220a) as a first trajectory (220a) and at least second items of movement sequence information (220b) as a second trajectory (220b) of the capability model, wherein the first trajectory (220a) defines a movement from the starting key location (210a) to an intermediate key location (210b) adjacent to the starting key location, and wherein the at least second trajectory (220b) defines a movement from the intermediate key location (210b) to the target key location (210c) or a movement from the intermediate key location (210b) to a further intermediate key location adjacent to the intermediate key location (210b).

3. Transport robot (120a) according to Claim 2, wherein the control unit (123) is further designed, after the movement of the transport robot (120a) from the starting key location (210a) to the intermediate key location (210b) on the basis of the first trajectory (220a), to determine a deviation in an actual position or actual pose (240b) from a desired position or desired pose (240b') of the transport robot (120a) relative to the intermediate key location (210b) and to vary the movement of the transport robot (120a) from the intermediate key location (210b) to the target key location (210c) or the further intermediate key location on the basis of the second trajectory (220b) in order to counteract the deviation in the actual position or actual pose (240b) from the desired position or desired pose (240b').

4. Transport robot (120a) according to Claim 2 or 3, wherein the control unit (123) is further designed to determine the movement of the transport robot (120a) from the starting key location (210a) to the target key location (210c) via the one or more intermediate key locations (210b) situated therebetween using a cost function.

5. Transport robot (120a) according to any of the preceding claims, wherein each key location of the multiplicity of key locations (210a-c) has a marker (230a,b) and wherein the transport robot (120a) further has a sensor unit (130a) which is designed to detect the marker (230a,b) of a respective key location (210a-c) and identify the respective key location (210a-c) on the basis of the detected marker (230a,b).

6. Transport robot (120a) according to Claim 5, wherein the control unit (123) is designed to determine the position or pose (240b,c) of the transport robot (120a) relative to the respective key location (210a-c) on the basis of the marker (230a,b) of a respective key location (210a-c) detected by the sensor unit (130a).

7. Transport robot (120a) according to Claim 6, wherein the marker (230a,b) is a visual marker (230a,b) and wherein the sensor unit (130a) comprises a camera (130a), in particular a depth camera (130a), for detecting the visual marker (230a,b).

8. Transport robot (120a) according to any of the preceding claims, wherein the transport robot (120a) further comprises a drive unit (121) which is designed to drive the movement of the transport robot (120a) on the basis of movement control signals from the control unit (123).

9. Transport robot (120a) according to any of the preceding claims, wherein the transport robot (120a) is designed as an industrial truck (120a), in particular a forklift truck (120a).

10. System (100) for operating a multiplicity of transport robots (120a,b) in a warehouse, wherein the system (100) comprises: a multiplicity of transport robots (120a,b) according to any of the preceding claims; and a central device (110) for operating the multiplicity of transport robots (120a,b) in the warehouse, wherein the central device (110) is designed to provide the surroundings model (200) of the warehouse and the capability model of the multiplicity of transport robots (120a,b) to a transport robot of the multiplicity of transport robots (120a,b).

11. Method (400) for operating a transport robot (120a) which is operable so as to carry out transport orders for transporting goods (140) in a warehouse with a multiplicity of transport robots (120a,b), wherein the method (400) comprises: receiving (401) a surroundings model (200) of the warehouse and a capability model of the multiplicity of transport robots (120a,b) in the warehouse, wherein the surroundings model (200) of the warehouse defines a multiplicity of key locations (210a-c) of the warehouse and, for each key location (210a-c), at least one adjacent key location, and wherein, for each of the key locations (210a-c) of the surroundings model, the capability model defines items of movement sequence information (220a-c) as at least one trajectory (220a-c) for automated movement to the at least one adjacent key location, wherein, for each key location of the multiplicity of key locations (210a-c) of the surroundings model (200), the at least one trajectory (220a-c) of the capability model is based on at least one movement learned by imitation learning of a manually operated transport robot (120b) of the multiplicity of transport robots (120a,b) between the key location and the at least one adjacent key location, wherein the trajectories (220a-c) are recorded and collected during manual operation of the transport robots (120a,b) in the warehouse in order to generate the capability model, wherein the capability model, by way of the items of movement sequence information (220a-c), in particular trajectories (220a-c), for each key location (210a-c) of the surroundings model (200), enables the transport robot (120a) to connect adjacent key locations (210a-c) of the surroundings model to one another; and controlling (403) the movement of the transport robot (120a) on the basis of the surroundings model and the capability model in order to carry out a transport order from a starting key location (210a) to a target key location (210c) of the multiplicity of key locations (210a-c).