Methods for determining the position of industrial trucks

Abstract

10. A method for determining the position of industrial trucks in a defined storage environment, comprising the following steps: • Recording odometry data on a change in vehicle pose, • Determining a successor pose using a filter unit that calculates a successor pose from an initial pose using the acquired odometry data, whereby • a prediction step is performed to determine the subsequent pose, taking into account the odometry data and • in addition, a correction step is performed in which the subsequent pose is corrected towards an orientation of zero or a multiple of π / 2 using a virtual measurement function.

Description

[0001] The present invention relates to a method for determining the position of industrial trucks in a defined warehouse environment. The industrial trucks can be manually controlled vehicles operated by a driver or automated vehicles, such as AGVs (automated guided vehicles).

[0002] Vehicle positioning forms the basis for a multitude of applications in logistics, such as tracking routes, navigating to a destination, locating specific vehicle events like collisions with other vehicles or warehouse equipment, or even relocating parked vehicles. Depending on the application and use case, the position data is required and processed with varying degrees of spatial accuracy. For example, locating a parked vehicle does not require the same level of accuracy as navigating to a specific destination.

[0003] Numerous localization systems exist for industrial trucks, enabling them to determine their position in various environments. For example, it is common practice to install additional infrastructure in warehouses, such as ultra-wideband (UWB) for short-range, radio-based communication technology to ensure fast and stable data transmission for indoor and outdoor positioning. Bluetooth transmitters and receivers can also be used. Another approach involves equipping the vehicle with additional sensors, such as cameras, laser scanners, or ultrasonic sensors, to locate it within its surroundings.Systems requiring additional infrastructure also include those that use existing infrastructure to roughly locate a vehicle, for example, via a magnetic fingerprint. This method detects local variations in the Earth's magnetic field and evaluates them to determine the vehicle's position. Another method uses odometry data, which determines the distance traveled by the number of wheel rotations and the direction of travel by the steering angle. Odometry data is always acquired from a reference position. Due to the inherent inaccuracy of odometry data, it can only be used for short periods or distances, for example, to predict a vehicle pose until a new, more accurate measurement is available, or to avoid ambiguities in vehicle poses.

[0004] The term "vehicle pose" refers to both the position and the orientation of the vehicle. The vehicle's position within a storage area can be specified, for example, by X and Y coordinates. The vehicle's orientation can be specified by the angle of its longitudinal axis. Absolutely determining a vehicle's pose requires a reference coordinate system and a reference orientation. However, for processing odometry data, such an absolute reference position is not necessary; it suffices to specify the degree by which the orientation of the vehicle's longitudinal axis has changed relative to its previous orientation, just as it is sufficient to specify the spatial change in the X and Y coordinates.

[0005] A disadvantage of known methods for determining position is that the methods either require additional infrastructure in the environment and / or additional sensors on the vehicle, or have a very limited spatial resolution that only lasts for a short time.

[0006] The invention is based on the objective of providing a method for determining the position of a forklift truck that allows improved position determination using the simplest possible means.

[0007] According to the invention, the problem is solved by a method with the features of claim 1. Advantageous embodiments are the subject matter of the dependent claims.

[0008] According to the invention, the problem is solved by a method with the features of claim 1. The method is designed and intended for determining the position of industrial trucks in a defined storage environment. A defined storage environment has, in particular, a spatial structure that influences the position of the vehicles and their movement within the storage environment. The method provides that odometry data relating to a change in the vehicle pose are acquired. This means that spatial movement of the vehicle and a change in vehicle orientation are detected. Furthermore, a subsequent pose is determined using a filter unit, wherein the filter unit calculates a subsequent pose from an initial pose using the acquired odometry data relating to the changes. A prediction step and a correction step are performed to determine the subsequent pose. The prediction step is executed taking the odometry data into account.The correction step is performed by using a virtual measurement function to correct the subsequent pose towards an orientation of zero or a multiple of π / 2. During the calculation, the filter unit determines an estimated value for the new vehicle position based on the detected change in the vehicle's position. This is done by successively determining the subsequent pose from the initial pose. In the correction step, the virtual measurement function applies a weighting, distinguishing between stronger and weaker weighting. The subsequent pose is weighted more heavily if the orientation of the forklift truck remains unchanged. It is also weighted more heavily if the orientation of the forklift truck changes by a multiple of π / 2, i.e., 90°.A different change in the orientation of the subsequent pose is permitted by the filter unit with a lower weight, i.e., it is weighted less. This weighting is performed via the virtual measurement function, which is referred to as virtual because it does not rely on measured values ​​but rather weights the changes in the orientation of the subsequent pose. According to the invention, the filter unit is designed to preferentially permit straight-line travel or a right-angle change in pose. This takes into account the fact that the defined storage environment typically has a structure for the industrial truck that allows right-angle movements within the storage area. By giving greater weight to, and thus preferential treatment for, changes in pose based on right-angle movements, the filter unit achieves significantly improved accuracy over a longer period when evaluating the odometry data.

[0009] In a preferred embodiment of the method according to the invention, a Kalman filter is provided as the filter unit. The Kalman filter allows the estimated values ​​for the subsequent pose, for example Δx, Δy, and Δθ, to be estimated more accurately and reliably by incorporating correlations. With the Kalman filter, not only the initial pose of the industrial truck is considered in determining the subsequent pose, but previous vehicle poses are also taken into account with a certain statistical weight. The Kalman filter produces only one subsequent pose, which is also updated taking the odometry data into account. In a subsequent correction step, the orientation is corrected towards a multiple of Pi / 2 using a virtual measurement function, where the multiple also includes the value zero.

[0010] Alternatively or additionally, a particle filter can be used for the filter unit. Particle filters are estimation methods used to estimate the position and velocity of an object based on inaccurate and erroneous measurements. Particle filters are sometimes also referred to as sequential Monte Carlo (SMC) methods. The particle filter uses a large number of independent particles, each representing a possible vehicle pose, to estimate the vehicle's position. In the prediction step, the individual particles are projected based on the measured odometry data and an assumed level of process noise. In a subsequent correction step, the individual particles are weighted using a virtual measurement function, with particles with an orientation of zero or a multiple of 90° being weighted more heavily than particles whose orientation deviates from this.In a resampling step, which is performed under certain conditions (e.g., after each step or when many particles have a weight close to zero, i.e., fall below a predetermined minimum weight), low-weight particles are deleted and replaced by duplicates of high-weight particles. A first predetermined threshold can be defined for deleting low-weight particles; below this threshold, the particle is deleted. A second threshold can also be defined, specifying the minimum weight required to generate new particles. In relation to the particle filter, the weighting here means that the individual particle poses are not changed. Instead, the particle cloud is corrected towards the preferred orientations through appropriate particle weighting and the resampling step.The Kalman filter has only one subsequent pose, which is also updated taking the odometry data into account. The weight of a particle indicates its probability of being "reproduced." Thus, the particles themselves are not changed; however, in the resampling step, particles with very low weights are deleted and replaced by particles with higher weights. In a subsequent step, the orientation is corrected towards a multiple of π / 2 (or zero) using the virtual measurement function.

[0011] Both methods have in common that the successor positions are corrected taking into account the preferred orientations, resulting in a weighting that causes the corrected successor positions to be preferentially accepted.

[0012] In a preferred embodiment of the inventive method, the correction of the subsequent poses is carried out via the virtual measurement function, which can also be viewed as a cost function. The virtual measurement function depends on the change in orientation in the acquired odometry data. Such cost functions weight the estimation process for the subsequent pose and can thus be easily implemented in existing filter units. When calculating the average, for example in a particle filter, the cost function can weight the contributions of the particles to the average.

[0013] The method according to the invention further provides that a reference pose is determined. This is done in the storage environment, for example, by means of one or more locally defined reference markers. The industrial truck can determine its position and orientation at the reference markers, thus providing a precise position for further positioning. RFID markers, UWB markers, or optically readable markers, for example, can be used as reference markers. These markers are permanently installed in the storage environment and contain information that enables the industrial truck to determine its pose.

[0014] In a preferred embodiment, it has been found that the filter unit can omit the preference given to the successor pose when the industrial truck is turning. The turning maneuver is preferably detected by the steering angle sensor. Upon detection of a turning maneuver, the filter unit estimates a new vehicle pose from the current vehicle pose without using any weighting.

[0015] In a preferred embodiment, the odometry data is acquired via the wheel speed and the steering angle position. The steering angle position can be acquired either at the steered wheel(s) or at the steering angle sensor. A distinction must be made here as to whether the industrial truck has a single steered wheel or multiple steered wheels.

[0016] To improve the data, it can be implemented to combine a sequence of subsequent positions and smooth the trajectory formed by the subsequent poses using pose graph optimization. When smoothing the trajectory, it can be assumed that the forklift moves essentially smoothly and steadily, without excessive steering angle deflections.

[0017] The method according to the invention is explained in more detail below with reference to a series of figures. Fig. Figure 1 shows a top view of a warehouse with a large number of storage racks and industrial trucks moving within the warehouse. Fig. Figure 2 shows a weight function for a filter unit. Fig. Figures 3 a, b show the comparison between laser localization and odometry on the one hand and laser localization and odometry with a restriction to 90° on the other.

[0018] Fig. Figure 1 shows a schematic view of warehouse 10, which has a rectangular footprint. Shelves 12a to d are arranged parallel to each other in warehouse 10. A second group of shelves 14a to d is positioned at a right angle to shelves 12. In the depicted warehouse, shelves 12 and 14 are aligned parallel to a wall of the warehouse. This results in an overall topology characterized by right angles relative to each other. Forklifts 16a to e move through the warehouse carrying their transport orders. The future travel path of each vehicle is shown as a dashed line. It can be seen that some vehicles, such as 16b, 16c, and 16e, are traveling parallel to a wall of the warehouse, while vehicles 16a and 16d have not yet turned but are about to turn, although they may already have their steering angle indicator engaged.

[0019] The method according to the invention assumes that, starting from a current vehicle pose, the subsequent vehicle pose is estimated via a filter unit. For this estimation process, the in Fig. The weighting of the particles shown in Figure 2 is used. With a particulate filter, the next vehicle pose is weighted based on a change in vehicle orientation. The weighting function shown, Figure 18, as an example of a virtual measurement function, shows that a new vehicle pose receives a high weight if the angular change in orientation is 0°, 90°, 180°, or 270°. The plotted curves 18 and 20 show different tolerance ranges with regard to the angular deviation. Thus, the weighting position assigns a low weight to a new vehicle pose if it results in the vehicle not maintaining its orientation or changing direction at a right angle. For example, consider the situation where a particle of the particulate filter calculates a subsequent pose with an orientation of 45°: This would result in a diagonal movement through bearing 10.Such a change in orientation is quite unlikely due to the topology of the bearing. This is taken into account in the invention by – as in . Fig. 2 shown - the particle becomes light with this vehicle pose and its occurrence is therefore unlikely.

[0020] Due to this low weighting, it is advisable that the industrial trucks 16a and 16d be made up of Fig. 1. The filter unit operating according to the invention is switched off for the duration of the turn. Thus, weighting of the subsequent pose is omitted during the turn until the turn is completed. The advantage of this is that a turn can be reliably detected by a steering angle sensor input, and therefore a change of orientation by 90° or maintaining the orientation is reliably prevented.

[0021] In Fig. Figure 3a shows the dashed line representing the pure use of odometry data, without additional weighting or filtering. The solid black lines are based on laser localization, which determines the position of the forklift truck in the warehouse. For the dashed lines, the subsequent pose is estimated based on the acquired odometry data, and it is clearly visible that in some areas of the warehouse, the odometric position determination deviates significantly from the laser localization. In the areas shown in the Fig.In the dashed lines shown in Figure 3b, the estimated localization results that do not correspond to a straight-ahead journey or a right-angled change are weighted less according to the invention. It is clearly evident that when weighting the right-angled changes in orientation, the estimated curve corresponds much better to the reference curve (solid line) determined by laser localization. The reference curve was determined using a laser scanner and describes the actual trajectory of the industrial truck in the warehouse. It is also possible to precisely define at least one pose as the reference pose and then select this as the starting point for the odometric localization.

Claims

[1] Method for determining the position of industrial trucks in a defined storage environment, comprising the following steps: • Recording odometry data on a change in vehicle pose, • Determining a successor pose using a filter unit that calculates a successor pose from an initial pose using the acquired odometry data, characterized by , that • a prediction step is performed to determine the subsequent pose, taking into account the odometry data and • in addition, a correction step is performed in which the subsequent pose is corrected towards an orientation of zero or a multiple of π / 2 using a virtual measurement function. [2] Method for determining position according to claim 1, characterized by that a Kalman filter is provided as the filter unit. [3] Method for determining position according to claim 1, characterized by, that a particle filter is provided as a filter unit, which, with the prediction step, updates the vehicle pose of the individual particles based on the measured odometry data and, with the correction step, weights the individual particles via the virtual measurement function, whereby particles with an orientation of zero or a multiple thereof are weighted more heavily than particles whose orientation deviates from this. [4] Method according to claim 3, characterized by, that an additional resampling step is provided, which is carried out under one or more of the following conditions: after each measurement step, if a predetermined number of particles or a predetermined proportion of all particles has a weight below a predetermined minimum weight, wherein during the resampling step particles with a weight below a specified first limit are deleted and replaced by particles with a weight above a second limit. [5] Method for determining position according to any of the preceding claims, characterized by that a reference pose is determined, in particular by at least one reference marker in the storage environment. [6] Method for determining position according to claim 5, characterized by that the reference marker has at least one RFID marker, one UWB marker or one optically readable marker. [7] Method for determining position according to any of the preceding claims, characterized by , that during cornering maneuvers, the filter unit does not favor subsequent poses, whereby the cornering maneuver is detected by a movement of the steering angle sensor. [8] Method for determining position according to any of the preceding claims, characterized by , that odometry data about wheel speeds and steering angle position are recorded. [9] Method for determining position according to any of the preceding claims, characterized by , that a sequence of successor poses with a predetermined number of calculated successor poses undergoes a pose graph optimization in which the trajectory of successor poses is smoothed.

Images