Method for predicting the trajectory of a vehicle
Patent Information
- Authority / Receiving Office
- EP · EP
- Patent Type
- Applications
- Current Assignee / Owner
- ROBERT BOSCH GMBH
- Filing Date
- 2024-06-27
- Publication Date
- 2026-05-27
Smart Images
Figure EP2024068121_23012025_PF_FP_ABST
Abstract
Description
[0001] Description
[0002] title
[0003] Method for trajectory prediction for a vehicle
[0004] State of the art
[0005] The present invention relates to a method for trajectory prediction for a vehicle, a system for trajectory prediction, and a vehicle.
[0006] Methods for predicting vehicle trajectories, i.e., routes the vehicle will take in a potential future journey, are known. Such predictions are often based on navigation data from the vehicle's navigation system. It is also known that such predicted trajectories can be exchanged with other vehicles, for example, via V2X transmission, to optimize traffic flow within the traffic system.
[0007] Disclosure of the invention
[0008] The method according to the invention with the features of claim 1 is distinguished by the fact that a particularly reliable and precise estimation of vehicle trajectories can be carried out in a simple and cost-effective manner. This makes it possible, in particular, to optimize a coordinated traffic flow of several vehicles in a networked traffic system. This is achieved according to the invention by a method for trajectory prediction for a vehicle, preferably for an electric bicycle, comprising the steps:
[0009] - Estimation of at least one predicted trajectory, which depicts a future route of the vehicle,
[0010] - Determining state history data of the vehicle over at least one predetermined, preferably past, period of time, and - Estimating a probability and / or an uncertainty of the predicted trajectory based on the at least one predicted trajectory and additionally based on the determined state history data.
[0011] In other words, the method estimates a future travel path for the vehicle. In particular, location information and preferably additional movement information, such as speed, of the vehicle's travel within a predetermined future period are regarded as the trajectory. The predicted trajectory is preferably determined for at least a predetermined future period starting from the current point in time, which is, for example, at least 10 seconds. In addition, the method determines state history data of the vehicle for at least one further predetermined period. In particular, past travel information of the vehicle is regarded as state history data. The state history data preferably includes location information and movement information of the vehicle's travel.Using the determined predicted trajectory and based on the determined state history data, an estimate of the probability and / or uncertainty of the determined predicted trajectory is made.
[0012] In particular, probability is considered to be a measure of how likely it is that the vehicle will move along a certain predicted trajectory, for example in comparison to other possible trajectories.
[0013] Uncertainty is considered, in particular, a measure of how precise the determined predicted trajectory is. In particular, uncertainty is considered a statistically calculated measure of the inaccuracy of a predicted future position of the vehicle.
[0014] The method thus offers the advantage of obtaining particularly detailed information about the vehicle's future route. This can improve the vehicle's ferry operation, for example, while a navigation system is operating. Furthermore, particularly in a networked transport system where trajectories of different vehicles can preferably be exchanged, the coordination of these different vehicles' routes can be optimized particularly efficiently and precisely.
[0015] The subclaims contain preferred developments of the invention.
[0016] The state history data is preferably determined using sensor data from the vehicle. For example, a vehicle's movement state can be monitored and recorded using a speed sensor and / or an inertial sensor system. In particular, past and current movement of the vehicle can advantageously be monitored using an inertial sensor system, allowing, for example, a simple and reliable estimate of the probability and / or uncertainty with which the vehicle will move along the predicted trajectory.
[0017] Particularly preferably, the predicted trajectory is estimated using map information, in particular from a vehicle's navigation system. For example, the predicted trajectory can be directly derived from navigation data from the navigation system. Further preferably, the predicted trajectory can be estimated based on the current position and direction of movement determined by the navigation system.
[0018] Preferably, the status history data is determined for a past period of time, in particular up to the current time, of at least 2 seconds, preferably a maximum of 10 seconds. Alternatively, the status history data can preferably be determined for an entire journey of the vehicle. This enables a simple and cost-effective implementation of the method.
[0019] Further preferably, the estimation of the probability and / or uncertainty is performed for several possible predicted trajectories. This means that the method estimates several potentially possible trajectories along which the vehicle could move in the future. The probability and / or uncertainty is estimated for each of these trajectories. This allows particularly precise information about the vehicle's future movement to be determined, enabling particularly advantageous coordination of multiple road users, particularly in a networked traffic system.
[0020] Preferably, the estimation of the probability and / or uncertainty of the predicted trajectory is performed using a neural network. In other words, the estimation of the probability and / or uncertainty is performed using deep learning. This allows for detailed information about the predicted trajectories to be obtained easily and with particularly high precision. In particular, the method allows the use of a particularly simple neural network, which allows for precise determination with limited resources, especially in terms of computing power.
[0021] Particularly preferably, in the method, map information, in particular from a vehicle's navigation system, is encoded using a multilayer perceptron. The state history data is encoded using a long short-term memory. Subsequently, based on the encoded map information and the encoded state history data, a classification of the determined predicted trajectory is performed. Based on this classification, the probability of the determined predicted trajectory can be determined. This means that a special calculation or analysis of the provided information, namely the determined predicted trajectory and the state history data, is performed in order to subsequently estimate the probability of the trajectory using simple classification. This enables a particularly simple implementation of the method, which requires few computing resources.
[0022] Further preferably, the determined predicted trajectory and also the state history data are normalized and reduced in dimensionality using a principal component analysis. This correspondingly reduced data is then clustered using the K-means algorithm. A statistical analysis is then performed for each cluster generated in this way to determine the uncertainty of the predicted trajectory. This means that, preferably using a simple machine model, a particularly simple and cost-effective calculation or analysis of the available information can be performed with minimal computing resources, in order to obtain knowledge about the uncertainty of the determined trajectory.
[0023] The method preferably further comprises the step of exchanging the at least one predicted trajectory together with the associated determined probability and / or uncertainty with at least one other road user. The exchange with the other road user preferably takes place via a V2X transmission (V2X: short for vehicle-to-everything). This enables simple data transmission between multiple road users, even over long distances. Using the exchanged information, for example, multiple road users in a networked traffic system can thus determine with particular precision along which trajectories the other road users will move and with what probability. Particularly preferably, this can be used to provide a collision warning system, for example for at least one of the road users.
[0024] Preferably, the method further comprises the step of determining an impending collision between the vehicle and another road user, in particular depending on the determined probability and / or uncertainty of the trajectory. For example, the vehicle trajectory with the highest probability can be considered. This enables particularly efficient and precise detection of an impending collision.
[0025] The method further preferably comprises the step of issuing a warning in response to a detected, particularly impending, collision. For example, the warning can be issued acoustically and / or visually to a user of the vehicle, preferably by means of an output device and / or a user device.
[0026] Furthermore, the invention leads to a system for trajectory prediction for a vehicle, preferably for an electric bicycle. The system comprises a control unit, which is in particular a control unit of the vehicle, wherein the control unit is configured to carry out the described
[0027] Method. Furthermore, the system can preferably comprise a navigation system and / or a sensor system, such as an environmental sensor system and / or an inertial sensor system.
[0028] Furthermore, the invention relates to a vehicle, preferably an electric bicycle, which comprises the described system for trajectory prediction.
[0029] Short description of the drawings
[0030] The invention is described below using exemplary embodiments in conjunction with the figures. In the figures, functionally identical components are identified by the same reference numerals. Here:
[0031] Figure 1 is a simplified schematic view of a vehicle in which a method for trajectory prediction is carried out according to a preferred embodiment of the invention,
[0032] Figure 2 is a simplified schematic view of the steps for estimating a probability of a predicted trajectory when carrying out the method of the preferred embodiment, and
[0033] Figure 3 is a simplified schematic view of the steps for estimating an uncertainty of a predicted trajectory when carrying out the method of the preferred embodiment.
[0034] Preferred embodiments of the invention
[0035] Figure 1 shows a simplified schematic view of a vehicle 100 in which a method for trajectory prediction according to a preferred embodiment of the invention is being carried out. The vehicle 100 is an electric bicycle comprising a drive unit 101 configured to assist a rider's pedaling force using a torque generated by an electric motor. A control unit 105 is integrated into the drive unit 101, which is configured in particular to carry out the method according to the invention.
[0036] Vehicle 100 can, for example, be part of a networked traffic system that includes other road users that can communicate with each other via V2X transmission. For example, the users of the networked traffic system can exchange data with each other, preferably to avoid collisions.
[0037] The method involves estimating one or more predicted trajectories 1, 2, 3, which depict possible future travel paths of the vehicle 100, as shown by way of example in Figure 1. The predicted trajectories 1, 2, 3 can be estimated using several different data and methods. In particular, the estimation is carried out based on sensor data from a sensor system of the vehicle 100 and based on navigation data from a navigation system of the vehicle 100. Preferably, the sensor data obtained is movement data from movement sensors, such as a speed sensor, an inertial sensor system, and the like, of the vehicle 100. Furthermore, the navigation data preferably includes map information for an environment in which the vehicle 100 is currently located.
[0038] In addition, the method involves determining status history data of the vehicle 100 over at least a predetermined period of time. In particular, data relating to the past travel history of the vehicle 100 up to the current point in time are determined as status history data. Particularly preferably, the status history data includes location information and movement information relating to the journey of the vehicle 100 up to the current point in time.
[0039] The predetermined period preferably comprises a period of at least 2 seconds. Alternatively, the state history data can also be recorded over the entire journey of the vehicle 100. Based on the predicted trajectories 1, 2, 3 thus determined and the determined state history data, a probability is then determined for each of the trajectories 1, 2, 3. This is illustrated in simplified schematic form in Figure 2, which schematically depicts a flowchart 20 for the steps for estimating the probability. These steps are carried out, in particular, using a neural network.
[0040] In the first step 21, the predicted trajectories 1, 2, 3 are estimated, and in particular path parameters of the trajectories 1, 2, 3 are determined, such as directions and curvatures, potential speeds of the vehicle 100 when moving along the trajectories, distances between relevant points, such as curves, of the trajectories 1, 2, 3 and the like.
[0041] Subsequently, in step 23, the corresponding data of the trajectories 1, 2, 3 are encoded using a multi-layer perceptron using the determined trajectories 1, 2, 3.
[0042] Furthermore, preferably simultaneously in step 22, the state history data is determined, which is encoded in step 24 by means of a long short-term memory (also: LSTM; long short-term memory).
[0043] The corresponding coded data of the trajectories 1, 2, 3 and state information are then jointly classified in step 25 using a multi-layer perceptron to determine the probabilities, which can be output in step 26.
[0044] In particular, the probability of each of the determined possible trajectories 1, 2, 3 is estimated separately. In particular, the trajectory 1, 2, 3 with the highest probability can be output in step 26.
[0045] Furthermore, the method determines an uncertainty of the determined predicted trajectory 1, 2, 3. This is shown schematically in Figure 3, which schematically represents a further flowchart 30 for the steps for estimating the uncertainty. These steps are carried out, in particular, using a simple classical machine model, enabling a particularly simple and cost-effective implementation with computationally limited hardware.
[0046] In the first step 31, the predicted trajectories 1, 2, 3 and the state history data are determined, for example, analogous to steps 21 and 22 in Figure 2. In the subsequent step 32, the data thus obtained are normalized and their dimensionality reduced using a principal component analysis (PCA). The reduced data are then clustered using a K-means algorithm in step 33. For each cluster, a statistical analysis is performed in the subsequent step 34, with the results of the statistical analysis representing the uncertainties of the various trajectories 1, 2, 3.
[0047] The method can thus determine and assess the predicted trajectories 1, 2, 3 along which the vehicle 100 can potentially travel in the future. The information thus obtained regarding the probabilities and uncertainties of these trajectories 1, 2, 3 can be exchanged with other road users in the networked traffic system in order to enable optimized ferry operations for all participants in the traffic system. Advantageously, a collision warning system can be provided, which can offer particularly precise and reliable functionality through knowledge of the probabilities and uncertainties of the predicted trajectories.
Claims
Claims 1. A method for trajectory prediction for a vehicle (100), in particular for an electric bicycle, comprising the steps: - estimating at least one predicted trajectory (1), wherein the predicted trajectory (1) depicts a future travel path of the vehicle (100), - determining status history data of the vehicle (100) over at least a predetermined period of time, and - Estimating a probability and / or an uncertainty of the predicted trajectory (1) based on the at least one predicted trajectory (1) and the determined state history data.
2. The method according to claim 1, wherein the state history data is determined using sensor data of the vehicle (100).
3. Method according to one of the preceding claims, wherein the estimation of the predicted trajectory (1) is carried out by means of map information, in particular a navigation system of the vehicle (100).
4. Method according to one of the preceding claims, wherein the status history data are determined for a past period of at least 2 seconds, preferably a maximum of 10 seconds.
5. Method according to one of the preceding claims, wherein the estimation of the probability and / or the uncertainty is carried out for several possible predicted trajectories (1, 2, 3).
6. Method according to one of the preceding claims, wherein the estimation of the probability and / or the uncertainty of the predicted trajectory (1) is carried out by means of a neural network.
7. Method according to claim 6, - where map information is encoded using a multi-layer perceptron, - where state history data is encoded using a long short-term memory, and - wherein a classification of the predicted trajectory (1) is carried out by means of the coded map information and state history data in order to determine the probability of the predicted trajectory (1).
8. Method according to one of the preceding claims, - where the predicted trajectory (1) and the state history data are normalized and reduced in dimensionality using principal component analysis, - where the data is then clustered using the K-Means algorithm, and - where a statistical analysis is then carried out for each cluster to determine the uncertainty of the predicted trajectory (1).
9. Method according to one of the preceding claims, further comprising the step: - exchanging the at least one predicted trajectory (1) together with the determined probability and / or uncertainty with at least one other road user, in particular by means of a V2X transmission.
10. The method according to any one of the preceding claims, further comprising the step: - Determining an impending collision between the vehicle (100) and another road user, in particular depending on the determined probability and / or uncertainty of the predicted trajectory (1), and - preferably issuing a warning in response to a detected collision.
11. System for trajectory prediction for a vehicle (100), in particular for an electric bicycle, comprising a control unit (105) which is designed to Carrying out the method according to one of the preceding claims.
12. Vehicle, in particular electric bicycle, comprising a system (110) according to claim 11.