Vehicle and method for controlling a vehicle that is at least partially automated

Abstract

The invention relates to a vehicle (1) that is at least partially automated, comprising a computing unit (3) for determining an optimal route between a starting node and a destination node using an A* search algorithm for cost determination based on a plurality of distance-based costs.In a vehicle that significantly reduces the computational effort required to plan an optimal route for at least partially automated driving, the computing unit (3) for planning a trajectory of the vehicle (1) using a modified A* search algorithm comprises an analysis unit (5) which evaluates characteristic parameters of the vehicle (1) at predecessor nodes from a trajectory history, which characterize dynamic influences, and a trajectory determination unit (7) which, together with the distance-based costs, evaluates the costs caused by the dynamic influences of the vehicle (1). The computing unit (3) is designed to control actuators (9) for longitudinal and lateral control of the vehicle (1) such that the vehicle (1) follows the trajectory determined by the modified A* search algorithm.

Description

[0001] The invention relates to a vehicle that is at least partially automated, comprising a computing unit for determining an optimal driving route between a starting node and a destination node by means of an A* search algorithm for cost determination based on a plurality of distance-based costs, and a method for controlling an at least partially automated vehicle.

[0002] US Patent 2025 / 0020476A1 describes a vehicle navigation procedure based on an improved A* algorithm. The procedure includes: converting a regional map to a raster map for path planning; planning a destination path for an automated guided vehicle from a starting point to a destination point on the raster map using an improved A* algorithm, where the improved A* algorithm is an A* algorithm that incorporates turn-by-turn costs into a cost function; and navigating the automated guided vehicle along the destination path.

[0003] From DE 693 13 527 T2, a method for determining the optimal vehicle course between two points is known. The vehicle course is determined using a synthetic evaluation function with a plurality of parameters relating to travel costs, such as distance, average travel time, path width, and the number of curves in the vehicle course. A formulaic A* search algorithm is used as the synthetic evaluation function, which always examines first the nodes that are likely to lead to the destination quickly.

[0004] A method for local path planning based on high-resolution maps (HD maps) is known from US Patent 2024 / 0001961 A1. The method involves verifying a position on a global path, generating a path candidate group based on this global path, and selecting an optimal path from this group. The selection is made using a cost function that takes into account, among other things, the cost of following the global path, in order to favor a path with the smallest lateral offset from the global path.

[0005] WO 2024 / 144826 A1 describes a method for determining a road route by reducing the search space for path planning. From map data indicating the boundaries of a road lane, a portion of this lane is identified as a narrowed road lane with a reduced width. This is achieved by excluding areas close to environmental features (e.g., curbs, construction sites) from the original road lane. The subsequent evaluation of candidate paths takes place within this narrowed search space, thus improving the computational efficiency of path determination.

[0006] The object of the invention is to provide a vehicle and a method for controlling an at least partially automated vehicle, which significantly reduce the computational effort required for planning an optimal route for the at least partially automated vehicle.

[0007] The invention is defined by the features of the independent claims. Advantageous further developments and embodiments are the subject of the dependent claims. Further features, applications, and advantages of the invention will become apparent from the following description and the explanation of exemplary embodiments of the invention illustrated in the figures.

[0008] The problem is solved with the subject matter of claim 1 or 8.

[0009] In the vehicle described above, which is at least partially automated, comprising a computing unit for determining an optimal route between a start node and a destination node using an A* search algorithm for cost determination based on a plurality of distance-based costs, the computing unit for planning a trajectory of the vehicle using a modified A* search algorithm comprises an analysis unit, which evaluates characteristic parameters of the vehicle at predecessor nodes from a trajectory history, which determines the costs caused by the dynamic influences of the vehicle, and a trajectory determination unit, which together with the distance-based costs evaluates the costs caused by the dynamic influences of the vehicle, and the computing unit is designed in such a way as to control actuators for longitudinal and lateral control of the vehicle.that the vehicle follows the trajectory determined by the modified A* search algorithm, wherein, the analysis unit (5) is designed to create a weighted hierarchy of the predecessor nodes with respect to the parameters characterizing the dynamic influences by analyzing the geometric trajectory history.

[0010] Another aspect of the invention relates to a method for controlling an at least partially automated vehicle, in which an A* search algorithm for cost determination based on a plurality of distance-based costs is used to determine an optimal route between a starting node and a destination node, wherein a trajectory of the vehicle is determined by means of a modified A* search algorithm by - Key figures characterizing dynamic influences at predecessor nodes are determined and evaluated from a trajectory history, - from the characteristic values ​​that are influenced by dynamic factors, the costs caused by the dynamic factors of the vehicle are determined, - Based on the costs caused by the dynamic influences of the vehicle and the distance-based costs, control signals for longitudinal and lateral control of the vehicle are generated so that it follows the trajectory determined by the modified A* search algorithm. According to the invention, a weighted hierarchy of the predecessor nodes is created by analyzing the geometric trajectory history with regard to the characteristic values ​​that characterize the dynamic influences.

[0011] The modified A* search algorithm, which is also formulated as a formula, extends the well-known A* search algorithm by introducing a multi-layered cost function that evaluates path smoothness and dynamic feasibility based on historical trajectory information. By analyzing the path geometry, dynamic vehicle properties are integrated into graph-based algorithms. The modified A* search algorithm operates by maintaining a weighted hierarchy of predecessor nodes, calculating trajectory properties at different scales, and incorporating these into the cost function through an adaptive weighting scheme. This generates a natural bias towards dynamically feasible paths without requiring explicit trajectory computation.The key advantage lies in the ability to derive pseudo-dynamic constraints from purely geometric path properties, thereby achieving improved path quality without requiring the computational effort of a complete cinematic dynamic planning.

[0012] In a further embodiment, the analysis unit is designed to limit the number of considered predecessor nodes in order to determine the parameters characterizing the dynamic influences. Despite limiting the number of considered predecessor nodes and thus reducing computational effort, a reliable calculation of the costs caused by the vehicle's dynamic influences is possible. Furthermore, adaptive parameter adjustments based on the environmental context are continuously performed.

[0013] In a further embodiment, the analysis unit is designed to incorporate a time decay factor when determining the parameters characterizing the dynamic influences. This means that parameters from predecessor nodes whose determination occurred some time ago are included in the trajectory history analysis with a lower factor than parameters from predecessor nodes that are closer in time to the analysis node. Thus, the analysis gives greater weight to the dynamic behavior of the vehicle immediately preceding the analysis, allowing for a better consideration of the local conditions along the path.

[0014] In a further embodiment, the analysis unit is designed to calculate the costs underlying the dynamic influences by considering values ​​for a change in direction and / or values ​​for local path curvature and / or continuity costs for each node. This creates graph-based motion planning scenarios that deliver smooth, natural paths without post-processing.

[0015] In a further refinement, the analysis unit weights the values ​​for a change in direction and / or values ​​for local path curvature and / or continuity costs differently. This implicitly takes physical constraints into account.

[0016] In a further embodiment, the trajectory determination unit is designed to generate successor nodes, calculating the distance costs for each successor node and, by retrieving the values ​​for the considered predecessor nodes, the costs caused by the vehicle's dynamic influences. This ensures that the determined path exhibits pleasing dynamic properties without requiring explicit state-space expansions.

[0017] In a further embodiment, the trajectory determination unit is designed to dynamically calculate the costs caused by the vehicle's dynamic influences during the generation of successor nodes. This on-the-fly calculation reduces memory overhead during computation, as only essential node data, such as orientation, nodes, and predecessors, are stored. This is particularly effective because the pseudo-dynamic constraints are directly embedded in the cost calculation of the A* search algorithm, thereby avoiding both the computational complexity of a complete kinodynamic planning process and the need for subsequent path smoothing. This allows the hierarchy depth to be dynamically adjusted and weights to be modified without recalculating the nodes. Furthermore, the balance between path smoothness and optimality can be fine-tuned.This creates smooth transitions between different types of vehicle movement.

[0018] Further advantages, features, and details will become apparent from the following description, in which at least one embodiment is described in detail. The described features can, individually or in any meaningful combination, constitute the subject matter of the invention, optionally also independently of the claims, and can, in particular, also be the subject matter of one or more separate applications.

[0019] This shows: Fig. 1 an embodiment of the vehicle according to the invention.

[0020] In Fig.Figure 1 shows an embodiment of the vehicle according to the invention. The vehicle 1 under consideration is operated at least partially semi-automatically and comprises a computing unit 3 for determining an optimal route between a starting node and a destination node. The starting node shall be understood as the starting point of the vehicle 1 and the destination node as the destination of the vehicle 1. Furthermore, nodes shall be understood as individual local points that are considered for determining the route of the vehicle.

[0021] The processing unit 3 comprises an analysis unit 5, which evaluates key parameters of vehicle 1 at predecessor nodes based on a trajectory history, characterizing dynamic influences, and determines the costs caused by these dynamic influences. Analysis unit 5 is coupled to a trajectory determination unit 7, which, together with the distance-based costs, evaluates the costs caused by the dynamic influences of vehicle 1. As a result of the cost evaluation, processing unit 3 generates signals to control vehicle actuator 9 for lateral and / or longitudinal control of vehicle 1.

[0022] Computing unit 3 stores a procedure for a modified A* search algorithm, which automatically determines an optimal route from a starting point to a desired destination point in the form of a cost-optimized trajectory within a predefined node network or grid, or generates successor nodes using motion primitives. A formulaic A* search algorithm is used, which always examines first the nodes that are likely to lead to the destination quickly. To identify the most promising node, each known node n is assigned a metric f(n) = g(n) + h(n), which provides an estimate of the path length from the starting node to the destination node using the considered node in the most favorable case.

[0023] A fundamental consideration of the optimized A* search algorithm using adaptive motion primitives will be explained in more detail below.

[0024] The modified A* search algorithm comprises the following steps: To determine the metric f(n) = g(n) + h(n) for node n, a search using an extended Cost to Come method is initiated. Distance-based costs gbase(n) are calculated using a known A* search algorithm. Simultaneously, further costs Agtrajectory(n) dependent on the dynamic properties of vehicle 1 are determined. g(n)=gbase(n)+λgtrajectory(n)

[0025] These include: - gbase(n): distance-based costs - gtrajectory(n): Trajectory costs caused by the dynamic properties of vehicle 1 and determined from a trajectory history, - h(n): Traditional heuristic, e.g. distance-based (Euclidean) - λ: Weighting factor (0 ≤ λ ≤ 1)

[0026] The following applies to the determination of trajectory costs: gtrajectory(n)=∑i=1kαi⋅C(n,i) where: • k: Number of predecessor nodes considered • α: Time decay factor (0 ≤ α ≤ 1) • C(n,i): Cost component of the considered node n at hierarchical depth i • n: Currently evaluated node

[0027] The cost components C(n,i) that characterize the dynamic properties of vehicle 1 consist of: C(n,i)=w1Δθi+w2ki+w3ci where: • Δθ i = |θ(n) - θ(n -i )|: Change of direction • κ i : Local path curvature • W1, W2, W3: Weighting factors • c i : Continuity costs (curvature of the path in relation to step size) • n -i : Predecessor node at depth i

[0028] This is followed by node expansion in the optimized A* search algorithm. During node expansion, successor nodes are generated using motion primitives. The endpoints of a set of motion primitives define the nearest possible intermediate nodes to the target node. For each successor node, the base costs are calculated, which depend on the distance to the predecessor node. This is achieved by retrieving a chain of predecessor nodes, followed by the determination of parameters characterizing the dynamic influences of the vehicle, subsequently referred to as trajectory costs. The costs g(n) determined for each successor node are then combined into an extended cost metric. The A* search algorithm is then continued with this extended cost metric. The weighting factors preferably decrease with increasing depth i; for example, they could each be multiplied by a factor of 1 / i.

[0029] The optimized A* search algorithm dynamically calculates the trajectory costs of the considered node n caused by the dynamic properties of vehicle 1 during node expansion, thereby reducing the recalculation overhead during node expansion. Recalculations are negligible if the number k of considered predecessor nodes is less than or equal to 5.

[0030] The inventive method will now be explained in more detail using an exemplary embodiment. To determine a path P with implied dynamic executability, a map M is specified with a start node s, a target node g, and a time depth k. Subsequently, a priority queue Q is initialized with the start node s. During the subsequent node expansion, a current node n and several predecessor nodes A = {n t-1 , n t-2 , ..., n t-k} selected, for which for each motion primitive m the successor state s with m, c base is calculated as the Euclidean distance (n, s), where c dynamik goes to 0. For every time scale i ε [1, k], the following holds: A. Δθ i as a change in direction of node s relative to its predecessor node Ai B. k i as local curvature of node s to its predecessor node Ai C. c Continuity costs D. c dynamik ← c dynamic + α i (ω θ Δθ i + ω κ k i ω i c i )

[0031] Then a C total educated: ctotal=λcbase+(1−λ)cdynamic which is inserted into the priority queue Q. This process is repeated until the target node g is reached.

[0032] After completion of the procedure, the vehicle's computing unit 3 sends control signals to the actuators 9 for longitudinal and lateral control of the vehicle 1, so that the vehicle 1 follows the trajectory determined according to the modified A* search algorithm.

Claims

[1] Vehicle (1) that is at least partially automated, comprising a computing unit (3) for determining an optimal route between a starting node and a destination node using an A* search algorithm for cost determination based on a plurality of distance-based costs, wherein The computing unit (3) for planning a trajectory of the vehicle (1) using a modified A* search algorithm comprises an analysis unit (5) which evaluates characteristic values ​​of the vehicle (1) at predecessor nodes from a trajectory history, which characterize dynamic influences, and from which it determines the costs caused by the dynamic influences of the vehicle (1), and a trajectory determination unit (7) which, together with the distance-based costs, evaluates the costs caused by the dynamic influences of the vehicle (1), and the computing unit (3) is configured to control actuators (9) for longitudinal and lateral control of the vehicle (1) such that the vehicle (1) follows the trajectory determined according to the modified A* search algorithm. characterized by , that the analysis unit (5) is designed to create a weighted hierarchy of the predecessor nodes with respect to the parameters characterizing the dynamic influences by analyzing the geometric trajectory history. [2] Vehicle according to claim 1, characterized by , that the analysis unit (5) is designed in such a way as to limit the number of considered predecessor nodes in order to determine the characteristic values ​​that characterize the dynamic influences. [3] Vehicle according to claim 1 or 2, characterized by , that the analysis unit (5) is designed in such a way as to take into account a time decay factor in order to determine the characteristic values ​​that characterize the dynamic influences. [4] Vehicle according to at least one of the preceding claims, characterized by, that the analysis unit (5) is designed to take into account values ​​for a change of direction and / or values ​​for a local path curvature and / or continuity costs for each node in order to calculate the costs underlying the dynamic influences. [5] Vehicle according to claim 4, characterized by , that the analysis unit (5) weights the values ​​for a change of direction and / or values ​​for a local path curvature and / or continuity costs differently. [6] Vehicle according to at least one of the preceding claims, characterized by , that the trajectory determination unit (7) is designed to generate successor nodes, whereby for each successor node the distance costs are calculated and, by retrieving the values ​​for the considered predecessor nodes, the costs caused by dynamic influences of the vehicle (1) are calculated. [7] Vehicle according to claim 6, characterized by, that the trajectory determination unit (7) is designed to dynamically calculate the costs caused by dynamic influences of the vehicle (1) during the generation of the successor nodes. [8] Method for controlling a vehicle that is at least partially automated (1) in which an A* search algorithm for cost determination based on a plurality of distance-based costs is used to determine an optimal route between a starting node and a destination node, where a trajectory of the vehicle (1) is determined using a modified A* search algorithm, by - Key figures characterizing dynamic influences at predecessor nodes are determined and evaluated from a trajectory history, - from the characteristic values ​​that are characterized by dynamic influences, the costs caused by the dynamic influences of the vehicle (1) are determined, - based on the costs caused by the dynamic influences of the vehicle (1) and the distance-based costs, control signals for longitudinal and lateral control of the vehicle (1) are generated so that it follows the trajectory determined by the modified A* search algorithm, characterized by , that by analyzing the geometric trajectory history, a weighted hierarchy of the predecessor nodes is created with regard to the parameters characterizing the dynamic influences.

Images