Automatic driving decision-making and planning algorithm applied to complex intersection

Abstract

The invention provides an automatic driving decision-making and planning algorithm applied to a complex intersection. The automatic driving decision-making and planning algorithm comprises a prediction module and a planning module, the prediction module is provided with a condition prediction model and is used for predicting and outputting predicted trajectories of surrounding intelligent agents under different candidate trajectories by taking the candidate trajectories of the vehicle as conditions; the planning module comprises a trajectory tree construction module which is used for generating candidate trajectories of a vehicle and forming a trajectory tree; the scene tree construction module is used for generating predicted trajectories of other traffic participants around the vehicle to form a scene tree; the cost evaluation module is combined with a neural network-based feature extractor and a weight decoder to calculate the cost of each planning trajectory branch in the joint trajectory of the intelligent agent; the dynamic planning module is used for exporting candidate planning trajectories from the candidate trajectories output by the trajectory tree construction module and outputting the candidate planning trajectories to a vehicle control system; and the node pruning module is used for pruning low-fraction nodes in the trajectory tree and reducing the number of branches in the scene tree.

Description

Technical Field

[0001] The present invention relates to the field of autonomous vehicles, and more particularly, to an autonomous driving decision-making and planning algorithm applied to complex intersections. Background Art

[0002] As the core of future transportation systems, the optimization and improvement of the decision-making and planning systems of autonomous driving technology are the keys to achieving safe, efficient, and user-friendly driving experiences. In complex and ever-changing traffic environments, especially at busy intersections, autonomous vehicles must possess highly accurate prediction capabilities to accurately predict the future behaviors of other traffic participants, including pedestrians, bicycles, and other vehicles, whose behaviors are variable and unpredictable. Based on these predictions, the autonomous driving system needs to plan a driving path that is both safe and compliant with traffic rules to ensure the safety of all road participants. Intersections are one of the most complex and dangerous areas in the urban road network, where traffic fluidity and diversity pose great challenges to the driving of autonomous vehicles. The problem of visual occlusion at intersections also increases the decision-making difficulty of the autonomous driving system. Autonomous vehicles must be able to handle multi-directional traffic flows while responding to various traffic signals and complying with traffic rules.

[0003] In practical applications, existing autonomous driving decision-making and planning methods often neglect the importance of learning the costs of vehicle planning trajectories and fail to fully utilize historical data and real-time data to optimize cost functions. In addition, the prediction and planning systems in current systems are usually designed as independent modules, lacking effective information exchange and collaboration with each other. This separation may lead to delays and losses in processing information, increasing the risk of decision-making errors, thereby affecting the performance and efficiency of the entire planning system. Summary of the Invention

[0004] Aiming at the deficiencies in the prior art, the purpose of the present invention is to provide an autonomous driving decision-making and planning algorithm applied to complex intersections, which comprehensively considers the interactive effects between the behavior of the host vehicle and other traffic participants when dealing with complex situations at intersections, generates more accurate and real-time behavior prediction results, and realizes a safer, more efficient, and more user-friendly autonomous driving experience.

[0005] According to the autonomous driving decision-making and planning algorithm applied to complex intersections provided by the present invention, the algorithm includes a prediction module and a planning module;

[0006] The prediction module is provided with a conditional prediction model for using the candidate trajectories of the autonomous vehicle as conditions to predict and output the predicted trajectories of surrounding agents under different candidate trajectories;

[0007] The planning module includes:

[0008] A trajectory tree construction module, based on the road centerline and the future driving direction of the autonomous vehicle, generates candidate trajectories of the autonomous vehicle at different driving speeds to form a trajectory tree;

[0009] A scenario tree construction module, based on a conditional prediction model, generates predicted trajectories of the agents around the autonomous vehicle under different candidate trajectories to form a scenario tree;

[0010] A cost evaluation module, in combination with a neural network-based feature extractor and a weight decoder, calculates the cost of each planned trajectory branch in the joint trajectory of the agents;

[0011] A dynamic programming module derives candidate planned trajectories from the candidate trajectories output by the trajectory tree construction module and outputs them to the vehicle control system;

[0012] A node pruning module is used to prune the low-score nodes in the trajectory tree and reduce the number of branches in the scenario tree.

[0013] Preferably, the conditional prediction model uses a center query-based Transformer encoder to encode the environmental information and the past attributes of the autonomous vehicle and the surrounding agents, and uses a Transformer decoder to output the predicted trajectories of the surrounding agents based on self-trajectory prediction.

[0014] Preferably, the joint trajectory includes the candidate trajectories in the trajectory tree of the autonomous vehicle and the trajectories of the surrounding agents in the scenario tree obtained through the predicted Transformer decoder.

[0015] Preferably, the cost evaluation module adopts a cost function based on the scenario tree and the trajectory tree, and the cost function enables joint differentiable learning of the conditional prediction model and the cost evaluation module.

[0016] Preferably, the prediction module and the planning module can be jointly trained to ensure differentiability between the modules.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] 1. The autonomous driving decision-making and planning algorithm for complex intersections provided by the present invention proposes a differentiable joint training framework for conditional prediction and cost evaluation, directly improving the final planning performance, and training the prediction module and the planning module together to ensure differentiability between these two modules.

[0019] 2. The autonomous driving decision-making and planning algorithm for complex intersections provided by the present invention introduces a Transformer model based on central query in the prediction module. This model can efficiently perform self-conditioned motion prediction and consider the impact of the ego-vehicle's behavior on other agents during the prediction process.

[0020] 3. The autonomous driving decision-making and planning algorithm for complex intersections provided by the present invention adopts a tree-structured planning method. By constructing a trajectory tree and a scenario tree, it selects the optimal planned trajectory obtained according to the cost function and makes the conditional prediction and cost evaluation models jointly learnable. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Other features, objectives, and advantages of the present invention will become more apparent by reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0022] Figure 1 It is a structural diagram of the autonomous driving decision-making and planning algorithm for complex intersections in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. Components of the embodiments of the present invention described and illustrated in the drawings here can be arranged and designed in various different configurations.

[0024] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0025] It should be noted that: similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. In addition, all directional indications (such as up, down, left, right, front, back, bottom, etc.) in the present invention are only used to explain the relative positional relationship and motion conditions between components in a specific posture (as shown in the drawings). If the specific posture changes, the directional indications will also change accordingly. Further, the descriptions involving "first", "second", etc. in the invention are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features.

[0026] As Figure 1As shown in the figure, the present embodiment provides an autonomous driving decision-making and planning algorithm applied to complex intersections, including a prediction module and a planning module. The prediction module and the planning module can be combined for training to ensure differentiability between the modules. This algorithm can dynamically adjust the output of the prediction model and the trajectory cost function based on real-time data to adapt to the changing traffic environment. Since the prediction and decision-making planning systems are integrated, the prediction information can be transmitted to the decision-making planning module more accurately and losslessly, enabling the decision-making planning module to achieve stronger planning performance.

[0027] The prediction module is provided with a conditional prediction model. The conditional prediction model uses a Transformer encoder based on center query to encode the environmental information and the past attributes of the ego vehicle and surrounding agents, and uses a Transformer decoder to output the predicted trajectories of the surrounding agents based on possible self-trajectories. The conditional prediction model takes the candidate trajectories of the autonomous driving vehicle as conditions, and predicts and outputs the predicted trajectories of the surrounding agents under different candidate trajectories;

[0028] The planning module includes a trajectory tree construction module, a scenario tree construction module, a cost evaluation module, a dynamic programming module, and a node pruning module. The algorithm constructs two tree structures, namely a trajectory tree and a scenario tree. Among them, the trajectory tree is a set of candidate trajectories generated based on the road centerline and the future driving direction of the autonomous driving vehicle, and through different driving speeds; the scenario tree is a set of predicted trajectories of the surrounding agents output by the conditional prediction model through the Transformer decoder. The trajectory tree construction module is used to generate candidate trajectories of the autonomous driving vehicle to form a trajectory tree; the scenario tree construction module is used to generate predicted trajectories of other traffic participants around the autonomous driving vehicle to form a scenario tree.

[0029] The cost evaluation module combines a neural network-based feature extractor and a weight decoder to calculate the cost of each planned trajectory branch in the joint trajectory of the agents. Among them, the joint trajectory includes the candidate trajectories in the trajectory tree of the autonomous driving vehicle and the trajectories of the surrounding agents in the scenario tree obtained through the predicted Transformer decoder. The cost evaluation module also adopts a cost function based on the scenario tree and the trajectory tree, and the cost function can enable the conditional prediction model and the cost evaluation module to jointly perform differentiable learning.

[0030] The dynamic programming module derives candidate planned trajectories from the candidate trajectories output by the trajectory tree construction module and outputs them to the vehicle control system;

[0031] The node pruning module effectively prunes the low-score nodes in the trajectory tree through the learned cost model, reduces the number of branches in the scenario tree, and improves the running efficiency of the model. Through continuous pruning, the ideal vehicle driving trajectory is finally obtained.

[0032] The autonomous driving decision-making and planning algorithm for complex intersections provided by the present invention aims to provide a new technical solution for the decision-making and planning of autonomous vehicles at traffic intersections, so as to achieve a safer, more efficient and more user-friendly autonomous driving experience. In the application of intersection scenarios, this algorithm is particularly suitable for dealing with complex situations at traffic intersections because it comprehensively considers the interaction effects between the behavior of the host vehicle and other traffic participants, and can generate more accurate and real-time behavior prediction results. At traffic intersections, this algorithm effectively predicts the impact of the host vehicle's actions on the behavior of surrounding traffic participants through advanced conditional prediction techniques. This prediction takes into account possible behavior strategies of the host vehicle, such as accelerating through the intersection or stopping before a red light, and how these behaviors affect the reactions of other drivers and pedestrians; at intersections, complex interaction scenarios and diverse trajectory options must be evaluated. The cost evaluation module of this algorithm can calculate the costs of different trajectory options in real time, and optimize factors such as compliance with traffic rules, maintaining a safe distance between vehicles, and driving efficiency.

[0033] The autonomous driving decision-making and planning algorithm for complex intersections provided by the present invention proposes a differentiable joint training framework for conditional prediction and cost evaluation, which directly improves the final planning performance, and trains the prediction module and the planning module together to ensure the differentiability between these two modules. At the same time, a Transformer model based on center query is introduced into the prediction module, which can efficiently perform self-conditional motion prediction and consider the impact of the host vehicle's behavior on other agents during the prediction process. The present invention also adopts a tree structure planning method. By constructing a trajectory tree and a scenario tree, it selects the optimal planned trajectory obtained according to the cost function, and enables the conditional prediction and the cost evaluation model to be jointly learnable.

[0034] The specific embodiments of the present invention have been described above. Through the above description, relevant staff can make various changes and modifications completely within the scope of not deviating from the technical idea of this invention.

Claims

1. An automatic driving decision-making and planning algorithm applied to complex intersections, characterized in that: The algorithm includes a prediction module and a planning module; The prediction module is provided with a conditional prediction model for taking the candidate trajectory of the autonomous driving vehicle as a condition to predict and output the predicted trajectory of the surrounding intelligent agents under different candidate trajectories; The planning module includes: The trajectory tree construction module generates candidate trajectories of the autonomous vehicle based on the road centerline and the future driving direction of the autonomous vehicle and at different driving speeds to form a trajectory tree; The scenario tree construction module generates the predicted trajectories of the intelligent agents around the autonomous driving vehicle under different candidate trajectories based on the conditional prediction model to form a scenario tree; The cost evaluation module combines a neural network-based feature extractor and a weight decoder to calculate the cost of each planned trajectory branch in the agent's joint trajectory; A dynamic planning module, which derives candidate planning trajectories from the candidate trajectories output by the trajectory tree construction module and outputs them to the vehicle control system; The node pruning module is used to prune low-score nodes in the track tree and reduce the number of branches in the scene tree.

2. The automatic driving decision-making and planning algorithm applied to complex intersections according to claim 1 is characterized in that: The conditional prediction model uses a central query-based Transformer encoder to encode environmental information and past attributes of the autonomous vehicle and surrounding agents, and uses a Transformer decoder to output the predicted trajectory of the surrounding agents based on the self-trajectory prediction.

3. The automatic driving decision-making and planning algorithm applied to complex intersections according to claim 2 is characterized in that: The joint trajectory includes the candidate trajectory in the autonomous driving vehicle trajectory tree and the surrounding agent trajectory in the scene tree obtained by the predictive Transformer decoder.

4. The automatic driving decision-making and planning algorithm applied to complex intersections according to claim 1, characterized in that: The cost evaluation module adopts a cost function based on a scene tree and a trajectory tree, and the cost function can enable the conditional prediction model and the cost evaluation module to be jointly differentiable for learning.

5. The automatic driving decision-making and planning algorithm applied to complex intersections according to claim 1 is characterized in that: The prediction module and planning module can be trained together to ensure the differentiability between the modules.

Images