CN116164748A - Decision planning system and decision planning device for complex road conditions in urban areas - Google Patents

Abstract

The present invention provides a decision-making planning system and a decision-making planning device for complex road conditions in urban areas. The decision-making planning system includes: detecting the driving parameters of the driving vehicle and surrounding environment parameters to obtain multiple original data; importing multiple original data into the AI model for processing to obtain multiple perception results; calculate the planned trajectory through multiple perception results to obtain the first planned trajectory result; the AI model obtains the second planned trajectory result based on multiple original data; combine the first planned trajectory result and the second planned trajectory The results are imported into the trajectory scoring arbitration module to verify the best trajectory and output the best trajectory. The embodiment of the present invention solves the problem of low efficiency and security in the existing algorithm.

Description

Decision-making planning system and decision-making planning device for complex road conditions in urban areas

technical field

The invention relates to the field of automation, in particular to a decision-making planning system and a decision-making planning device for complex road conditions in urban areas.

Background technique

At present, the mainstream autonomous driving algorithm technology stack is mainly based on the layered "perception-positioning-decision-making planning-control", and each module is relatively independent and decoupled. From the perspective of parallel development efficiency, it is obviously impossible to leave code problems to be exposed during multi-module joint debugging of the real vehicle. Each module should first be responsible for its own output. The problem with the decision-making planning module is that it does not have a clear label of the true value of the perceived target object like the perception module (for example, whether the currently seen object is a truck or a car, and it is easy to judge if the output result of the perception algorithm does not match the actual one), Both the positioning module and the control module have their own positioning accuracy and control accuracy as evaluation indicators. But for the output of the decision-making planning module, apart from the safety requirement that collisions cannot occur, there is a lack of a similar unique correct truth value as a reference. Therefore, the effect of the current decision-making planning algorithm still largely depends on the joint debugging effect of the control module on the real vehicle. This is not conducive to the agile development and rapid iteration of algorithms.

In addition, in actual mass production, due to limitations in computing power and hardware costs, the decision-making and planning algorithm technical solutions used cannot give a good balance between solving efficiency and complex scene processing capabilities. For example, on a road in a certain urban area, assuming that there are more than 20 vehicles traveling at different speeds around the current vehicle, it is difficult for traditional decision-making and planning algorithms to plan interactive trajectories for multiple traffic participants at the same time, causing its strategy to tend to Because it is conservative, it is not easy to search and solve the trajectory with higher driving efficiency.

Contents of the invention

Therefore, the embodiments of the present invention provide a decision-making planning system and a decision-making planning device for complex road conditions in urban areas, which solve the problems of low efficiency and safety in existing algorithms.

In order to solve the above problems, the present invention provides a decision-making and planning system for complex road conditions in urban areas, including: detecting the driving parameters of the driving vehicle and surrounding environment parameters, and obtaining multiple original data; importing multiple original data into the AI model for processing, and obtaining Multiple perception results; calculate the planned trajectory through multiple perception results to obtain the first planned trajectory result; the AI model obtains the second planned trajectory result based on multiple original data; import the first planned trajectory result and the second planned trajectory result into the trajectory The scoring arbitration module verifies the best trajectory and outputs the best trajectory.

Compared with the existing technology, the technical effect achieved by adopting this technical solution: the original data is obtained by setting the driving parameters of the detected vehicle and the surrounding environment parameters, so that the source of the data is more suitable for the actual situation of the current vehicle driving, according to the actual situation According to the situation, the following trajectory planning is carried out, so that the trajectory is more in line with the current driving environment of the vehicle, and the raw data is imported into the AI model for processing, and the perception results are obtained, so as to perform traditional trajectory calculations and obtain the first planned trajectory results , and then use the AI model to process the second planning trajectory based on the original data, so that there are more ways to calculate the trajectory, and different trajectories can be obtained through different methods, so that the practicability and safety of the trajectory are higher. At the same time, the The results of the two planned trajectories are imported into the scoring arbitration module to verify the best trajectory, which minimizes the risk of the trajectory and improves the efficiency of the algorithm. It can more quickly obtain the best driving trajectory of the vehicle in the current driving environment. The safety of the driver and the driving vehicle is guaranteed. At the same time, the AI model is used to process the data to obtain different trajectory models, which makes the trajectory planning of the automatic driving system more comprehensive, safe and complete. At the same time, the AI model can better respond to The current driving environment makes a response decision, and the algorithm is more efficient, enabling the vehicle to safely drive automatically in a complex driving environment, improving the safety and practicality of automatic driving, and ensuring the safety of the driver.

In an example of the present invention, multiple raw data are imported into the AI model for processing to obtain multiple perceptual results, which also includes: using convolutional neural networks to preprocess multiple raw data and extract local features to obtain multiple features vector.

Compared with the existing technology, the technical effect achieved by adopting this technical solution: by setting and using the convolutional neural network to preprocess the original data and extract local features, the obtained original data can quickly remove impurities and interference factors, and obtain The required data features, and local feature extraction can quickly extract the features required by the system and convert them into corresponding feature vectors for subsequent operations, making the algorithm more efficient, the process more accurate, and the results provided clearer. The planning trajectory is more convenient and safe, making the decision-making and planning system for complex road conditions in urban areas more practical and safe, and ensuring the safety of drivers.

In an example of the present invention, the AI model obtains the second planned trajectory result based on multiple original data, which also includes: combining multiple feature vectors through splicing operations to obtain the combined feature vector; through a fully connected neural network structure and Transformer The network structure acts as an intermediate layer to perform feature conversion and encoding on the merged feature vectors to obtain intermediate results.

Compared with the existing technology, the technical effect achieved by adopting this technical solution: by splicing and merging multiple feature vectors through setting, the subsequent operation and processing are more convenient and fast, and the efficiency of the subsequent algorithm is faster, and then through the full connection The neural network structure and the Transformer network structure are used to perform feature conversion and encoding on the merged feature vectors to obtain intermediate results, so that the combined feature vectors can be used for subsequent processing operations, and the conversion process is more efficient. At the same time, the obtained intermediate results can be used directly, which is more efficient. The convenience makes the algorithm more efficient.

In an example of the present invention, the AI model obtains the second planning trajectory result based on a plurality of original data, and further includes: inputting the intermediate result into the LSTMDecoder network to generate a series of trajectory sequences, and pre-generating decision planning trajectory points; The planned trajectory points are processed, and the output is the second planned trajectory result.

Compared with the existing technology, the technical effect achieved by adopting this technical solution: a series of trajectory sequences are generated by inputting the intermediate results into the LSTMDecoder network, so that the trajectory can be better predicted through the LSTMDecoder network. The trajectory is a set of continuous Data information has time continuity and spatial randomness, and the LSTMDecoder network can predict and process it very well, making trajectory prediction more accurate. The decision planning trajectory points are processed through the trajectory model, and the output is the second planning trajectory result , and the model combined with the LSTMDecoder network can make the accuracy of the model higher, making subsequent predictions more accurate, making the decision-making planning system more comprehensive and safe, better protecting the safety of drivers, and making autonomous driving more intelligent .

In an example of the present invention, importing the result of the first planned trajectory and the result of the second planned trajectory into the trajectory scoring arbitration module to verify the optimal trajectory, and outputting the optimal trajectory, it also includes: performing trajectory prediction through the prediction module to obtain the predicted trajectory, The credibility of the predicted trajectory is evaluated to obtain a third planned trajectory result; the third planned trajectory result and the second planned trajectory result are subjected to collision screening to obtain a fourth planned trajectory result.

Compared with the existing technology, the technical effect achieved by adopting this technical solution: by setting the prediction module to predict the trajectory, the predicted trajectory is obtained, and the credibility of the predicted trajectory is evaluated, and the result of the third planned trajectory is obtained, so that the predicted trajectory To assist the trajectory generated by the AI model to judge and evaluate, combine multiple trajectories to make the final output trajectory more comprehensive and accurate, more in line with the current environment of the driving vehicle, and protect the safety of the driver. The results of the second planning trajectory are subjected to collision screening to obtain the result of the fourth planning trajectory. On the one hand, the overall solution efficiency of the algorithm is improved, and a large amount of computing power is avoided on invalid trajectories for complete feasibility inspection and comfort scoring; at the same time, it is also done once. Security screening. At the same time, the collisions of some low-confidence trajectory points may be ignored. The effective self-vehicle planning trajectory will not be deleted blindly because of some low-confidence predicted trajectories, and the scope and completeness of the algorithm solution space will be expanded.

In an example of the present invention, the first planned trajectory result and the fourth planned trajectory result are subjected to comfort score, anthropomorphic score and efficiency score; as a result of the comfort score, anthropomorphic score and efficiency score, the trajectory with a high score is evaluated. output.

Compared with the existing technology, the technical effect achieved by adopting this technical solution: by setting the results of the first planned trajectory and the result of the fourth planned trajectory for comfort score, anthropomorphic score and efficiency score, the output results are passed through three After scoring, it can be more representative and practical, and more in line with the current road conditions of the driving vehicle, so that the trajectory can ensure the safe driving of the driving vehicle and the safety of the driver.

In an example of the present invention, the comfort score further includes: comfort=normalized (1/track curvature)+normalized (1/acceleration rate).

Compared with the existing technology, the technical effect achieved by adopting this technical solution: the trajectory is evaluated by setting the comfort level, and the comfort level is mainly measured by the two indicators of the trajectory curvature change rate and the planning acceleration change rate: that is, when the Under the premise of safety, the lower the rate of curvature change and acceleration rate of the trajectory, the better, so that the trajectory with a high score can be safer and more practical.

In an example of the present invention, the efficiency score further includes: efficiency=normalization (total length of track).

Compared with the existing technology, the technical effect achieved by adopting this technical solution: the trajectory is scored by setting the efficiency, so that the longer the path, the higher the speed and the better the efficiency, so that the trajectory obtained through efficiency scoring can be better to drive at the current intersection.

In an example of the present invention, the anthropomorphic score further includes: anthropomorphic = normalization (1/Frescher distance (corresponding to the average trajectory of the human driver in the scene, the target trajectory)).

Compared with the existing technology, the technical effect achieved by adopting this technical solution: scoring the trajectory by setting the anthropomorphic degree, so that the driving of the trajectory can be more in line with human standards, and at the same time more consistent with the driving habits of existing drivers, ensuring While driving on the best route under the current driving situation, it makes the trajectory more humanized and ensures driving safety.

The present invention also provides a decision-making and planning device, including: acquisition module: detect driving parameters of the driving vehicle and surrounding environment parameters, and obtain multiple original data; processing module: import multiple original data into AI model for processing, and obtain multiple perception Result; calculate the planned trajectory through multiple perception results to obtain the first planned trajectory result; the AI model obtains the second planned trajectory result based on multiple original data; scoring module: import the first planned trajectory result and the second planned trajectory result into the trajectory The scoring arbitration module verifies the best trajectory and outputs the best trajectory.

Compared with the existing technology, the technical effect achieved by adopting this technical solution: the original data is obtained by setting the acquisition module to obtain the current driving road condition environment of the vehicle, so that the trajectory planning of the vehicle is more in line with the actual situation of the current environment, and the planning It is more practical, and the original data is processed by the processing module to obtain the first planned trajectory result and the second planned trajectory result, so that the obtained two trajectory results are more accurate and in line with the actual situation, and the evaluation of multiple trajectory results at the same time Comparison can make the final output track more safe and practical. At the same time, the scoring module is used to score and verify the two tracks, so that the output home is the safest, fastest and most humanized track. Through the cooperation of the three modules, The algorithm is more efficient, and at the same time, it can take into account different driving environments and has higher practicability.

Adopting this technical scheme has the following beneficial effects:

(1) The original data is obtained by setting the driving parameters of the detected vehicle and the surrounding environment parameters, so that the source of the data is more in line with the actual situation of the current vehicle driving, and the next trajectory planning is carried out according to the actual situation, so that the trajectory is more in line with the current situation The driving environment of the vehicle is processed by importing the original data into the AI model, and the perception result is obtained, and the traditional trajectory calculation is performed to obtain the first planning trajectory result, and then the AI model is processed based on the original data to obtain the second The result of the planned trajectory enables more calculation methods of the trajectory, and different trajectories are obtained through different methods, so that the practicability and safety of the trajectory are higher. At the same time, the two planned trajectory results are imported into the scoring arbitration module to verify that the best The trajectory minimizes the risk of the trajectory, and also improves the efficiency of the algorithm. It can more quickly obtain the optimal driving trajectory of the vehicle in the current driving environment, ensuring the safety of the driver and the driving vehicle. At the same time, through the AI model To process the data to obtain different trajectory models, so that the trajectory planning of the automatic driving system is more comprehensive, safe and complete. At the same time, the AI model can better respond to the current driving environment and make decisions, and the algorithm is more efficient. It enables the vehicle to safely drive automatically in a complex driving environment, improves the safety and practicability of automatic driving, and ensures the safety of the driver.

(2) By setting and using convolutional neural network to preprocess the original data and extract local features, the obtained original data can quickly remove impurities and interference factors, and obtain the required data features. Quickly extract and convert to corresponding feature vectors for subsequent operations, making the algorithm more efficient, the process more accurate, and the results provided clearer, making the subsequent planning trajectory more convenient and safe, and making the complex road conditions in urban areas more convenient and safe. The decision-making planning system is more practical and safe, which ensures the safety of drivers.

(3) Predict the trajectory by setting the prediction module, obtain the predicted trajectory, and evaluate the credibility of the predicted trajectory to obtain the third planning trajectory result, so that the trajectory generated by the predicted trajectory to assist the AI model can be judged and evaluated, combined with multiple The trajectory makes the final output trajectory more comprehensive and accurate, which is more in line with the current environment of the driving vehicle and protects the safety of the driver. Then, the third planning trajectory result and the second planning trajectory result are subjected to collision screening to obtain the fourth planning trajectory As a result, on the one hand, the overall solution efficiency of the algorithm is improved to avoid spending a lot of computing power on invalid trajectories for a complete feasibility check and comfort score; at the same time, a safety screening is also done. At the same time, the collisions of some low-confidence trajectory points may be ignored. The effective self-vehicle planning trajectory will not be deleted blindly because of some low-confidence predicted trajectories, and the scope and completeness of the algorithm solution space will be expanded.

(4) By setting the results of the first planned trajectory and the fourth planned trajectory to score comfort, anthropomorphism and efficiency, the output results can be more representative and practical after passing the three scores. It conforms to the driving conditions of the current driving vehicle, so that the trajectory can ensure the safe driving of the driving vehicle and the safety of the driver.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the accompanying drawings to be used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For Those of ordinary skill in the art can also obtain other drawings based on these drawings without making creative efforts.

FIG. 1 is a flowchart of a decision-making and planning system for urban complex road conditions provided by the present invention.

FIG. 2 is a schematic structural diagram of a decision planning device provided by the present invention.

Fig. 3 is a detailed flow chart of a decision-making planning system for urban complex road conditions provided by the present invention.

Fig. 4 is a schematic diagram of a multi-task unified AI model provided by the present invention.

FIG. 5 is a flow chart of trajectory scoring arbitration of a decision-making planning system for urban complex road conditions provided by the present invention.

Fig. 6 is a schematic diagram of the TrajectoryModel trajectory model provided by the present invention.

Explanation of reference signs:

100 is a decision planning device; 110 is an acquisition module; 120 is a processing module; 130 is a scoring module.

Detailed ways

In order to make the above objects, features and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than Full examples. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

Referring to Figures 1-6, the present invention provides a decision-making and planning system for complex road conditions in urban areas, including:

Step S100: Detect the driving parameters of the driving vehicle and the surrounding environment parameters, and obtain a plurality of original data;

Step S200: Import multiple raw data into the AI model for processing to obtain multiple perception results; calculate the planned trajectory through multiple perception results to obtain the first planned trajectory result; the AI model obtains the second planned trajectory result based on the multiple raw data ;

Step S300: Import the first planned trajectory result and the second planned trajectory result into the trajectory scoring arbitration module to verify the best trajectory, and output the best trajectory.

Specifically, based on the original sensor collection and input of driving parameters and surrounding environment parameters of the driving vehicle, the unified multi-task AI model proposed by the present invention is used for processing, and the perception results obtained specifically include: traffic participants such as vehicles, bicycles, and pedestrians, etc. Target detection results; target perception results such as lane lines, traffic signs, traffic lights, etc.; predicted behavior trajectories of traffic participants; positions of other stationary obstacles, etc. Then these results will be directly output to the traditional decision-making planning algorithm, and the corresponding planning trajectory calculation will be performed to obtain the first planning trajectory result. At the same time, the AI model will also directly output the planned trajectory result based on the data-driven method, that is, the second planned trajectory result. These two types of trajectories pass through a trajectory scoring arbitration module before final output.

Further, a unified AI model framework with multi-tasking functions: it can support multiple types of sensors (visual cameras, lidar, millimeter-wave radar, etc.), multi-channel sensors (such as forward-facing cameras, side-facing cameras, input to the rear camera, etc.). For the input of different sensors, the corresponding CNN (convolutional neural network) is used for preprocessing and local feature extraction. The feature vectors extracted from different sensors are then merged by a Concat operation. Then, a DNN (fully connected neural network) structure and a Transformer network structure are used as the middle layer to perform further feature transformation and encoding on the merged feature vector: the result of the middle layer of the entire AI model is obtained. The result of this intermediate layer can be followed by some typical task network blocks, such as prediction, target object detection and lane line segmentation, etc., to output the corresponding multi-task perception results; at the same time, the result of this intermediate layer can be directly output to LSTMDecoder The network generates a series of trajectory sequences for pre-generation of trajectory points for decision planning. Then, the pre-generated trajectory points are processed and output through the TrajectoryModel trajectory model proposed by the present invention. These modules above share the same intermediate layer results, maximizing the reuse of the trained AI network model.

Specifically, the present invention proposes a tree-like decision planning trajectory representation model, that is, the Trajectory Model trajectory model, which finally needs to be realized with a "tree" data structure. Considering the function, the tree decision planning trajectory model proposed by the present invention can better deal with the multimodal problem of decision planning. That is, in the same scene, the self-vehicle has multiple feasible solutions, and other vehicles will also have multiple possibilities for future driving behavior. The tree-based structure can fully capture the various possibilities of the trajectory of the own vehicle and the interaction with other vehicles. Finally, the module will convert the trajectory points generated by LSTM into tree-like trajectory for output to represent the various possibilities of the future state of the vehicle.

Further, the prediction results of the prediction module (i.e., a plurality of traffic participants around the vehicle, if 10, a credibility evaluation is performed. If the predicted trajectory, that is, the third planning trajectory result, is in the form of 10 [a, b,c,d,e,f] trajectory coordinate point sequence. Then this module will assign corresponding credibility discounts to each point in these trajectories, such as [(a,0.95),(b,0.9) ,(c,0.85),d(0.80),e(0.75),f(0.7)]. The predicted trajectory after the credibility evaluation will be collided with the tree-shaped trajectory model output by the AI model. In this step Mainly consider the possibility of collision between those predicted trajectory points with high confidence (such as above 0.8) and the tree trajectory of the ego vehicle. The trajectory tree with high confidence potential collision risk is directly pruned, that is, in Delete the corresponding branches in the tree structure. On the one hand, this improves the overall solution efficiency of the algorithm and avoids spending a lot of computing power on invalid trajectories for complete feasibility checks and comfort scores; at the same time, it also performs a security screening. At the same time, the collisions of some low-confidence (such as lower than 0.3) trajectory points may be ignored. The advantage of this step is that the effective self-vehicle planning trajectory will not be blindly deleted because of some low-confidence predicted trajectories. The scope and completeness of the algorithm solution space are expanded. Finally, the trajectory tree after collision screening, that is, the result of the fourth planning trajectory, and the traditional decision planning trajectory that has done complete collision detection, that is, the result of the first planning trajectory, pass through the follow-up The three scoring stages (comfort, anthropomorphism, and efficiency) of the three scoring stages, and finally select a single track sequence with the highest total score for output.

Preferably, the original data is obtained by setting the driving parameters of the detected vehicle and the surrounding environment parameters, so that the source of the data is more suitable for the actual situation of the current vehicle driving, and the next trajectory planning is carried out according to the actual situation, so that the trajectory is more in line with the current situation. The driving environment of the vehicle is processed by importing the original data into the AI model, and the perception result is obtained, and the traditional trajectory calculation is performed to obtain the first planning trajectory result, and then the AI model is processed based on the original data to obtain the second The result of the planned trajectory enables more calculation methods of the trajectory, and different trajectories are obtained through different methods, so that the practicability and safety of the trajectory are higher. At the same time, the two planned trajectory results are imported into the scoring arbitration module to verify that the best The trajectory minimizes the risk of the trajectory, and also improves the efficiency of the algorithm. It can more quickly obtain the optimal driving trajectory of the vehicle in the current driving environment, ensuring the safety of the driver and the driving vehicle. At the same time, through the AI model To process the data to obtain different trajectory models, so that the trajectory planning of the automatic driving system is more comprehensive, safe and complete. At the same time, the AI model can better respond to the current driving environment and make decisions, and the algorithm is more efficient. It enables the vehicle to safely drive automatically in a complex driving environment, improves the safety and practicability of automatic driving, and ensures the safety of the driver.

Specifically, multiple raw data are imported into the AI model for processing to obtain multiple perception results, which also includes: using convolutional neural networks to preprocess multiple raw data and extract local features to obtain multiple feature vectors.

Preferably, by setting and using a convolutional neural network to perform preprocessing and local feature extraction on the original data, the obtained original data can quickly remove impurities and interference factors, and obtain the required data features. Quickly extract and convert to corresponding feature vectors for subsequent operations, making the algorithm more efficient, the process more accurate, and the results provided clearer, making the subsequent planning trajectory more convenient and safe, and making the complex road conditions in urban areas more convenient and safe. The decision-making planning system is more practical and safe, which ensures the safety of drivers.

Specifically, the AI model obtains the second planning trajectory result based on multiple original data, which also includes: combining multiple feature vectors through splicing operations to obtain the combined feature vector; using a fully connected neural network structure and Transformer network structure as the middle layer , perform feature transformation and encoding on the merged feature vectors to obtain intermediate results.

Preferably, by setting and merging multiple feature vectors, the subsequent operations and processing are more convenient and quicker, and the efficiency of subsequent algorithms is faster, and then the combined feature vectors are characterized by a fully connected neural network structure and a Transformer network structure. Converting and coding to obtain intermediate results, so that the merged feature vectors can be used for subsequent processing operations, and the conversion process is more efficient. At the same time, the obtained intermediate results can be used directly, which is more convenient and makes the algorithm more efficient.

Specifically, the AI model obtains the second planning trajectory results based on multiple original data, which also includes: inputting the intermediate results into the LSTMDecoder network to generate a series of trajectory sequences, and pre-generating the decision planning trajectory points; processing the decision planning trajectory points through the trajectory model , the output is the result of the second planned trajectory.

Preferably, a series of trajectory sequences are generated by inputting the intermediate results into the LSTMDecoder network, so that the trajectory can be better predicted by the LSTMDecoder network. The trajectory is a set of continuous data information with time continuity and spatial randomness, and LSTMDecoder The network can predict and process it very well, making the trajectory prediction more accurate. The decision-making planning trajectory points are processed through the trajectory model, and the output is the second planning trajectory result, and the model combined with the LSTMDecoder network can make the model more accurate. , making subsequent predictions more accurate, making the decision-making planning system more comprehensive and safe, better ensuring the safety of drivers, and making autonomous driving more intelligent.

Specifically, the results of the first planned trajectory and the second planned trajectory are imported into the trajectory scoring arbitration module to verify the optimal trajectory, and output the optimal trajectory, which also includes: performing trajectory prediction through the prediction module to obtain the predicted trajectory, and performing a calculation on the predicted trajectory The reliability evaluation is performed to obtain the third planned trajectory result; the third planned trajectory result and the second planned trajectory result are subjected to collision screening to obtain the fourth planned trajectory result.

Preferably, the trajectory prediction is performed by setting the prediction module to obtain the predicted trajectory, and the predicted trajectory is evaluated for credibility to obtain the third planning trajectory result, so that the trajectory generated by the predicted trajectory to assist the AI model is judged and evaluated, combined with multiple The trajectory makes the final output trajectory more comprehensive and accurate, which is more in line with the current environment of the driving vehicle and protects the safety of the driver. Then, the third planning trajectory result and the second planning trajectory result are subjected to collision screening to obtain the fourth planning trajectory As a result, on the one hand, the overall solution efficiency of the algorithm is improved to avoid spending a lot of computing power on invalid trajectories for a complete feasibility check and comfort score; at the same time, a safety screening is also done. At the same time, the collisions of some low-confidence trajectory points may be ignored. The effective self-vehicle planning trajectory will not be deleted blindly because of some low-confidence predicted trajectories, and the scope and completeness of the algorithm solution space will be expanded.

Specifically, the results of the first planned trajectory and the fourth planned trajectory are scored for comfort, anthropomorphism, and efficiency; the resulting comfort, anthropomorphic, and efficiency scores are scored, and trajectories with high scores are output.

Preferably, by setting the first planned trajectory result and the fourth planned trajectory result to carry out comfort score, anthropomorphic score and efficiency score, the output results can be more representative and practical after passing the three scores, and more It conforms to the driving conditions of the current driving vehicle, so that the trajectory can ensure the safe driving of the driving vehicle and the safety of the driver.

Specifically, the comfort score further includes: comfort=normalized (1/track curvature)+normalized (1/acceleration rate).

Preferably, the trajectory is evaluated by setting the comfort level. The comfort level is mainly measured by the two indicators of the trajectory curvature change rate and the planning acceleration change rate: that is, under the premise of satisfying safety, the trajectory curvature change rate and acceleration change rate The lower the better, making higher-scoring trajectories safer and more practical.

Specifically, the efficiency score also includes: efficiency=normalization (total length of track).

Specifically, the efficiency of the trajectory can be measured by the total length of the planned trajectory, that is, when each trajectory planning is planned for the state of the vehicle in the future (say) 10 seconds, the longer the path, the higher the speed , The better the efficiency.

Preferably, the trajectory is scored by setting the efficiency, so that the longer the path, the higher the speed and the better the efficiency, so that the trajectory obtained through the efficiency score can better drive at the current intersection.

Specifically, the score of anthropomorphism also includes: anthropomorphism = normalization (1/Frescher distance (corresponding to the average trajectory of human drivers in the scene, target trajectory)).

Specifically, the anthropomorphic index measures the similarity between the current trajectory and the trajectory of a human driver. The closer to the human standard, the easier it is to be accepted by users and the higher the score. The average trajectory of human drivers in corresponding scenarios/working conditions.

Furthermore, when the Frescher distance of two curves is smaller, it can be understood that the similarity between the two curves is higher.

Specifically, the above index scores, namely the comfort score, anthropomorphism score and efficiency score, are all normalized, so that the final total score will not be greatly negatively affected by the difference in the calculation order of magnitude between different parameters.

Preferably, the trajectory is scored by setting the degree of anthropomorphism, so that the driving of the trajectory can be more in line with human standards, and at the same time more consistent with the driving habits of existing drivers, so as to ensure the best route in the current driving situation while driving. Higher degree of humanization to ensure driving safety.

The present invention also provides a decision-making and planning device 100, including: acquisition module 110: detecting driving parameters of the driving vehicle and surrounding environment parameters, and obtaining a plurality of original data; processing module 120: importing a plurality of original data into the AI model for processing, and obtaining Multiple perception results; planning trajectory calculation through multiple perception results to obtain the first planning trajectory result; the AI model obtains the second planning trajectory result based on multiple original data; scoring module 130: compare the first planning trajectory result and the second planning trajectory result The trajectory results are imported into the trajectory scoring arbitration module to verify the best trajectory and output the best trajectory.

Preferably, by setting the acquisition module 110 to obtain the driving road condition environment of the current vehicle, the original data is obtained, so that the trajectory planning of the vehicle is more in line with the current environmental reality, making the planning more practical, and the original data is processed by the processing module 120 After processing, the results of the first planned trajectory and the second planned trajectory are obtained, so that the obtained two trajectory results are more accurate and in line with the actual situation. At the same time, the evaluation of multiple trajectory results can make the final output trajectory more safe and practical. At the same time, the scoring module 130 is used to score and verify the two trajectories, so that the output home is the safest, fastest and most humanized trajectory. Through the cooperation of the three modules, the algorithm is more efficient and can take into account different The driving environment is more practical.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims (10)

1. A decision-making and planning system for complex road conditions in urban areas, characterized in that it comprises: Detect the driving parameters of the driving vehicle and the surrounding environment parameters to obtain multiple raw data; Importing the multiple raw data into the AI model for processing to obtain multiple perception results; Perform planning trajectory calculation through the plurality of perception results to obtain a first planning trajectory result; The AI model obtains a second planned trajectory result based on the plurality of raw data; Importing the first planned trajectory result and the second planned trajectory result into the trajectory scoring arbitration module to verify the best trajectory, and output the best trajectory. 2. The decision-making and planning system for complex road conditions in urban areas according to claim 1, wherein said importing said plurality of raw data into an AI model for processing to obtain a plurality of perception results also includes: A convolutional neural network is used to perform preprocessing and local feature extraction on the plurality of original data to obtain a plurality of feature vectors. 3. The decision-making and planning system for urban complex road conditions according to claim 2, wherein the AI model obtains the second planning track result based on the plurality of raw data, and also includes: Merging the plurality of feature vectors through a splicing operation to obtain a merged feature vector; A fully connected neural network structure and a Transformer network structure are used as an intermediate layer to perform feature conversion and encoding on the merged feature vector to obtain an intermediate result. 4. The decision-making and planning system for complex road conditions in urban areas according to claim 3, wherein the AI model obtains the second planning trajectory result based on the plurality of raw data, and also includes: The intermediate result is input to the LSTMDecoder network to generate a series of trajectory sequences, and pre-generated decision planning trajectory points; The decision-planning trajectory points are processed by a trajectory model, and output as a second planning trajectory result. 5. The decision-making planning system for complex road conditions in urban areas according to claim 4, wherein said importing said first planned trajectory result and said second planned trajectory result into a trajectory scoring arbitration module to verify the best trajectory, And output the best trajectory, also includes: Predicting the trajectory through the prediction module to obtain a predicted trajectory, and evaluating the credibility of the predicted trajectory to obtain a third planning trajectory result; Perform collision screening on the third planned trajectory result and the second planned trajectory result to obtain a fourth planned trajectory result. 6. the decision-making planning system for urban complex road conditions according to claim 5, characterized in that, Carrying out comfort scoring, anthropomorphic scoring and efficiency scoring on the first planned trajectory result and the fourth planned trajectory result; As a result of the comfort score, the anthropomorphic score, and the efficiency score, tracks with high scores are output. 7. The decision-making and planning system for urban complex road conditions according to claim 6, wherein the comfort score comprises: Comfort = normalized (1/track curvature) + normalized (1/acceleration rate). 8. The decision-making and planning system for urban complex road conditions according to claim 6, wherein the efficiency score comprises: Efficiency = normalized (total track length). 9. The decision-making and planning system for urban complex road conditions according to claim 6, wherein the anthropomorphic scoring comprises: Anthropomorphism = normalization (1/Fresher distance (corresponding to the average trajectory of human drivers in the scene, target trajectory)). 10. A decision planning device, characterized in that it comprises: Acquisition module: detect the driving parameters of the driving vehicle and the surrounding environment parameters, and obtain multiple raw data; Processing module: import the plurality of raw data into the AI model for processing to obtain a plurality of perception results; calculate the planned trajectory through the plurality of perception results to obtain the first planned trajectory result; the AI model is based on the plurality of The raw data obtains the second planning trajectory result; Scoring module: Import the first planned trajectory result and the second planned trajectory result into the trajectory scoring arbitration module to verify the best trajectory, and output the best trajectory.

Images