Autonomous navigation control method and system for unmanned vehicle

Abstract

The present invention provides an autonomous navigation control method and system for an unmanned vehicle, which relates to the field of navigation control technology, including sensing the surrounding environment of the vehicle through multi-sensor fusion, establishing a semantic map model, extracting driving information of the unmanned vehicle, performing dimensionality reduction processing on the driving information of the unmanned vehicle, constructing a Markov decision model based on low-dimensional state characteristics, and obtaining an initial behavior strategy; using a trajectory planning algorithm to generate local trajectory segments, and using an adaptive step size to discretize the local trajectory segments, generating a candidate trajectory set, calculating the expected cost of each candidate trajectory in each scenario, and selecting the candidate trajectory with the smallest expected cost as the optimal driving strategy; obtaining the current speed of the vehicle and the curvature information of the road ahead according to the optimal driving strategy, adjusting the preview distance in real time, obtaining the predicted time domain length, generating the optimal control instruction within the predicted time domain length according to a preset model prediction controller, and performing autonomous navigation control on the unmanned vehicle through the optimal control instruction.

Description

Technical Field

[0001] The present invention relates to navigation control technology, and in particular to an autonomous navigation control method and system for an unmanned vehicle. Background Art

[0002] With the rapid development of artificial intelligence and autonomous driving technology, unmanned vehicles have received more and more attention. Unmanned vehicles can autonomously perceive, make decisions and control in complex traffic environments, which is expected to greatly improve traffic efficiency and travel safety. Traditional unmanned vehicle navigation and control methods generally predefine a series of rules, judge the current state and select the corresponding control strategy based on the environmental information collected by the vehicle sensors. This method has achieved autonomous navigation and control of unmanned vehicles to a certain extent, but there are still problems such as poor decision-making adaptability, complex environmental modeling, and untimely response to dynamic changes, which makes it difficult to meet the requirements of safety, real-time and generalization when facing complex traffic scenes. Therefore, there is an urgent need for a more intelligent, efficient and robust unmanned vehicle autonomous navigation and control method to adapt to the ever-changing traffic environment and improve the autonomous decision-making and control capabilities of unmanned vehicles. Summary of the invention

[0003] The embodiments of the present invention provide an autonomous navigation control method and system for an unmanned vehicle, which can solve the problems in the prior art.

[0004] According to a first aspect of the embodiments of the present invention,

[0005] Provided is an autonomous navigation control method for an unmanned vehicle, comprising:

[0006] The vehicle surrounding environment is perceived through multi-sensor fusion, a semantic map model is established, the driving information of the unmanned vehicle is extracted using the semantic map model, the driving information of the unmanned vehicle is reduced in dimension using the principal component analysis method to obtain low-dimensional state features, a Markov decision model is constructed based on the low-dimensional state features, and an initial behavior strategy is obtained by solving the Markov decision model;

[0007] Based on the initial behavior strategy, combined with the road topology and vehicle dynamics model, a trajectory planning algorithm is used to generate local trajectory segments. At the same time, an adaptive step size is used to discretize the local trajectory segments to generate a candidate trajectory set. For the candidate trajectory set, a scenario-based evaluation mechanism is introduced to calculate the expected cost of each candidate trajectory in each scenario, and the candidate trajectory with the smallest expected cost is selected as the optimal driving strategy.

[0008] According to the optimal driving strategy, the current speed of the vehicle and the curvature of the road ahead are obtained. The preview distance is adjusted in real time according to the changes in the current speed of the vehicle and the curvature of the road ahead to obtain the predicted time domain length. According to the preset model prediction controller, the optimal control instructions within the predicted time domain length are generated, and the unmanned vehicle is autonomously navigated and controlled through the optimal control instructions.

[0009] In an optional embodiment,

[0010] Based on the initial behavior strategy, combined with the road topology and vehicle dynamics model, the trajectory planning algorithm is used to generate local trajectory segments. At the same time, the adaptive step size is used to discretize the local trajectory segments. The candidate trajectory set generated includes:

[0011] Decoding the initial behavior strategy into a motion constraint set, the motion constraint set including lateral constraints and longitudinal constraints, optimizing the motion constraint set in combination with a road topology structure, the road topology structure including road curvature, lane width, and intersection location, and performing a feasibility check on the motion constraint set using a pre-built vehicle dynamics model to obtain feasible motion constraints;

[0012] A trajectory planning algorithm using layered spatiotemporal sampling is used to generate motion primitives covering the achievable motion space of the unmanned vehicle. The layered spatiotemporal sampling divides the sampling space into a lateral subspace and a longitudinal subspace. The lateral motion primitives are generated by sampling in the lateral subspace, and the longitudinal motion primitives are generated by sampling in the longitudinal subspace.

[0013] The lateral motion primitive and the longitudinal motion primitive are combined to generate a local trajectory segment that satisfies the feasible motion constraint, the trajectory discrete step length is adaptively adjusted according to the road curvature and the obstacle density, the local trajectory segment is discretized according to the obtained discrete step length, and a candidate trajectory set is generated.

[0014] In an optional embodiment,

[0015] Optimizing the motion constraint set in combination with the road topology structure includes:

[0016] The cost terms of road curvature, lane width and intersection position in the road topology are calculated respectively, and the comprehensive objective function of motion constraint is constructed by combining the road curvature cost term, lane width cost term and intersection position cost term. The calculation formula is as follows:

[0017] ;

[0018] Among them, J represents the comprehensive objective function, T represents the time domain range, ω1 represents the weight coefficient of the road curvature cost term, α1 represents the balance factor of the curvature size, η represents the road curvature, s represents the path arc length, t represents the time, α2 represents the balance factor of the curvature change, represents the rate of change of road curvature, ω2 represents the weight coefficient of the lane width cost term, β1 represents the amplification coefficient of the exponential cost term from the lane boundary, β2 represents the attenuation coefficient related to the exponential cost term from the lane boundary, d(t) represents the distance from the vehicle to the nearest lane boundary at time t, ω3 represents the weight coefficient of the intersection position cost term, C int represents the number of intersections, i represents the i-th intersection, γ1 represents the amplification coefficient related to the exponential cost term of the distance intersection, γ2 represents the attenuation coefficient related to the exponential cost term of the distance intersection, Represents the distance from the vehicle to the i-th intersection at time t.

[0019] In an optional embodiment,

[0020] For the candidate trajectory set, a scenario-based evaluation mechanism is introduced to calculate the expected cost of each candidate trajectory in each scenario, and the candidate trajectory with the smallest expected cost is selected as the optimal driving strategy, including:

[0021] Determine multiple driving scenarios corresponding to each candidate trajectory in the candidate trajectory set and construct a scenario set, and use the scenario set to segment the candidate trajectory, each segment of the candidate trajectory corresponds to a driving scenario;

[0022] The risk metric of each candidate trajectory in the corresponding driving scenario is calculated using the trained risk assessment model to obtain the trajectory segment risk sequence. The scenario transition probability is introduced to construct the transition relationship of driving scenarios between different candidate trajectories. The scenario transition probability is used to map the trajectory segment risk sequence into the scenario path risk distribution.

[0023] Based on the scenario path risk distribution, trajectory evaluation indicators and cost functions are designed to obtain a comprehensive trajectory evaluation system for measuring candidate trajectories in different driving scenarios. The comprehensive trajectory evaluation system is used to calculate the cost of each candidate trajectory in each driving scenario. The comprehensive cost of the trajectory is obtained by combining the costs under different driving scenarios, and the candidate trajectory with the smallest comprehensive cost is selected as the optimal driving strategy.

[0024] In an optional embodiment,

[0025] The calculation formula of trajectory evaluation index is as follows:

[0026] ;

[0027] Among them, G represents the trajectory evaluation index, M represents the number of scenarios, and δ b represents the weight coefficient of the b-th scenario, K represents the number of trajectory segments, a represents the a-th trajectory, j represents the j-th segment, and r a，j represents the risk measure of the jth segment of the ath trajectory, and P represents the scenario transition probability.

[0028] In an optional embodiment,

[0029] The optimal control instructions within the predicted time domain length are generated according to the preset model predictive controller, and the autonomous navigation control of the unmanned vehicle is performed through the optimal control instructions, including:

[0030] Predicting the vehicle's trajectory tracking error within the prediction time domain length by using a pre-built vehicle dynamics model and a trajectory tracking error model, wherein the trajectory tracking error includes a lateral position error, a heading angle error, and a longitudinal velocity error;

[0031] The vehicle's trajectory tracking error within the prediction time domain length is used as the input of the model predictive controller. The control variable change rate is introduced, and the controller multi-objective function is constructed in combination with the trajectory tracking error. The numerical optimization algorithm is used to solve the controller multi-objective function, and the optimal control instructions within the prediction time domain length are generated. The unmanned vehicle is autonomously navigated and controlled through the optimal control instructions.

[0032] In an optional embodiment,

[0033] The calculation formula of the controller multi-objective function is as follows:

[0034] ;

[0035] Where M represents the controller multi-objective function, N p represents the prediction time domain length, represents the error prediction value, k represents the current time, i represents the offset, Q represents the weight coefficient of the error prediction value, N c represents the control time domain length, u represents the control quantity sequence, △u(·) represents the control quantity change rate, R represents the weight coefficient of the control quantity change rate, and S represents the weight coefficient of the control quantity.

[0036] According to a second aspect of the embodiments of the present invention,

[0037] Provided is an autonomous navigation control system for an unmanned vehicle, comprising:

[0038] Unit 1: Use multi-sensor fusion to perceive the surrounding environment of the vehicle, establish a semantic map model, use the semantic map model to extract the driving information of the unmanned vehicle, use the principal component analysis method to reduce the dimension of the driving information of the unmanned vehicle to obtain low-dimensional state features, build a Markov decision model based on the low-dimensional state features, and obtain the initial behavior strategy by solving the Markov decision model;

[0039] The second unit is used to generate local trajectory segments based on the initial behavior strategy, combined with the road topology and vehicle dynamics model, using the trajectory planning algorithm, and using the adaptive step size to discretize the local trajectory segments to generate a candidate trajectory set. For the candidate trajectory set, a scenario-based evaluation mechanism is introduced to calculate the expected cost of each candidate trajectory in each scenario, and the candidate trajectory with the smallest expected cost is selected as the optimal driving strategy;

[0040] The third unit is used to obtain the current speed of the vehicle and the curvature information of the road ahead according to the optimal driving strategy, adjust the preview distance in real time according to the changes in the current speed of the vehicle and the curvature information of the road ahead, obtain the predicted time domain length, generate the optimal control instructions within the predicted time domain length according to the preset model prediction controller, and perform autonomous navigation control on the unmanned vehicle through the optimal control instructions.

[0041] According to a third aspect of the embodiments of the present invention,

[0042] An electronic device is provided, comprising:

[0043] processor;

[0044] a memory for storing processor-executable instructions;

[0045] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0046] A fourth aspect of the embodiments of the present invention is:

[0047] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0048] In this embodiment, a semantic map model is established through multi-sensor fusion technology, which can fully perceive and describe the environmental information around the unmanned vehicle, including lane lines, obstacles, traffic rules, etc., and provide high-quality environmental perception data for subsequent decision-making and planning. The principal component analysis method is used to reduce the dimension of vehicle driving information and extract low-dimensional state features, which can concisely and efficiently describe the current state of the vehicle, laying the foundation for behavioral strategy planning based on the Markov decision process. By constructing a Markov decision model and solving the initial behavior strategy, decisions can be made at an abstract level to generate high-level behavioral intentions, such as keeping lanes, changing lanes, etc., to guide downstream trajectory planning and control modules. Based on the initial behavior strategy and road topology information, a trajectory planning algorithm with spatiotemporal sampling is used to generate a set of candidate trajectories that meet the constraints, and a scenario evaluation mechanism is introduced to select the optimal driving trajectory. The trajectory has high safety and comfort. The preview distance is adjusted in real time according to the vehicle speed and road conditions, and the optimal control sequence is generated in combination with the model predictive controller. This can accurately control the vehicle to travel along the planned trajectory, greatly improving the trajectory tracking accuracy and stability. At the same time, it can realize high-precision autonomous navigation control of unmanned vehicles in complex environments, and has good safety, comfort and robustness, which can provide strong support for the practical application of unmanned driving technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of a flow chart of an autonomous navigation control method for an unmanned vehicle according to an embodiment of the present invention;

[0050] Figure 2 The figure is a schematic diagram of the structure of an autonomous navigation control system for an unmanned vehicle according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0053] Figure 1 FIG. 1 is a flow chart of an autonomous navigation control method for an unmanned vehicle according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0054] S101. The vehicle's surrounding environment is perceived through multi-sensor fusion, and a semantic map model is established. The driving information of the unmanned vehicle is extracted using the semantic map model. The driving information of the unmanned vehicle is reduced in dimension using the principal component analysis method to obtain low-dimensional state features. A Markov decision model is constructed based on the low-dimensional state features. The initial behavior strategy is obtained by solving the Markov decision model.

[0055] For example, multiple sensors such as LiDAR, millimeter-wave radar, and visual cameras are selected to achieve spatiotemporal registration and fusion of different sensor data through steps such as data synchronization, coordinate system 1, and feature matching. On the basis of multi-sensor fusion, a semantic map model is further constructed. The semantic map contains not only the geometric information of objects in the environment, but also their semantic attributes and scene semantic relationships.

[0056] The constructed semantic map model is used to extract key information that affects the driving decisions of the unmanned vehicle, including the vehicle's position, speed, acceleration, heading angle and other motion state information, as well as the vehicle's size, dynamic parameters and other attribute information, the road type, road width, road curvature and other geometric parameters around the vehicle, as well as semantic information such as road speed limits and traffic signs, as well as the position, speed, acceleration and other state information of obstacles around the vehicle, and the type, size and other attribute information of the obstacles.

[0057] The extracted unmanned vehicle driving information constitutes a high-dimensional state space, and then the driving information is reduced in dimension. Specifically, the principal component analysis (PCA) method is used to reduce the dimension of the unmanned vehicle driving information. PCA projects the original high-dimensional data into a set of orthogonal low-dimensional feature spaces through linear transformation, while retaining the main features of the data, it removes redundant and noise information.

[0058] After obtaining the low-dimensional state features, this paper uses the Markov decision process (MDP) to model the decision optimization problem of the unmanned vehicle. MDP consists of state space, action space, state transition probability and reward function. Based on the MDP model, the decision problem of the unmanned vehicle can be transformed into the problem of selecting the optimal action to maximize the long-term cumulative reward under a given state.

[0059] The decision optimization algorithm combining temporal difference (TD) learning and Monte Carlo tree search (MCTS) can be used to solve the MDP model and obtain the optimal behavior strategy. Specifically, the Q function estimate obtained by TD learning is used as the heuristic information of MCTS to guide the selection and expansion of tree search. At the same time, the simulation results of MCTS can be used as training samples for TD learning to update the estimate of the Q function. By iteratively alternating TD learning and MCTS, the quality of the strategy can be continuously improved to obtain the initial behavior strategy.

[0060] In this embodiment, the multi-sensor fusion technology can be used to obtain comprehensive and accurate information about the vehicle's surroundings, including the geometry, semantic attributes, and relationships of objects, and provide high-quality environmental perception data support for unmanned vehicle decision-making and planning. The original high-dimensional vehicle driving information is reduced in dimension through the principal component analysis algorithm, and low-dimensional state features are extracted, which can concisely and efficiently describe the current state and reduce the computational complexity of subsequent decisions. Modeling the decision-making of unmanned vehicles as a Markov decision process can achieve optimal decision-making within a theoretical framework and generate reasonable high-level behavior strategies. Combining the optimization algorithm of temporal difference learning and Monte Carlo tree search can efficiently solve complex Markov decision models and obtain near-optimal initial behavior strategies, which can provide high-quality and highly robust decision support for unmanned driving in complex environments.

[0061] S102. Based on the initial behavior strategy, combined with the road topology and vehicle dynamics model, a trajectory planning algorithm is used to generate local trajectory segments. At the same time, an adaptive step size is used to discretize the local trajectory segments to generate a candidate trajectory set. For the candidate trajectory set, a scenario-based evaluation mechanism is introduced to calculate the expected cost of each candidate trajectory in each scenario, and the candidate trajectory with the smallest expected cost is selected as the optimal driving strategy.

[0062] Among them, the trajectory planning algorithm takes the initial behavior decision sequence as input, and generates a motion primitive library covering the achievable motion range of the unmanned vehicle by sampling the motion space of the unmanned vehicle in a hierarchical and adaptive step size.

[0063] In an optional embodiment,

[0064] Based on the initial behavior strategy, combined with the road topology and vehicle dynamics model, the trajectory planning algorithm is used to generate local trajectory segments. At the same time, the adaptive step size is used to discretize the local trajectory segments. The candidate trajectory set generated includes:

[0065] Decoding the initial behavior strategy into a motion constraint set, the motion constraint set including lateral constraints and longitudinal constraints, optimizing the motion constraint set in combination with a road topology structure, the road topology structure including road curvature, lane width, and intersection location, and performing a feasibility check on the motion constraint set using a pre-built vehicle dynamics model to obtain feasible motion constraints;

[0066] A trajectory planning algorithm using layered spatiotemporal sampling is used to generate motion primitives covering the achievable motion space of the unmanned vehicle. The layered spatiotemporal sampling divides the sampling space into a lateral subspace and a longitudinal subspace. The lateral motion primitives are generated by sampling in the lateral subspace, and the longitudinal motion primitives are generated by sampling in the longitudinal subspace.

[0067] The lateral motion primitive and the longitudinal motion primitive are combined to generate a local trajectory segment that satisfies the feasible motion constraint, the trajectory discrete step length is adaptively adjusted according to the road curvature and the obstacle density, the local trajectory segment is discretized according to the obtained discrete step length, and a candidate trajectory set is generated.

[0068] Exemplarily, the initial behavior strategy is decoded into motion constraints, where the initial behavior strategy includes operations such as lane keeping, lane changing, and deceleration, and information such as target speed, acceleration, and steering angle in the strategy is extracted to construct a motion constraint set. The motion constraint set includes lateral constraints and longitudinal constraints, where the lateral constraints limit the lateral motion range of the vehicle, and the longitudinal constraints limit the speed and acceleration range of the vehicle.

[0069] Combined with the road topology, the path topology information between the vehicle's current position and the target position is obtained, including road curvature, lane width, intersection location, etc. The topological information is used to optimize the motion constraints and adjust the constraint boundaries to ensure that the generated trajectory meets the road geometry constraints.

[0070] A vehicle dynamics model is established, taking into account parameters such as vehicle mass, moment of inertia, and tire cornering stiffness, and motion constraints are mapped to the vehicle control space. The vehicle dynamics model is used to check whether the optimized motion constraint set is feasible under the current vehicle state. If not, the constraints are appropriately relaxed and re-optimized until a feasible motion constraint set is obtained.

[0071] The sampling space is divided into a transverse subspace and a longitudinal subspace. The transverse subspace includes transverse displacement, transverse velocity, etc., and the longitudinal subspace includes longitudinal displacement, longitudinal velocity, etc.

[0072] The motion space of the unmanned vehicle is sampled in layers in space-time, dividing the motion space into three dimensions: longitudinal, lateral, and time. The longitudinal dimension sampling generates longitudinal motion primitives with different speeds and accelerations, the lateral dimension sampling generates lateral motion primitives with different lateral offsets and curvatures, and the time dimension sampling generates trajectory fragments of different durations.

[0073] The sampling results of the three dimensions of longitudinal, lateral and time are arranged and combined to generate local trajectory segments that meet the feasible motion constraints. The local trajectory segments are discretized using adaptive step sizes, and the trajectory discrete step sizes are adaptively adjusted according to the road curvature and obstacle density, thereby improving the sampling efficiency while ensuring the smoothness of the trajectory.

[0074] Combine local trajectory segments into a global trajectory to form a complete trajectory from the current position to the target position. Smooth the global trajectory to eliminate sharp changes in the trajectory and improve the comfort of the trajectory. Use the cost function to evaluate the candidate trajectories and select the trajectory with the minimum comprehensive cost as the optimal trajectory.

[0075] In an optional embodiment,

[0076] Optimizing the motion constraint set in combination with the road topology structure includes:

[0077] The cost terms of road curvature, lane width and intersection position in the road topology are calculated respectively, and the comprehensive objective function of motion constraint is constructed by combining the road curvature cost term, lane width cost term and intersection position cost term. The calculation formula is as follows:

[0078] ;

[0079] Among them, J represents the comprehensive objective function, T represents the time domain range, ω1 represents the weight coefficient of the road curvature cost term, α1 represents the balance factor of the curvature size, η represents the road curvature, s represents the path arc length, t represents the time, α2 represents the balance factor of the curvature change, represents the rate of change of road curvature, ω2 represents the weight coefficient of the lane width cost term, β1 represents the amplification coefficient of the exponential cost term from the lane boundary, β2 represents the attenuation coefficient related to the exponential cost term from the lane boundary, d(t) represents the distance from the vehicle to the nearest lane boundary at time t, ω3 represents the weight coefficient of the intersection position cost term, C int represents the number of intersections, i represents the i-th intersection, γ1 represents the amplification coefficient related to the exponential cost term of the distance intersection, γ2 represents the attenuation coefficient related to the exponential cost term of the distance intersection, Represents the distance from the vehicle to the i-th intersection at time t.

[0080] In this embodiment, by decoding the initial behavior decision into lateral and longitudinal motion constraints, and integrating the road topology and vehicle dynamics model, it is ensured that the generated trajectory meets various constraints, thereby improving the feasibility and safety of the trajectory. By adopting the layered spatiotemporal sampling method, the combined sampling of the lateral and longitudinal subspaces can efficiently cover the entire reachable motion space of the unmanned vehicle and avoid missing feasible solutions. By constructing a motion primitive library and generating local trajectory segments in a combined manner, a variety of trajectory samples can be generated to provide sufficient alternatives for subsequent selection. The discrete step length of the trajectory is dynamically adjusted according to the road curvature and obstacle density, and the sampling is denser in sections with large curvature or dense obstacles, and the sampling is sparser in straight or open sections, which improves the sampling efficiency while ensuring the necessary accuracy. By decomposing the planning process into modules such as behavior decision, constraint fusion, sampling trajectory and optimization, each module is relatively independent, computationally efficient and can meet the needs of online real-time trajectory planning of unmanned vehicles in complex urban road environments, providing high-quality trajectory input for the final decision of the unmanned driving system.

[0081] In an optional embodiment,

[0082] For the candidate trajectory set, a scenario-based evaluation mechanism is introduced to calculate the expected cost of each candidate trajectory in each scenario, and the candidate trajectory with the smallest expected cost is selected as the optimal driving strategy, including:

[0083] Determine multiple driving scenarios corresponding to each candidate trajectory in the candidate trajectory set and construct a scenario set, and use the scenario set to segment the candidate trajectory, each segment of the candidate trajectory corresponds to a driving scenario;

[0084] The risk metric of each candidate trajectory in the corresponding driving scenario is calculated using the trained risk assessment model to obtain the trajectory segment risk sequence. The scenario transition probability is introduced to construct the transition relationship of driving scenarios between different candidate trajectories. The scenario transition probability is used to map the trajectory segment risk sequence into the scenario path risk distribution.

[0085] Based on the scenario path risk distribution, trajectory evaluation indicators and cost functions are designed to obtain a comprehensive trajectory evaluation system for measuring candidate trajectories in different driving scenarios. The comprehensive trajectory evaluation system is used to calculate the cost of each candidate trajectory in each driving scenario. The comprehensive cost of the trajectory is obtained by combining the costs under different driving scenarios, and the candidate trajectory with the smallest comprehensive cost is selected as the optimal driving strategy.

[0086] Exemplarily, for each trajectory in the candidate trajectory set, the driving scenarios that may be encountered during the execution of the trajectory are predicted, where the driving scenario is a set of parameters that describe the environment around the unmanned vehicle, including lane lines, obstacle positions and motion states, traffic rules, etc. For each candidate trajectory, it is divided into multiple driving scenarios according to its relative position relationship with the obstacles, and a scenario set is constructed. For each driving scenario, the feature vector related to risk assessment is extracted, and the risk assessment model is trained using a deep learning method. The driving scenarios and trajectory segments are input into the trained risk assessment model, and the risk measurement of each candidate trajectory segment in the corresponding driving scenario is output to construct a trajectory segment risk sequence.

[0087] The scenario transition probability is defined by means of data statistics or modeling, that is, the probability of transitioning to another scenario under a certain driving scenario. Using the scenario transition probability, the trajectory segment risk sequence is mapped to the scenario path risk distribution, where the scenario path risk distribution represents the probability distribution of the corresponding risk value when driving along the candidate trajectory.

[0088] Design multiple evaluation indicators, such as expected risk, risk variance, probability of dangerous events, etc., to describe the different statistical characteristics of the scenario path risk distribution. Construct a comprehensive cost function, the value of which reflects the comprehensive risk level of the candidate trajectory in various scenarios.

[0089] Based on the trajectory evaluation index and cost function, a comprehensive trajectory evaluation system is obtained to measure the candidate trajectories in different driving scenarios. The comprehensive trajectory evaluation system is used to calculate the cost of each candidate trajectory in each driving scenario. The comprehensive cost of the trajectory is obtained by combining the costs under different driving scenarios, and the candidate trajectory with the smallest comprehensive cost is selected as the optimal driving strategy.

[0090] In an optional embodiment,

[0091] The calculation formula of trajectory evaluation index is as follows:

[0092] ;

[0093] Among them, G represents the trajectory evaluation index, M represents the number of scenarios, and δ b represents the weight coefficient of the b-th scenario, K represents the number of trajectory segments, a represents the a-th trajectory, j represents the j-th segment, and r a，j represents the risk measure of the jth segment of the ath trajectory, and P represents the scenario transition probability.

[0094] In this embodiment, by introducing the concept of scenario, the candidate trajectories are segmented according to the driving scenario, and the risk and cost of the trajectory are evaluated for each scenario, which is more in line with the real scenario and the evaluation is more objective and accurate. By constructing a scenario transfer probability model, the trajectory segment risk is mapped to the scenario path risk distribution, which reflects the dynamic time-varying characteristics of the scenario and avoids the limitations of static evaluation. The trajectory segment risk is calculated using a trained risk assessment model. The risk assessment model is trained by machine learning technology and has strong generalization ability, and the risk assessment is more accurate and reliable. By designing a cost function that comprehensively considers multiple evaluation indicators such as risk, comfort, and efficiency, the evaluation is more comprehensive and can weigh multiple optimization goals. The evaluation process of segmented evaluation and comprehensive ranking is efficient and concise. Compared with the simultaneous evaluation of all trajectory scenario combinations, the calculation efficiency is greatly improved, and the evaluation results are more reliable and stable. This trajectory optimization scheme based on scenario evaluation breaks through the limitations of traditional single static evaluation and proposes a new scenario-based dynamic evaluation paradigm. By introducing technologies such as scenario segmentation, scenario transfer, and comprehensive evaluation system, it can more accurately, comprehensively, and efficiently evaluate and optimize candidate trajectories in complex situations, providing strong support for high-quality decision-making of unmanned vehicles.

[0095] S103. Obtain the current speed of the vehicle and the curvature information of the road ahead according to the optimal driving strategy, adjust the preview distance in real time according to the changes in the current speed of the vehicle and the curvature information of the road ahead, obtain the predicted time domain length, generate the optimal control instructions within the predicted time domain length according to the preset model prediction controller, and perform autonomous navigation control on the unmanned vehicle through the optimal control instructions.

[0096] Among them, the preview distance is a key parameter that affects the trajectory tracking performance. A reasonable preview distance can improve the accuracy and stability of trajectory tracking, while an inappropriate preview distance will cause trajectory deviation or system oscillation. The main factors affecting the selection of the preview distance include first, vehicle speed. The higher the speed, the larger the preview distance should be to provide sufficient advance prediction and response time; second, road curvature. The greater the road curvature, the smaller the preview distance should be to avoid trajectory deviation caused by excessive preview distance; third, vehicle dynamics characteristics. Different vehicles have different dynamic response characteristics, and the preview distance needs to be adjusted according to the vehicle characteristics; fourth, control system bandwidth. The response speed and stability of the control system will also affect the selection of the preview distance.

[0097] In view of the above-mentioned influencing factors, real-time adaptive adjustment of the preview distance is realized through an adaptive preview distance adjustment algorithm. Specifically, the benchmark preview distance is calculated according to the current vehicle speed and the vehicle speed adjustment coefficient, and the curvature adjustment factor is calculated according to the curvature of the road ahead. The vehicle speed adjustment coefficient and the curvature adjustment factor are multiplied to obtain a comprehensive adjustment factor. Finally, the benchmark preview distance is multiplied by the comprehensive adjustment factor to obtain an adaptive preview distance, and the adaptive preview distance is used as the prediction time domain length.

[0098] In an optional embodiment,

[0099] The optimal control instructions within the predicted time domain length are generated according to the preset model predictive controller, and the autonomous navigation control of the unmanned vehicle is performed through the optimal control instructions, including:

[0100] Predicting the vehicle's trajectory tracking error within the prediction time domain length by using a pre-built vehicle dynamics model and a trajectory tracking error model, wherein the trajectory tracking error includes a lateral position error, a heading angle error, and a longitudinal velocity error;

[0101] The vehicle's trajectory tracking error within the prediction time domain length is used as the input of the model predictive controller. The control variable change rate is introduced, and the controller multi-objective function is constructed in combination with the trajectory tracking error. The numerical optimization algorithm is used to solve the controller multi-objective function, and the optimal control instructions within the prediction time domain length are generated. The unmanned vehicle is autonomously navigated and controlled through the optimal control instructions.

[0102] Exemplarily, to realize model predictive control, it is first necessary to establish a vehicle dynamics model and a trajectory tracking error model, wherein the vehicle dynamics model includes a lateral motion model and a longitudinal motion model. By establishing a lateral motion differential equation such as a two-wheel model or a unit mass model, the lateral motion dynamics characteristics of the vehicle are described, and a longitudinal motion differential equation is constructed to describe the longitudinal motion dynamics of the vehicle, including acceleration, velocity, etc.

[0103] The trajectory tracking error state quantities such as lateral position error, heading angle error and longitudinal velocity error are defined, and an error model describing the trajectory tracking error dynamics is established using the vehicle dynamics model and the error state quantity definitions.

[0104] The constructed vehicle dynamics model and trajectory tracking error model are used to predict the lateral position error, heading angle error and longitudinal velocity error within the prediction time domain length.

[0105] The vehicle's trajectory tracking error within the prediction time domain length is used as the input of the model predictive controller. The control variable change rate is introduced, and the multi-objective function of the controller is constructed in combination with the trajectory tracking error. Multiple objectives are combined in the form of weighted sum or product, and the importance of different objectives is balanced by adjusting the weight coefficient.

[0106] Optionally, robustness constraints are introduced in the design of the model predictive controller, and the robustness constraints include using state feedback to compensate for modeling errors and external disturbances, describing the uncertainty range of vehicle model parameters and incorporating it into constraints, and introducing closed-loop control system robustness evaluation indicators as constraints.

[0107] The multi-objective function is solved by numerical optimization algorithms (such as quadratic programming, gradient descent, etc.) to obtain the optimal control sequence. At each sampling moment, the first optimal control instruction is applied to control the vehicle.

[0108] In an optional embodiment,

[0109] The calculation formula of the controller multi-objective function is as follows:

[0110] ;

[0111] Where M represents the controller multi-objective function, N p represents the prediction time domain length, represents the error prediction value, k represents the current time, i represents the offset, Q represents the weight coefficient of the error prediction value, N c represents the control time domain length, u represents the control quantity sequence, △u(·) represents the control quantity change rate, R represents the weight coefficient of the control quantity change rate, and S represents the weight coefficient of the control quantity.

[0112] In this embodiment, adaptive adjustment of the preview distance is achieved by comprehensively considering multiple influencing factors such as vehicle speed and road curvature. Appropriately increase the preview distance of the straight high-speed section to enhance the forward-looking nature of the planning; appropriately reduce the preview distance of the curved low-speed section to improve the immediate response capability. It is conducive to the unmanned vehicle to maintain good trajectory tracking performance under different working conditions. Through the vehicle dynamics model and the trajectory tracking error model, the trajectory tracking error in the future time domain can be accurately predicted, including the lateral position, heading angle and longitudinal speed errors, to provide high-quality trajectory error feedback for the controller. The controller integrates multiple objectives such as trajectory tracking error and control quantity change rate into a single optimization objective function, and solves it through a numerical optimization algorithm. The generated control instructions can balance tracking accuracy and control smoothness to achieve multi-objective optimization. The model predictive controller combines the feedforward effect of the prediction model with the feedback effect of the trajectory tracking error, and can correct the accumulated error in time to improve the accuracy and stability of the control. The controller takes the optimal driving strategy given by the trajectory planning as input, and passes the calculated optimal control instructions to the execution module, which is effectively connected with the upstream and downstream modules to form a complete closed-loop control system.

[0113] Figure 2 FIG. 1 is a schematic diagram of the structure of an autonomous navigation control system for an unmanned vehicle according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0114] Unit 1: Use multi-sensor fusion to perceive the surrounding environment of the vehicle, establish a semantic map model, use the semantic map model to extract the driving information of the unmanned vehicle, use the principal component analysis method to reduce the dimension of the driving information of the unmanned vehicle to obtain low-dimensional state features, build a Markov decision model based on the low-dimensional state features, and obtain the initial behavior strategy by solving the Markov decision model;

[0115] The second unit is used to generate local trajectory segments based on the initial behavior strategy, combined with the road topology and vehicle dynamics model, using the trajectory planning algorithm, and using the adaptive step size to discretize the local trajectory segments to generate a candidate trajectory set. For the candidate trajectory set, a scenario-based evaluation mechanism is introduced to calculate the expected cost of each candidate trajectory in each scenario, and the candidate trajectory with the smallest expected cost is selected as the optimal driving strategy;

[0116] The third unit is used to obtain the current speed of the vehicle and the curvature information of the road ahead according to the optimal driving strategy, adjust the preview distance in real time according to the changes in the current speed of the vehicle and the curvature information of the road ahead, obtain the predicted time domain length, generate the optimal control instructions within the predicted time domain length according to the preset model prediction controller, and perform autonomous navigation control on the unmanned vehicle through the optimal control instructions.

[0117] According to a third aspect of the embodiments of the present invention,

[0118] An electronic device is provided, comprising:

[0119] processor;

[0120] a memory for storing processor-executable instructions;

[0121] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0122] A fourth aspect of the embodiments of the present invention is:

[0123] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0124] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0125] 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 it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An autonomous navigation control method for an unmanned vehicle, characterized in that: include: The vehicle surrounding environment is perceived through multi-sensor fusion, a semantic map model is established, the driving information of the unmanned vehicle is extracted using the semantic map model, the driving information of the unmanned vehicle is reduced in dimension using the principal component analysis method to obtain low-dimensional state features, a Markov decision model is constructed based on the low-dimensional state features, and an initial behavior strategy is obtained by solving the Markov decision model; Based on the initial behavior strategy, combined with the road topology and vehicle dynamics model, a trajectory planning algorithm is used to generate local trajectory segments. At the same time, an adaptive step size is used to discretize the local trajectory segments to generate a candidate trajectory set. For the candidate trajectory set, a scenario-based evaluation mechanism is introduced to calculate the expected cost of each candidate trajectory in each scenario, and the candidate trajectory with the smallest expected cost is selected as the optimal driving strategy. According to the optimal driving strategy, the current speed of the vehicle and the curvature information of the road ahead are obtained. According to the changes in the current speed of the vehicle and the curvature information of the road ahead, the preview distance is adjusted in real time to obtain the predicted time domain length. According to the preset model prediction controller, the optimal control command within the predicted time domain length is generated, and the unmanned vehicle is autonomously navigated and controlled through the optimal control command. Based on the initial behavior strategy, combined with the road topology and vehicle dynamics model, the trajectory planning algorithm is used to generate local trajectory segments. At the same time, the adaptive step size is used to discretize the local trajectory segments. The candidate trajectory set generated includes: Decoding the initial behavior strategy into a motion constraint set, the motion constraint set including lateral constraints and longitudinal constraints, optimizing the motion constraint set in combination with a road topology structure, the road topology structure including road curvature, lane width, and intersection location, and performing a feasibility check on the motion constraint set using a pre-built vehicle dynamics model to obtain feasible motion constraints; A trajectory planning algorithm using layered spatiotemporal sampling is used to generate motion primitives covering the achievable motion space of the unmanned vehicle. The layered spatiotemporal sampling divides the sampling space into a lateral subspace and a longitudinal subspace. The lateral motion primitives are generated by sampling in the lateral subspace, and the longitudinal motion primitives are generated by sampling in the longitudinal subspace. The lateral motion primitive and the longitudinal motion primitive are combined to generate a local trajectory segment that satisfies the feasible motion constraint, the trajectory discrete step length is adaptively adjusted according to the road curvature and the obstacle density, the local trajectory segment is discretized according to the obtained discrete step length, and a candidate trajectory set is generated.

2. The method according to claim 1, characterized in that Optimizing the motion constraint set in combination with the road topology structure includes: The cost terms of road curvature, lane width and intersection position in the road topology are calculated respectively, and the comprehensive objective function of motion constraint is constructed by combining the road curvature cost term, lane width cost term and intersection position cost term. The calculation formula is as follows: ； Among them, J represents the comprehensive objective function, T represents the time domain range, ω1 represents the weight coefficient of the road curvature cost term, α1 represents the balance factor of the curvature size, η represents the road curvature, s represents the path arc length, t represents the time, α2 represents the balance factor of the curvature change, represents the rate of change of road curvature, ω2 represents the weight coefficient of the lane width cost term, β1 represents the amplification coefficient of the exponential cost term from the lane boundary, β2 represents the attenuation coefficient related to the exponential cost term from the lane boundary, d(t) represents the distance from the vehicle to the nearest lane boundary at time t, ω3 represents the weight coefficient of the intersection position cost term, C int represents the number of intersections, i represents the i-th intersection, γ1 represents the amplification coefficient related to the exponential cost term of the distance intersection, γ2 represents the attenuation coefficient related to the exponential cost term of the distance intersection, Represents the distance from the vehicle to the i-th intersection at time t.

3. The method according to claim 1, characterized in that For the candidate trajectory set, a scenario-based evaluation mechanism is introduced to calculate the expected cost of each candidate trajectory in each scenario, and the candidate trajectory with the smallest expected cost is selected as the optimal driving strategy, including: Determine multiple driving scenarios corresponding to each candidate trajectory in the candidate trajectory set and construct a scenario set, and use the scenario set to segment the candidate trajectory, each segment of the candidate trajectory corresponds to a driving scenario; The risk metric of each candidate trajectory in the corresponding driving scenario is calculated using the trained risk assessment model to obtain the trajectory segment risk sequence. The scenario transition probability is introduced to construct the transition relationship of driving scenarios between different candidate trajectories. The scenario transition probability is used to map the trajectory segment risk sequence into the scenario path risk distribution. Based on the scenario path risk distribution, trajectory evaluation indicators and cost functions are designed to obtain a comprehensive trajectory evaluation system for measuring candidate trajectories in different driving scenarios. The comprehensive trajectory evaluation system is used to calculate the cost of each candidate trajectory in each driving scenario. The comprehensive cost of the trajectory is obtained by combining the costs under different driving scenarios, and the candidate trajectory with the smallest comprehensive cost is selected as the optimal driving strategy.

4. The method according to claim 3, characterized in that: The calculation formula of trajectory evaluation index is as follows: ； Among them, G represents the trajectory evaluation index, M represents the number of scenarios, and δ b represents the weight coefficient of the b-th scenario, K represents the number of trajectory segments, a represents the a-th trajectory, j represents the j-th segment, and r a，j represents the risk measure of the j-th segment of the a-th trajectory, and P represents the scenario transition probability.

5. The method according to claim 1, characterized in that The optimal control instructions within the predicted time domain length are generated according to the preset model predictive controller, and the autonomous navigation control of the unmanned vehicle is performed through the optimal control instructions, including: Predicting the vehicle's trajectory tracking error within the prediction time domain length by using a pre-built vehicle dynamics model and a trajectory tracking error model, wherein the trajectory tracking error includes a lateral position error, a heading angle error, and a longitudinal velocity error; The vehicle's trajectory tracking error within the prediction time domain length is used as the input of the model predictive controller. The control variable change rate is introduced, and the controller multi-objective function is constructed in combination with the trajectory tracking error. The numerical optimization algorithm is used to solve the controller multi-objective function, and the optimal control instructions within the prediction time domain length are generated. The unmanned vehicle is autonomously navigated and controlled through the optimal control instructions.

6. The method according to claim 5, characterized in that The calculation formula of the controller multi-objective function is as follows: ； Where M represents the controller multi-objective function, N p represents the prediction time domain length, represents the error prediction value, k represents the current time, i represents the offset, Q represents the weight coefficient of the error prediction value, N c represents the control time domain length, u represents the control quantity sequence, △u(·) represents the control quantity change rate, R represents the weight coefficient of the control quantity change rate, and S represents the weight coefficient of the control quantity.

7. An autonomous navigation control system for an unmanned vehicle, used to implement the method described in any one of claims 1 to 6, characterized in that: include: Unit 1: Use multi-sensor fusion to perceive the surrounding environment of the vehicle, establish a semantic map model, use the semantic map model to extract the driving information of the unmanned vehicle, use the principal component analysis method to reduce the dimension of the driving information of the unmanned vehicle to obtain low-dimensional state features, build a Markov decision model based on the low-dimensional state features, and obtain the initial behavior strategy by solving the Markov decision model; The second unit is used to generate local trajectory segments based on the initial behavior strategy, combined with the road topology and vehicle dynamics model, using the trajectory planning algorithm, and using the adaptive step size to discretize the local trajectory segments to generate a candidate trajectory set. For the candidate trajectory set, a scenario-based evaluation mechanism is introduced to calculate the expected cost of each candidate trajectory in each scenario, and the candidate trajectory with the smallest expected cost is selected as the optimal driving strategy; The third unit is used to obtain the current speed of the vehicle and the curvature information of the road ahead according to the optimal driving strategy, adjust the preview distance in real time according to the changes in the current speed of the vehicle and the curvature information of the road ahead, obtain the predicted time domain length, generate the optimal control instructions within the predicted time domain length according to the preset model prediction controller, and perform autonomous navigation control on the unmanned vehicle through the optimal control instructions.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images