A motion control method for realizing modular bus station stop

Abstract

This invention discloses a motion control method for realizing modular bus stop stops, characterized by the following steps: constructing a motion control model and defining the objective function; acquiring modular bus data and environmental perception data, and inputting them into a trained first neural network model to obtain weight coefficients; weighting the objective function using the weight coefficients, and solving for the optimal control quantity under the motion control model; this invention utilizes the first neural network model to obtain weight coefficients for multiple objectives based on input features such as bus status, thus resolving the multi-objective conflict problem.

Description

Technical Field

[0001] This invention relates to the field of autonomous vehicle planning and control technology, and more specifically to a motion control method for realizing modular buses entering and stopping at stations. Background Technology

[0002] Currently, traditional public transportation systems rely on buses with fixed capacity. Because passenger demand may not match bus supply, fixed capacity can lead to longer wait times and higher operating costs. Emerging modular buses, which can adjust the number of modules in a fleet to change bus capacity, offer greater flexibility and represent a potential solution to these problems.

[0003] Current research in the field of modular public transport mainly focuses on scheduling, operation, and timetable optimization, and has made significant progress. To further promote the practical application of modular public transport, research on related control issues is also crucial. However, current research on modular public transport control is relatively scarce. Based on real-world scenarios, public transport control can be divided into intra-segment and intra-station control. Current control-related research focuses on the docking or separation of modular public transport within segments, lacking research on control within station scenarios. How modular public transport smoothly and with low energy consumption moves to the station and accurately stops at designated windows is crucial, affecting both passenger experience and operator image. Currently, there is no research on the motion control of modular public transport specifically addressing this process.

[0004] Parallel parking scenarios for private cars are similar to those for modular buses stopping at bus stops, thus offering potential avenues for method borrowing. In this field, Nonlinear Model Predictive Control (NMPC) is highly regarded for its ease of handling high-precision problems and constructing constrained multi-objective optimization models. Therefore, NMPC has emerged as a potential solution for controlling modular buses stopping at bus stops. The vehicle dynamics model and the weights of the objective function are the main factors affecting NMPC performance. In parallel parking, a common approach is to manually derive a dynamic model based on physical laws. Weights are then manually tuned offline through trial and error, and fixed during online testing. Current research has validated the feasibility of these strategies for achieving high-performance parallel parking. However, due to the numerous differences between modular buses and private cars, the aforementioned strategies are insufficient to meet the performance requirements of stopping scenarios.

[0005] On the one hand, modular buses can flexibly dock and disengage, thus their dynamics system may undergo significant changes. This flexibility requires a more adaptable vehicle dynamics model. Enumerating all conditions and manually deriving the physical model is extremely labor-intensive, and the physical parameters needed for the model may not be available. On the other hand, modular buses have high requirements for ride comfort and low energy consumption, and need to stop precisely at designated windows. However, the objective functions corresponding to these requirements may conflict, making it difficult to achieve the best overall performance.

[0006] Therefore, how to deal with the variability of modular buses, balance the conflicts between multiple objectives, and achieve safe, efficient, and comfortable entry and stopping control of modular buses in station scenarios is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a motion control method for realizing modular bus stop stops, which uses a first neural network model to obtain the weight coefficients of multiple targets based on input features such as bus status, thereby solving the multi-target conflict problem.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A motion control method for implementing modular bus stop stops includes the following steps:

[0010] Construct a motion control model and define the objective function.

[0011] The module bus data and environmental perception data are acquired and input into the trained first neural network model to obtain weight coefficients.

[0012] The objective function is weighted by the weighting coefficients, and the optimal control quantity is obtained by solving under the motion control model.

[0013] Preferably, the first neural network model fuses the input features based on the encoder module in the Transformer, then passes through a multilayer perceptron with residual structure, and finally passes through the output layer to obtain the weight values.

[0014] Preferably, the input features include the quadratic error between the current pose and the desired pose of the bus, the current speed, the current control quantity, the mass and moment of inertia of the bus.

[0015] Preferably, the first neural network model is trained through reinforcement learning, and the reward function used during training is:

[0016] R = (J energy +J smooth +J precise )

[0017] Where R is the reward function, J energy J represents the actual energy consumption of public transportation. smooth For the resultant acceleration, J precise This represents the deviation between the final docking location and the expected location.

[0018] Preferably, the construction of the motion control model and the definition of the objective function specifically include:

[0019] Construct the objective function:

[0020]

[0021] Where J is the objective function; ζ r =ζ-ζ d Let ζ = [s, e, θ] represent the longitudinal and lateral positional deviations of the bus. T The actual position is represented by the components: longitudinal position, lateral position, and heading angle in the Frenet coordinate system, ζ. d The corresponding expected value; u = [f t ,δ] T To coordinate with the control parameters of the bus, the components are, in order, tire driving force and steering angle, Q. p Q a Q j Q t This is the weight coefficient matrix;

[0022] Construct a vehicle dynamics model;

[0023] Construct constraints, including dynamic constraints based on the vehicle dynamics model:

[0024]

[0025] Where, x = [v x ,v y ,ω] T Represents state variables. v represents the reciprocal of the state vector. x ,v y ω represents the longitudinal and lateral velocities and the angular velocity, respectively, and F is the vehicle dynamics model.

[0026] Preferably, the constraints also include lane boundary restrictions based on the shape of the modular bus and physical constraints on the state and control quantities required for the docking bus.

[0027] Preferably, the lane boundary constraint based on the shape of the modular bus includes:

[0028] e+l f sin(θ)+w lcos(θ)≤G l (s)

[0029] el r sin(θ)+w l cos(θ)≤G l (s)

[0030] e+l f sin(θ)-w r cos(θ)≥G r (s)

[0031] el r sin(θ)-w r cos(θ)≥G r (s)

[0032] Among them, w l and w r Let l represent the distances from the centroid to the left and right ends of the bus, respectively. f ,l r G represents the distance from the center of mass to the front and rear of the bus, respectively. l G r Functions representing the upper and lower lane boundaries, respectively;

[0033] The physical constraints required for the state and control variables of the connected buses include:

[0034] u min ≤u≤u max

[0035] Among them, u min ,u max To control the physical limits of the quantity.

[0036] Preferably, the construction of the vehicle dynamics model specifically includes:

[0037] Obtain the state of each axle and use the resulting state sequence as input information.

[0038] The second neural network model is used to extract feature vectors and output the resultant force and resultant torque.

[0039] The resultant acceleration is calculated based on the resultant force and resultant moment, and longitudinal and transverse coupling terms are added to obtain the model output.

[0040] Preferably, the construction of the vehicle dynamics model further includes training the parameters in the second neural network model based on a labeled dataset.

[0041] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a motion control method for realizing modular bus stop stops. By constructing a motion control model and designing a specific objective function, a first neural network model is used to train the model through reinforcement learning. The weight coefficients are then used to solve the optimal control quantity by weighting the objective function with the weight coefficients, thus solving the multi-objective conflict problem. At the same time, a second neural network model is used to extract features from the axle state sequence to output the resultant force and resultant torque. Combined with the resultant acceleration and coupling terms, a vehicle dynamics model is constructed. The parameters are trained based on the dataset to achieve adaptive modeling of the multi-variable dynamic system of the modular bus, thereby realizing safe, efficient and comfortable stop control of the modular bus in the station scenario. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0043] Figure 1 This is a schematic diagram of the motion control framework for realizing modular bus stop stopping proposed in this invention.

[0044] Figure 2 This is a schematic diagram of a modular bus stopping at a station.

[0045] Figure 3 The parameter diagrams for different vehicles used in the experiment were used for verification.

[0046] Figure 4 This is a graph showing the fitting error of the dynamic model.

[0047] Figure 5 This is a diagram showing the high-speed extrapolation effect of the dynamic model.

[0048] Figure 6 The overall performance results obtained by setting different objective function weights are shown in the figure. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] like Figure 1 and Figure 2 This invention discloses a motion control method for realizing modular bus stop stopping, comprising the following steps:

[0051] S1: Construct a motion control model and define the objective function.

[0052] S2: Obtain module bus data and environmental perception data, and input them into the trained first neural network model to obtain weight coefficients.

[0053] S3: The objective function is weighted by weighting coefficients, and the optimal control quantity is obtained by solving under the motion control model.

[0054] The motion control model is based on NMPC theory and is constructed offline.

[0055] As a feasible implementation method, the constructed motion control model includes the following specific features:

[0056] The objective function is designed by comprehensively considering minimizing docking error, minimizing energy consumption, and maximizing comfort. Specifically:

[0057]

[0058] Where J is the objective function, t f The final stop; ζ r =ζ-ζ d , ζ d Including the corresponding expected value, ζ=[s,e,θ] T The components of the bus's longitudinal and lateral positions and heading angle are, in order, velocity, longitudinal position in the Frenet coordinate system, lateral position, and heading angle; u = [f t ,δ] T To coordinate with the control parameters of the bus, the components are, in order, tire driving force and steering angle; Q p Q a Q j Q t This is a weighting coefficient matrix, used to assign weights to each state variable or control variable in relation to docking error, energy consumption, comfort, and final state.

[0059] in, The error term represents the minimum docking error, u. T Q a u represents energy consumption. It is used to react to impact, in order to reflect comfort.

[0060] Constraints:

[0061] (1) The state and control variables of the modular bus must satisfy the dynamic constraints:

[0062]

[0063] Where, x = [v x ,v y ,ω] T Represents a state variable. v x ,v y ω represents the longitudinal and lateral velocities and the angular velocity, respectively. F defines the vehicle dynamics model.

[0064] A road model based on a Cartesian coordinate system is chosen to describe the relationship between velocity and position:

[0065]

[0066] in, and These represent the rates of change of longitudinal position, lateral position, and heading angle, respectively; θ is the heading angle; v x and v y These are the components of velocity along the x-axis and y-axis, respectively.

[0067] (2) The shape of the modular bus must not violate lane boundary restrictions:

[0068] e+l f sin(θ)+w l cos(θ)≤G l (s)

[0069] el r sin(θ)+w l cos(θ)≤G l (s)

[0070] e+l f sin(θ)-w r cos(θ)≥G r (s)

[0071] el r sin(θ)-w r cos(θ)≥G r (s)

[0072] Among them, l f ,l r G represents the distance from the center of mass to the front and rear of the bus, respectively. l G r These are functions representing the boundaries of the upper and lower lanes, respectively.

[0073] (3) The state and control variables of the connected bus must meet physical constraints:

[0074] u min ≤u≤u max

[0075] Among them, u min ,u max To control the physical limits of the quantity.

[0076] (4) The initial state of the model must be equal to the current state of the bus: x0 = x(0).

[0077] To further implement the above technical solution, a vehicle dynamics model is constructed based on PINN. The specific method is as follows:

[0078] Obtain the state of each axle and use the resulting state sequence as input information; for each axle, take the following input information:

[0079] x in =[f t ,f y ,l,δ,f t δ] T

[0080] Among them, f t ,f y These represent the longitudinal and lateral forces on the tire of this axle, respectively; δ is the tire's steering angle; l is the distance from the axle to the center of gravity; f y ≈c y α, c y Let α be the tire's lateral stiffness and slip angle, respectively. For the front axle, α ≈ δ - (v y +l f ω) / v x For the rear axle, α≈δ-(v y -l r ω) / v x .

[0081] A second neural network model (such as a recurrent neural network model) is used to extract feature vectors and output the resultant force and resultant torque;

[0082]

[0083] in, This is a feature vector (which can be five-dimensional) containing force and torque information. Gelu(*) is the activation function. hi To hide features (forces and moments calculated from the previous axis). in ,w hi b is a learnable parameter.

[0084]

[0085] Among them, f aFor the resultant force and resultant torque, w a ,b a w are learnable parameters a The dimension can be selected as 3×3.

[0086] The resultant acceleration is calculated based on the resultant force and resultant moment, and longitudinal and transverse coupling terms are added to obtain the model output.

[0087]

[0088] y = a + [v y ω,-v x ω,0] T

[0089] Where y is the final output of the neurodynamic model, i.e., the second neural network model, and a is the resultant acceleration. This indicates element-wise division; w1, w2, and w3 are learnable parameters.

[0090] Furthermore, the parameters in the second neural network model are trained based on the labeled dataset. During the training process, the training set undergoes the aforementioned forward propagation process to obtain predicted values. The model loss is then calculated based on the predicted values, and the parameters are updated.

[0091] As a feasible implementation method, the present invention achieves dynamic adjustment of multiple target weights in the objective function through a first neural network model, and trains the network offline based on reinforcement learning.

[0092] Specifically, the network input features include: the quadratic error between the bus's current pose and the desired pose, the current speed, the current control input, and the bus's mass and moment of inertia. The network output is Q. p Q a Q j The values ​​in the table represent the weighting coefficients of each state variable or control variable for docking error, energy consumption, and comfort, respectively.

[0093] The network structure of the first neural network model is as follows: First, linear layers are used to increase the dimension of the input features to 14. The input features are then fused using the encoder module in the Transformer, which has two layers and two heads, with a model dimension of 14. Next, the model passes through a multilayer perceptron with residual structures, resulting in a dimension of 14×14×14. Finally, the output layer is applied, with a dimension of 14×7.

[0094] The reward function settings required for training a network based on reinforcement learning are as follows:

[0095] R = (J energy +J smooth +J precise )

[0096] Where R is the reward function. J energy This represents the actual energy consumption of public transportation. smooth J is the resultant acceleration. precise This represents the deviation between the final docking location and the desired location, and can be defined as follows:

[0097]

[0098] Among them, s d ,e d These represent the desired longitudinal and transverse positions, respectively. d m This is the threshold for docking error.

[0099] The network is trained using a deep reinforcement learning algorithm, such as the SAC algorithm, based on the first network model and its reward function. The Critic's network architecture can be the same as the Actor's. The default loss function is used for both.

[0100] In this embodiment, at the start of each generation of training, the current state information of the modular bus is acquired, including: the quadratic error between the current pose and the desired pose, the current speed, the current control input, and the bus's mass and moment of inertia. This data is then formatted to be suitable for input into the first neural network model. The model then fuses the input features based on the Transformer's encoder module, capturing the correlation between the modular bus data and the environmental perception data. The fused features are further processed by a multilayer perceptron with residual structures, and finally, weight values ​​are obtained through the output layer. These weight coefficients are used to weight the objective function subsequently. Since the objective function is constantly changing, the newly generated weight coefficients will affect the importance of each sub-objective (such as stopping accuracy, comfort, efficiency, etc.) in the objective function. For example, if the current road is congested, the weight of avoiding sudden braking (to ensure comfort) in the objective function may increase; if approaching a bus stop, the weight of stopping accuracy in the objective function may increase. Under the updated objective function and motion control model, an appropriate optimization algorithm (such as gradient descent, Newton's method, etc.) is used to solve for the optimal control quantity. This optimal control quantity is then applied to the bus to enable it to perform corresponding actions, such as acceleration, deceleration, and steering. The goal of this optimal control quantity is to ensure the bus operates optimally under the current objective function and bus state, for example, stopping at a bus stop with appropriate speed and acceleration.

[0101] After a bus performs an action, it interacts with its environment. The environment updates the bus's status (such as changes in position and speed) based on the bus's actions, and generates corresponding rewards (or penalties). For example, if the bus stops accurately at a bus stop, it will receive a higher reward; if the bus collides with something or deviates from its route, it will receive a penalty.

[0102] Finally, data such as the new state after the bus performs its actions, the reward obtained, and the control amount executed are collected. This data is used to train the first neural network model. The model parameters are adjusted through reinforcement learning algorithms (such as deep Q-networks, policy gradient algorithms, etc.) so that the weight coefficients generated in subsequent steps can better adapt to changes in the objective function.

[0103] The present invention will further verify the above method with specific examples below.

[0104] A parking scenario is created based on Carla. Carsim / Trucksim provides small, medium, and large vehicle sizes; see details for specific parameters. Figure 3 Write the algorithm in Python. The initial condition for the reinforcement learning scenario is set to v. x ∈[4,6]m / s,v y ∈[-0.05,0.05]m / s,ω∈[-0.05,0.05]rad / s,s∈[-40,-20]m,e∈[4,6]m,θ∈[-1,1]°. The desired docking pose is [0,1,0]. A successful docking is considered achieved if the distance between the vehicle and the desired position is less than 0.05m, the heading angle deviation is less than 1°, and the speed is less than 0.01m / s. The entire training process comprises 600 rounds.

[0105] The original NMPC model was used as the baseline, with the vehicle dynamics model being a common physics-based single-scale model. This dynamics model was then linearized to form an LMPC control model as another baseline. A neurodynamics model and an adaptive weighting method were used for an enhanced model predictive control (RMPC). The overall performance statistics of the three methods under different vehicle models are shown in the table below.

[0106] Table 1 Overall performance statistics of different methods

[0107]

[0108] As can be seen from the table above, the proposed RMPC outperforms the benchmark methods in all performance metrics, especially in docking error. Experiments revealed that LMPC and NMPC suffer from significant errors in their dynamic models, leading to large deviations from the true model and frequently resulting in unsolvable problems.

[0109] Under the scenario settings corresponding to Table 1, the fitting error of the dynamic model is as follows: Figure 4 As shown in the figure, the neurodynamic model has higher accuracy. Setting the initial lateral velocity and angular velocity to 0, sinusoidal signals for driving force and steering angle are applied, with amplitudes f0 and f1 respectively. t =1KN,δ f=0.1 rad, period of 2 seconds. Gradually increase the longitudinal velocity, the extrapolation result is as follows: Figure 5 As shown, we found that although the neurodynamic model was trained using only data with speeds less than 6 m / s, it still exhibited strong extrapolation capabilities for high-speed conditions.

[0110] Using Bayesian optimization, expert-tuned objective function weights, and randomly generated objective function weights as benchmarks, the overall performance is evaluated using the reward value calculated by equation (31). In the randomly generated scenario, the proposed neural network-based parameter tuning method is compared with the results of each benchmark. Figure 6 As shown in the figure, it can be seen that the overall performance of the proposed method is significantly better than that of other methods.

[0111] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0112] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A motion control method for implementing modular bus stop boarding, characterized by, Includes the following steps: Construct a motion control model and define the objective function; specifically including: Construct the objective function: wherein, is the objective function, is the final stop time; is the longitudinal and lateral position deviation of the bus, is the actual position, with components being the longitudinal position, lateral position and heading angle in the Frenet coordinate system, is the corresponding expected value; is the control amount of the docking bus, with components being the tire driving force and the steering angle, is the weight coefficient matrix; Construct a vehicle dynamics model; The constraints are constructed, including dynamic constraints based on the vehicle dynamics model, lane boundary limit constraints based on the shape of the modular bus, and physical constraints on the state and control quantities of the docking bus. Dynamic constraints based on the aforementioned vehicle dynamics model: in, Represents state variables. Represents the reciprocal of the state vector. These represent the longitudinal and transverse velocities and the angular velocity, respectively. For vehicle dynamics model; The lane boundary constraint based on the shape of the modular bus includes: in, and These represent the distances from the center of mass to the left and right ends of the bus, respectively. These represent the distances from the center of mass to the front and rear of the bus, respectively. Functions representing the upper and lower lane boundaries, respectively. The vertical position in the Frenet coordinate system The horizontal position in the Frenet coordinate system. For heading angle; The physical constraints required for the state and control variables of the connected buses include: in, The physical limit of the control quantity The module bus data and environmental perception data are acquired and input into the trained first neural network model to obtain weight coefficients; The objective function is weighted by the weighting coefficients, and the optimal control quantity is obtained by solving under the motion control model.

2. The motion control method for realizing modular bus stop stopping according to claim 1, characterized in that, The first neural network model fuses input features based on the encoder module in Transformer, then passes through a multilayer perceptron with residual structure, and finally passes through the output layer to obtain the weight values.

3. The motion control method for realizing modular bus stop stopping according to claim 2, characterized in that, The input features include the quadratic error between the current and desired poses of the bus, the current speed, the current control input, the mass and moment of inertia of the bus.

4. A motion control method for realizing modular bus stop stopping according to claim 1, 2, or 3, characterized in that, The first neural network model is trained through reinforcement learning, and the reward function used during training is: in, For the reward function, This represents the actual energy consumption of public transportation. For the resultant acceleration, This represents the deviation between the final docking location and the expected location.

5. The motion control method for realizing modular bus stop stopping according to claim 1, characterized in that, The construction of the vehicle dynamics model specifically includes: Obtain the state of each axle and use the resulting state sequence as input information; The second neural network model is used to extract feature vectors and output the resultant force and resultant torque; The resultant acceleration is calculated based on the resultant force and resultant moment, and longitudinal and transverse coupling terms are added to obtain the model output.

6. The motion control method for realizing modular bus stop stopping according to claim 5, characterized in that, The construction of the vehicle dynamics model also includes training the parameters in the second neural network model based on a labeled dataset.

Images