Local dynamic path planning method based on dynamic window method

Abstract

The invention relates to a local dynamic path planning method based on a dynamic window method. The method comprises the following steps: step 1, constructing a kinematic model of a robot; 2, performing speed sampling, and simulating a sampled speed combination based on the kinematic model to obtain a plurality of simulation paths; 3, constructing an evaluation function in dynamic window method iteration, dynamically adjusting a weight parameter of the evaluation function through reinforcement learning, and obtaining a scoring result of the simulation path based on the evaluation function; and 4, comparing the scoring results of all the simulation paths, and selecting the robot driving path of the next time step based on the comparison result. According to the invention, through reinforcement learning, the evaluation function weight parameter can be continuously adjusted to a proper value according to the change of the environment state; and meanwhile, the robot can be guided to select a smoother path by utilizing an evaluation function, so that the adaptability of the robot to the environment and the flexibility of path planning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of robot path planning, and in particular to a local dynamic path planning method based on the dynamic window approach. Background Art

[0002] With the rapid development of artificial intelligence and robot technology, mobile robots are playing an increasingly important role in fields such as military reconnaissance, industrial automation, agricultural production, logistics distribution, and service robots. In tasks such as autonomous navigation and target tracking, mobile robots often face complex dynamic environments, and the presence of dynamic obstacles poses a severe challenge to path planning.

[0003] Path planning is usually divided into global path planning and local path planning. Global path planning mainly focuses on the search for the overall optimal path and is applicable to known static environments. In contrast, local path planning pays more attention to the real-time decision-making ability of mobile robots in unknown or dynamic environments. Typical local path planning algorithms include the artificial potential field method and the dynamic window approach, etc. The artificial potential field method realizes path planning by constructing a gravitational field and a repulsive field, but the disadvantage is that it is easy to fall into local optima and cannot fully consider the dynamic attributes of obstacles. The dynamic window approach (DWA) samples candidate velocities in the robot control space and selects the optimal velocity command by combining an evaluation function, thereby realizing obstacle avoidance and path planning. DWA performs well in static environments due to its high computational efficiency and simple implementation. However, in dynamic environments, due to the failure to fully consider the speed and movement direction of obstacles and the influence of the robot's own speed on the planned path, situations such as untimely obstacle avoidance and obstacle avoidance failure may occur; due to the speed changing continuously under the influence of the environment, the problem of uneven paths may also occur.

[0004] In recent years, there have been numerous improvement studies on the DWA algorithm. However, the improvements of existing DWA algorithms all rely on fixed parameters or preset models, lacking real-time feedback and dynamic adjustment capabilities, and unable to effectively cope with the rapid changes in the environment. Summary of the Invention

[0005] The present invention provides a local dynamic path planning method based on the dynamic window approach to solve the above technical problems.

[0006] To solve the above technical problems, the present invention provides a local dynamic path planning method based on the dynamic window approach, including the following steps:

[0007] Step 1: Construct a kinematic model of the robot;

[0008] Step 2: Perform velocity sampling and simulate the sampled velocity combinations based on the kinematic model to obtain multiple simulated paths;

[0009] Step 3: Construct an evaluation function in the iteration of the dynamic window method, dynamically adjust the weight parameters of the evaluation function through reinforcement learning, and obtain the scoring result of the simulation path based on the evaluation function;

[0010] Step 4: Compare the scoring results of all simulation paths, and select the robot driving path for the next time step based on the comparison results.

[0011] Preferably, in Step 1, the kinematic model is a circular arc model.

[0012] Preferably, in Step 3, dynamically adjusting the weight parameters of the evaluation function through reinforcement learning includes: establishing a Markov decision process model based on the evaluation function.

[0013] Preferably, establishing a Markov decision process model includes: designing a state set, designing an action set, and designing a reward and punishment function.

[0014] Preferably, the vector representation s of the state set is:

[0015] s = [s1, s2, s3]

[0016] where s1 represents the state associated with the distance between the robot and the nearest obstacle, s2 represents the state associated with the distance between the current position of the robot and the target point, and s3 represents the linear velocity state of the robot.

[0017] Preferably, the action set is the adjustment of the three weight parameters, and the weight parameters are the weight of the target direction, the weight of the obstacle distance, and the weight of the motion speed.

[0018] Preferably, the reward and punishment function R is:

[0019] R = R1 + R2 + R3

[0020] where R1 is the reward related to the distance to the nearest obstacle, R2 is the reward related to whether approaching the target, and R3 is the reward related to whether reaching the target or colliding with an obstacle.

[0021] Preferably, the evaluation function G(v, ω) is:

[0022] G(v, ω) = δ · (β · dist(v, ω) + γ · heading(v, ω) + σ · v(v) + ρ · k(v, ω))

[0023] Among them, dist(v, ω) is the obstacle distance evaluation function, heading(v, ω) is the target direction evaluation function, v(v) is the speed evaluation function, k(v, ω) is the smoothness evaluation function, β, γ, σ, and ρ are the weight factors of obstacle distance, target direction, movement speed, and smoothness respectively, and δ is the normalization coefficient.

[0024] Preferably, the calculation formula of the smoothness evaluation function k(v, ω) is:

[0025]

[0026] where v is the linear velocity of the robot and ω is the angular velocity of the robot.

[0027] Preferably, in step 4, the path with the highest score in the scoring result is used as the driving path of the robot in the next time step, and the corresponding linear velocity and angular velocity are used as the speed control commands of the robot at the next moment.

[0028] Compared with the prior art, the local dynamic path planning method based on the dynamic window approach provided by the present invention has the following advantages:

[0029] 1. The method (Q-DWA) of dynamically adjusting the weight parameters of the evaluation function based on reinforcement learning (Q-learning) in the present invention establishes a Markov decision process (MDP) model, including the design of a state set, an action set, and a reward function, and can continuously adjust the weight parameters of the evaluation function to appropriate values according to the changes in the environmental state;

[0030] 2. On the basis of Q-DWA, the present invention adds a smoothness evaluation function, borrowing the idea of curvature, which can guide the robot to select a smoother path. Description of the Drawings

[0031] Figure 1 is a flowchart of the local dynamic path planning method based on the dynamic window approach in a specific embodiment of the present invention;

[0032] Figure 2 is the motion model of the mobile robot;

[0033] Figure 3 is a schematic diagram of the simulated path of the mobile robot;

[0034] Figure 4a 、 4b and 4c are respectively the path planning result diagrams of three methods (traditional DWA, Q-DWA, and the complete improved scheme) in environment 1;

[0035] Figure 5a 、 5bFigures 5a and 5c respectively show the path planning results of three methods (traditional DWA, Q-DWA, and the complete improved solution) in Environment 2. Detailed implementation manners

[0036] To describe the technical solutions of the above invention in more detail, specific embodiments are listed below to prove the technical effects; it should be emphasized that these embodiments are used to illustrate the present invention and not to limit the scope of the present invention.

[0037] The local dynamic path planning method based on the dynamic window method provided by the present invention, as Figure 1 shown, includes the following steps:

[0038] Step 1: Construct the kinematic model of the robot;

[0039] Step 2: Perform speed sampling, and simulate the sampled speed combinations based on the kinematic model to obtain multiple simulated paths;

[0040] Step 3: Construct an evaluation function in the dynamic window method iteration, dynamically adjust the weight parameters of the evaluation function through reinforcement learning, and obtain the scoring results of the simulated paths based on the evaluation function;

[0041] Step 4: Compare the scoring results of all simulated paths, and select the driving path of the robot at the next time step based on the comparison results.

[0042] By introducing the idea of reinforcement learning and using Q-learning to dynamically adjust the weight parameters of the evaluation function in the DWA algorithm, the present invention enables the robot to adaptively adjust the obstacle avoidance strategy according to different environmental changes, thereby improving the real-time performance and flexibility of path planning; and on the basis of Q-DWA, the idea of curvature is borrowed and used as a new evaluation index to be added to the evaluation function to balance the smoothness and flexibility of the robot path planning.

[0043] The dynamic window method can provide real-time obstacle avoidance and navigation paths for the robot in a dynamic environment. Its core idea is to simulate and predict all possible speed combinations in combination with environmental information within the current movement speed range of the robot, and screen out the optimal speed combination through a comprehensive evaluation function, so that the robot can avoid obstacles and move towards the target point. DWA is widely used in various mobile robot systems due to its advantages such as high computational efficiency and strong real-time performance. This application elaborates in detail the implementation process of DWA from three aspects: the robot kinematic model, speed sampling, and path scoring.

[0044] The core of the dynamic window method lies in predicting the future motion trajectory of the mobile robot through speed sampling. Therefore, it is necessary to conduct a detailed study and modeling of the kinematic model of the mobile robot. Assume that the current linear velocity of the mobile robot is v and the angular velocity is ω. Its pose at time t is described by the position coordinates (x t , y t ) and the orientation angle θ t , that is, (x t , y t , θ t ). In the global coordinate system XOY, these parameters jointly define the motion state of the robot at a specific moment. The kinematic model of the mobile robot is as shown in Figure 2 .

[0045] For the convenience of analysis, assume that the robot moves in a uniform straight line within the adjacent time interval [t, t + 1], that is, the linear velocity and angular velocity remain unchanged. Under this assumption, the position and orientation of the robot at time t + 1 can be calculated by the following formulas:

[0046] x(t + 1) = x(t) + v(t)·cos(θ(t))·Δt (1)

[0047] y(t + 1) = y(t) + v(t)·sin(θ(t))·Δt (2)

[0048] θ(t + 1) = θ(t) + ω(t)·Δt (3)

[0049] Among them, Δt is the time interval between adjacent sampling points, and v(t) and ω(t) respectively represent the linear velocity and angular velocity at time t. The above formulas show that the linear velocity v determines the linear displacement of the robot's motion, while the angular velocity ω affects its direction change.

[0050] However, in actual situations, since the robot is often in a state of non-zero linear velocity and angular velocity, its motion trajectory is usually a curve rather than an ideal straight line. To more accurately describe this curvilinear motion, this application uses an arc model instead of a straight line model. Assume that the linear velocity and angular velocity of the robot are constant at time t. Then the curvature of the robot's motion trajectory can be represented by the radius of the arc R, and the calculation formula is:

[0051]

[0052] Based on this arc model, the pose of the robot at time t + 1 can be deduced by the following formula:

[0053] x(t + 1) = x(t) - R·sin(θ(t)) + R·sin(θ(t) + ω(t)·Δt) (5)

[0054] y(t + 1) = y(t) + R·cos(θ(t)) - R·cos(θ(t) + ω(t)·Δt) (6)

[0055] θ(t + 1) = θ(t) + ω(t)·Δt (7)

[0056] The above formulas indicate that the pose change of the robot within one time step depends on the linear velocity v(t), the angular velocity ω(t), and the current orientation angle θ(t). Among them, the arc radius R is determined by the ratio of the linear velocity to the angular velocity, reflecting the degree of curvature of the robot's motion trajectory.

[0057] The dynamic window method uses the above kinematic model to simulate the sampled velocity combinations to predict the possible future trajectories of the robot. By comprehensively evaluating multiple possible trajectories with the target point, the obstacle distribution, and the motion characteristics of the robot, the algorithm selects the optimal path, thereby guiding the robot's dynamic obstacle avoidance and path planning in a complex environment.

[0058] Velocity sampling is crucial for the dynamic window method to simulate the next trajectory. The velocity sampling space of the mobile robot is restricted by the magnitude of the mobile robot's velocity, the acceleration of the mobile robot, and the obstacles in the environment.

[0059] Due to its own velocity magnitude, the linear velocity and angular velocity of the robot are within a certain range, and the range is as follows:

[0060] V a = {(v, ω)|v ∈ [v min , v max ∩ ω ∈ [ω min , ω max} (8)

[0061] v is the minimum linear velocity, and ω min and ω max are the minimum and maximum angular velocities of the robot respectively. Affected by its maximum acceleration, the linear velocity and angular velocity of the robot need to be within the following ranges:

[0062] V b = {(v, ω)|v ∈ [v0 - α v Δt, v0 + α v Δt], ω ∈ [ω0 - α ω Δt, ω0 + α ω Δt]} (9)

[0063] Among them, v0 is the current linear velocity of the robot, and α v is the maximum linear acceleration; ω0 is the current angular velocity of the robot, and α ω is the maximum angular acceleration.

[0064] In a real environment, the robot's linear velocity and angular velocity at the next moment will be limited by obstacles, and the limitations are as follows:

[0065]

[0066] where dist(v,ω) is the distance between the corresponding robot and the nearest obstacle.

[0067] In summary, the velocity space of the robot's dynamic window is shown as follows:

[0068] V d =V a ∩V b ∩V c (11)

[0069] Based on the robot kinematic model, combined with the speed space restriction of the robot dynamic window, the feasible motion trajectory of the robot is generated. The predicted trajectory of the mobile robot is shown in the figure below: Figure 3 shown.

[0070] Each simulated path generated by state prediction is evaluated through a scoring process, comprehensively considering safety, goal orientation, and motion efficiency, ultimately selecting the optimal path and corresponding speed combination. The core of the scoring process lies in defining a reasonable evaluation function and normalizing various indicators.

[0071] The obstacle distance evaluation function is used to evaluate the distance between the predicted path and the obstacle to ensure the safety of the path. It is defined as:

[0072]

[0073] The greater the distance between the path and the obstacle, the higher the score.

[0074] The target direction evaluation function is used to measure the direction deviation between the robot's current position and the target point. It is defined as:

[0075]

[0076] The smaller the deviation, the higher the score.

[0077] The speed evaluation function is used to encourage the robot to maintain a higher speed to improve movement efficiency, and is defined as:

[0078]

[0079] Based on the above three indicators, DWA obtains the total score of the path through weighted summation:

[0080] G(v,ω)=δ·(β·dist(v,ω)+γ·heading(v,ω)+σ·v(v)) (15)

[0081] Among them, β, γ, and σ are the weight factors of the obstacle distance, target direction, and speed evaluation function respectively, and δ is the normalization coefficient used to balance the relative importance of each index.

[0082] Finally, by comparing the scoring results of all paths, the path with the highest score is selected as the robot's driving path for the next time step, and the corresponding v and ω are used as the speed control commands for the robot at the next moment.

[0083] This application dynamically adjusts the weight parameters of the evaluation function of DWA through reinforcement learning, enabling the robot to adaptively plan paths in different environments. Q-learning is a type of reinforcement learning algorithm that introduces a reward and punishment mechanism and is suitable for exploring unstructured environments. The key lies in establishing a Markov decision process (MDP) model based on the evaluation function in the iteration of the dynamic window method, including the design of the state set, action set, and reward function.

[0084] In path planning, the design of the state set directly affects the algorithm's ability to model the environment and the effectiveness of the robot's decision-making. A good state set can comprehensively describe the characteristics of the environment where the robot is currently located and provide sufficient information required for decision-making. In this application, combined with the characteristics of the dynamic window method, a state set is designed to describe the motion state and relative position relationship of the robot in the environment, and the state set can be represented by the following vector:

[0085] s = [s1, s2, s3] (16)

[0086] Among them, s1 represents the state associated with the distance between the robot and the nearest obstacle, s2 represents the state associated with the distance between the robot's current position and the target point, and s3 represents the robot's linear velocity state.

[0087] Among them, s1 and s2 have 2 possibilities, and s3 has 3 possibilities.

[0088] The definition of s1 is as follows:

[0089]

[0090] Among them, d o is the distance between the robot and the nearest obstacle, and d1 is the distance state judgment coefficient between the robot and the nearest obstacle.

[0091] The definition of s2 is as follows:

[0092]

[0093] Among them, d gLet d1 be the distance between the current position of the robot and the target point, and d2 be the distance status judgment coefficient between the current position of the robot and the target point.

[0094] The definition of s3 is as follows:

[0095]

[0096] Where v is the linear velocity of the robot, v1 is the low-speed judgment coefficient of the robot's linear velocity, and v2 is the high-speed judgment coefficient of the robot's linear velocity.

[0097] Finally, the multi-dimensional state combination is mapped to a one-dimensional index through a formula, and the formula definition is as follows:

[0098] s id =(s1 - 1)·6+(s2 - 1)·3+(s3 - 1) (20)

[0099] Where s id is the one-dimensional index of the state.

[0100] The design of the action set determines the types and ranges of behaviors that the robot can take. In the dynamic window approach, the core of path planning is to optimize the path by adjusting the evaluation weight parameters in the objective function. In this application, the action is designed as the adjustment of three key evaluation weight parameters in DWA, namely the weight of the target direction, the weight of the obstacle distance, and the weight of the movement speed. In this embodiment, 16 special combinations are designed, and these combinations form the action dimension.

[0101] The design of the reward function is the core of Q-learning, which directly determines the learning efficiency and final performance of the algorithm. The reward function in this application consists of three parts, which respectively measure the target proximity, safety, and movement efficiency, and comprehensively reflect the quality of path planning. In order to dynamically adjust the evaluation function weight coefficient with Q-learning, the reward function R is defined as:

[0102] R = R1 + R2 + R3 (21)

[0103] Where R1 is the reward related to the distance to the nearest obstacle, and the definition of R1 is as follows:

[0104]

[0105] R2 is the reward related to whether approaching the target, and the definition of R2 is as follows:

[0106]

[0107] R3 is the reward related to whether reaching the target or colliding with an obstacle, and the definition of R3 is as follows:

[0108]

[0109] Through these reward and punishment functions, the dynamic window method can better use appropriate evaluation weight parameters in different states of the robot, enhancing the flexibility of the algorithm.

[0110] In order to improve the smoothness of the path generated by the dynamic window method, this application borrows the idea of curvature and designs a smoothness evaluation function, namely function k, and adds it as an evaluation index to the total evaluation function to balance the smoothness of the robot. The definition of k is shown in the following formula:

[0111]

[0112] where v is the linear velocity of the robot and ω is the angular velocity of the robot.

[0113] Similar to curvature, the more frequently the robot turns and the less smooth the trajectory is, the larger k is; conversely, the smaller k is. Therefore, by adding the new smoothness evaluation function k to the evaluation function, the robot can be encouraged to choose a smoother path.

[0114] The total evaluation function after adding k is shown in the following formula:

[0115] G(v, ω) = δ · (β · dist(v, ω) + γ · heading(v, ω) + σ · v(v) + ρ · k(v, ω)) (26)

[0116] where ρ is the weight factor of function k.

[0117] To verify the timeliness and flexibility of the method proposed in this application, simulation experiments on path planning are respectively carried out for the traditional DWA algorithm, the Q-DWA algorithm, and the complete improved scheme in two different environments, and then various indicators are compared. The operating environment of the simulation experiment selected in this application is MATLAB R2022a.

[0118] First, the simulation results of the Q-DWA algorithm are compared with the simulation results of the traditional DWA algorithm, and then the simulation results of the Q-DWA algorithm are compared with the simulation results of the complete improved scheme. Fig. 4(a) is the path diagram generated by traditional DWA path planning, Fig. 4(b) is the path diagram generated by Q-DWA path planning, and Fig. 4(c) is the path diagram generated by the complete improved DWA algorithm path planning. The red obstacles in the figure are static obstacles, and the black obstacles are dynamic obstacles. This application designs that there are 2 dynamic obstacles and 9 static obstacles in Environment 1.

[0119] As can be seen from Figures 4(a) and 4(b), when the robot passes by the obstacle at the path coordinate axis (6,5), the robot in the traditional DWA path planning is significantly too close to the obstacle, which will cause the robot to easily collide with the obstacle and also result in unnecessary turning. The Q-DWA algorithm designed in this application plans a more reasonable path compared to the traditional DWA, which greatly improves the robot's ability to avoid obstacles successfully and also has better real-time performance. Table 1 shows the operation results in Environment 1.

[0120] Table 1 Operation Results in Environment 1

[0121] Algorithm Algorithm running time / s Optimal path length / m Traditional DWA 64.550 14.49 Q-DWA 49.539 14.48 Full improved DWA 47.678 14.44

[0122] As can be seen from Table 1, the Q-DWA algorithm designed in this application plans a shorter path in a shorter time compared to the traditional DWA algorithm, improving the speed of DWA path planning.

[0123] As can be seen from Figures 4(b) and 4(c), adding a new evaluation function k to the Q-DWA algorithm in this application can enhance the smoothness of the generated path. As can be seen from Table 1, this new evaluation function not only makes the path smoother, but also reduces the algorithm running time and shortens the optimal path length. Generally speaking, compared with the traditional DWA algorithm, the improved DWA algorithm reduces the running time by 16.872 s and the path length by 0.05 m.

[0124] In Environment 2, 2 dynamic obstacles are added and the positions of the obstacles also change. The simulation results of each algorithm are still compared. Figure 5(a) is the path map generated by the traditional DWA path planning, Figure 5(b) is the path map generated by the Q-DWA path planning, and Figure 5(c) is the path map generated by the fully improved DWA algorithm path planning. Table 2 shows the operation results in Environment 2.

[0125] Table 2 Operation Results in Environment 2

[0126] Algorithm Algorithm running time / s Optimal path length / m Traditional DWA 98.758 15.87 Q-DWA 78.239 15.32 Full improved DWA 66.415 15.00

[0127] The experimental results show that the traditional DWA algorithm has a poor path planning effect in Environment 2. The distance between the robot and the obstacle is short when the robot passes by the path coordinate axis (6,4), which easily causes the robot to collide with the obstacle and also makes the robot make unnecessary turns, generating additional paths. Compared with the traditional DWA algorithm in Environment 2, the Q-DWA algorithm generates a safer and more effective path. As can be seen from Table 2, the Q-DWA algorithm can still generate a shorter path in a shorter time in Environment 2, proving that using Q-learning to dynamically adjust the evaluation weight parameters of DWA can enhance the obstacle avoidance ability and real-time performance of the algorithm.

[0128] As can be seen from FIGS. 5(b) and 5(c), near the path coordinate axis (5, 5), the path generated by the algorithm with the evaluation function k added is smoother than that generated without adding the evaluation function k; and when passing between the two obstacles at (6, 2) and (6, 4), the path generated by the algorithm is also safer after adding the evaluation function k. As can be seen from Table 2, the path generated by the improved DWA algorithm is shortened by 5.48% compared with the traditional DWA algorithm, and the running time of the algorithm is reduced by 32.74%.

[0129] As can be seen from these two experiments, the Q-DWA algorithm designed in this application has enhanced adaptability to the dynamic environment and improved real-time performance and flexibility of path planning compared with the traditional DWA algorithm. On this basis, the added new evaluation function can make the path generated by the algorithm smoother.

[0130] In summary, the local dynamic path planning method based on the dynamic window method provided by the present invention has made a number of innovations and improvements on the basis of the traditional dynamic window method. A method (Q-DWA) for dynamically adjusting the weight parameters of the evaluation function based on Q-learning is proposed, and a Markov decision process (MDP) model is established, including the design of the state set, action set and reward function. This method can continuously adjust the weight parameters of the evaluation function to appropriate values according to the changes of the environmental state. On this basis, a new evaluation function is added, borrowing the idea of curvature, which can guide the robot to choose a smoother path. The experimental results show that the Q-DWA algorithm designed in this application can improve the adaptability of the robot to the environment and the flexibility of path planning; the newly added evaluation function can make the robot generate a smoother path.

[0131] Obviously, those skilled in the art can make various changes and modifications to the invention without departing from the spirit and scope of the invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A local dynamic path planning method based on the dynamic window approach, characterized in that It includes the following steps: Step 1: Construct the kinematic model of the robot; Step 2: Conduct speed sampling, and simulate the sampled speed combinations based on the kinematic model to obtain multiple simulated paths; Step 3: Construct an evaluation function in the dynamic window method iteration, dynamically adjust the weight parameters of the evaluation function through reinforcement learning, and obtain the scoring results of the simulated paths based on the evaluation function; Step 4: Compare the scoring results of all simulated paths, and select the driving path of the robot at the next time step based on the comparison results.

2. The local dynamic path planning method based on the dynamic window approach according to claim 1, characterized in that In Step 1, the kinematic model is an arc model.

3. The local dynamic path planning method based on the dynamic window approach according to claim 1, characterized in that In Step 3, dynamically adjusting the weight parameters of the evaluation function through reinforcement learning includes: based on the evaluation function, establishing a Markov decision process model.

4. The local dynamic path planning method based on the dynamic window approach according to claim 3, characterized in that Establishing a Markov decision process model includes: designing a state set, designing an action set, and designing a reward and punishment function.

5. The local dynamic path planning method based on the dynamic window approach according to claim 4, characterized in that, The vector representation s of the state set is: s = [s1, s2, s3] where s1 represents the state associated with the distance between the robot and the nearest obstacle, s2 represents the state associated with the distance between the current position of the robot and the target point, and s3 represents the linear velocity state of the robot.

6. The local dynamic path planning method based on the dynamic window method according to claim 5, wherein, The action set is the adjustment of the three weight parameters, and the weight parameters are the weight of the target direction, the weight of the obstacle distance, and the weight of the movement speed.

7. The local dynamic path planning method based on the dynamic window approach according to claim 6, characterized in that The reward and punishment function R is: R = R1 + R2 + R3 where R1 is the reward related to the distance to the nearest obstacle, R2 is the reward related to whether approaching the target, and R3 is the reward related to whether reaching the target or colliding with an obstacle.

8. The local dynamic path planning method based on the dynamic window approach according to claim 7, wherein The evaluation function G(v, ω) is: G(v, ω) = δ·(β·dist(v, ω) + γ·heading(v, ω) + σ·v(v) + ρ·k(v, ω)) where dist(v, ω) is the obstacle distance evaluation function, heading(v, ω) is the target direction evaluation function, v(v) is the speed evaluation function, k(v, ω) is the smoothness evaluation function, β, γ, σ, ρ are the weight factors of the obstacle distance, target direction, movement speed, and smoothness respectively, and δ is the normalization coefficient.

9. The local dynamic path planning method based on the dynamic window approach according to claim 8, characterized in that The calculation formula of the smoothness evaluation function k(v, ω) is: where v is the linear velocity of the robot and ω is the angular velocity of the robot.

10. The local dynamic path planning method based on the dynamic window approach according to claim 9, characterized in that, In Step 4, the path with the highest scoring result is used as the driving path of the robot at the next time step, and the corresponding linear velocity v and angular velocity ω are used as the speed control commands of the robot at the next moment.

Images