Unmanned ship dynamic obstacle avoidance control method

Abstract

The invention discloses a dynamic obstacle avoidance control method for an unmanned ship, and the method comprises the steps: multi-sensor data fusion and environment modeling, dynamic obstacle trajectory prediction, deep reinforcement learning path planning, nonlinear model prediction control, and final planning algorithm summarization. Obstacle avoidance control is carried out through multi-sensor data fusion and environment modeling, dynamic obstacle trajectory prediction, deep reinforcement learning path planning, nonlinear model prediction control and final planning algorithm summarization, and the method comprises multi-modal sensor fusion, LSTM trajectory prediction, deep reinforcement learning decision and nonlinear model prediction control. High-precision real-time obstacle avoidance under a complex water area is achieved, the whole method combines dynamic modeling and optimization control, the path safety, efficiency and energy utilization rate are remarkably improved, and the method is suitable for the fields of agricultural unmanned feeding, unmanned water quality intelligent detection and the like.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of unmanned ships, and in particular to a dynamic obstacle avoidance control method for unmanned ships. Background Art

[0002] As water scenarios become increasingly complex and operational tasks become increasingly diverse, autonomous navigation has become one of the key development trends in future water operations and an inevitable requirement for informationization and intelligentization. Furthermore, as unmanned driving technology matures, surface unmanned driving technology is attracting increasing attention and attention from scholars.

[0003] Traditional unmanned boat obstacle avoidance methods have the following technical defects:

[0004] 1. Insufficient environmental perception capabilities: Relying on a single sensor, such as radar or camera, which is susceptible to interference from rain, fog, and waves, and has a dynamic obstacle miss detection rate of >25%;

[0005] 2. Rigid obstacle avoidance strategy: Based on static path planning, it cannot adapt to sudden obstacles (such as fast-moving ships);

[0006] 3. Poor real-time performance: Global path replanning takes more than 3 seconds, and collisions are prone to occur during emergency obstacle avoidance. Summary of the Invention

[0007] 1. Technical problem to be solved by the invention:

[0008] The present invention provides a dynamic obstacle avoidance control method for an unmanned vessel, which is used to solve the technical problems existing in the above-mentioned background technology.

[0009] 2. Technical solution:

[0010] In order to achieve the above object, the technical solution provided by the present invention is:

[0011] A method for dynamic obstacle avoidance control of an unmanned vessel, comprising the following steps:

[0012] S1. Multi-sensor data fusion and environmental modeling: Based on the driving path acquired before the flight, a sensing array consisting of lidar, millimeter-wave radar, binocular vision, and multi-beam sonar simultaneously collects on-site environmental data, pan-tilt meteorological data, and global track information generated by the planning algorithm to build an environmental and navigation model. Specifically, this includes:

[0013] Sensor configuration and data synchronization, multi-sensor spatiotemporal synchronization is achieved through timestamp alignment and spatial coordinate transformation. The formula is:

[0014]

[0015] in, is the sensor extrinsic parameter Lie algebra, is the transformation matrix from the base coordinate system to the world coordinate system;

[0016] Environmental grid map construction;

[0017] S2. Dynamic obstacle trajectory prediction: Based on the LSTM network, obstacle trajectory prediction is performed, outputting the position distribution and confidence interval for the next 5 seconds. The presence of obstacles and obstacle avoidance solutions are determined based on the site navigation model. The dynamic obstacle trajectory prediction includes trajectory feature extraction and uncertainty quantification. According to the obstacle information transmitted by the perception system, trajectory feature extraction is performed, and an obstacle motion model is constructed based on the LSTM long short-term memory network. The historical trajectory sequence is input. (T = 10 seconds), output the position prediction for the next τ = 5 seconds The network structure is as follows:

[0018] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0019] i t =σ(W i ·[h t-1 ,x t ]+b i );c t =f t ☉c t-1 +i t ⊙tanh(W c ·[h t-1 ,x t ]+b c );o t =σ(W o ·[h t-1 ,x t ]+b o );h t =o t ☉tanh(c t )

[0020] Where σ is the Sigmoid function and ⊙ is the Hadamard product;

[0021] S3, Deep Reinforcement Learning Path Planning: including state-action space modeling, multi-objective reward function and strategy optimization and deployment, where the state space is marked as S Including the unmanned ship pose x = [x, y, ψ, u, v, r] T , target point g, obstacle prediction trajectory And environmental disturbance d=[w wind , w current ]T , the action space is marked as Heading angle increment Δψ[-30°, 30°], propulsion force F∈[0, F max ]、Transverse thrust T lat ∈[-T max , T max ], the multi-objective optimization function is path tracking, obstacle avoidance, safety, and energy consumption, and the reward function formula is:

[0022]

[0023] Where λ1 = 0.5, λ2 = 1.2, λ3 = 0.3, σ = 5m; the strategy optimization and deployment is to use the PPO algorithm to train the strategy network π θ (a|s);

[0024] S4. Nonlinear Model Predictive Control: This system solves the optimal control input under the ship's dynamic constraints in real time, performs nonlinear prediction of the ship's travel path, establishes a nonlinear model, and predictively controls unknown travel paths. The nonlinear model predictive control includes ship dynamics modeling and rolling horizon optimization.

[0025] S5. Final planning algorithm summary: The model prediction and optimization solution algorithm obtained by the above steps are handed over to the propulsion control module of the unmanned ship, so that it controls the state of the ship to achieve the calculated desired velocity vector direction.

[0026] As a preferred technical solution of the present invention, the sensor configuration in step S1 has a data-synchronized laser radar detection distance of 200m and an accuracy of ±0.1m. The millimeter-wave radar is a rain and fog interference-resistant radar, the binocular vision has a resolution of 3840×2160 and a frame rate of 30fps, and the multi-beam sonar has an underwater detection depth of 50m.

[0027] As a preferred technical solution of the present invention, the environmental grid map construction in step S1 is to fuse the perception data into a 2.5D grid map with a resolution of 0.2m×0.2m×0.1m, and each grid contains an obstacle probability P obs ∈[0, 1] and motion velocity vector

[0028] The update formula is:

[0029]

[0030] Among them, α=0.7 is the attenuation factor, Detection indicator function for the current frame.

[0031] As a preferred technical solution of the present invention, the uncertainty quantization in step S2 is to generate N=100 samples through Monte Carlo Dropout and calculate the predicted position confidence interval:

[0032]

[0033] Among them, z 0.95 =1.96 is the Z value at the 95% confidence level.

[0034] As a preferred technical solution of the present invention, the strategy network π of the strategy optimization and deployment in step S3 θ (a|s) The objective function is:

[0035]

[0036] Among them, ∈ = 0.2 is the cutoff threshold and A is the advantage function.

[0037] As a preferred technical solution of the present invention, the ship dynamics modeling in step S4 adopts a three-degree-of-freedom dynamics equation, and the three-degree-of-freedom dynamics equation formula is as follows:

[0038]

[0039] Where m is the mass, is the hydrodynamic derivative, and l is the thrust arm length.

[0040] As a preferred technical solution of the present invention, the rolling time domain optimization in step S4 is performed in the prediction time domain T p = Solve the following optimization problem within 5 seconds, and the rolling time domain optimization formula is:

[0041]

[0042] Among them, Q=diag(10,10,5,1,1,1), R=diag(0.1,0.1).

[0043] As a preferred technical solution of the present invention, the model constructed by multi-sensor data fusion and environmental modeling in step S1 is a spatiotemporal alignment model of lidar point cloud-visual semantics-hydrological geographic information.

[0044] As a preferred technical solution of the present invention, the multi-sensor data fusion and the environmental modeling data fusion in step S1 adopt a Bayesian network to realize multi-source data fusion, and the fusion formula is as follows:

[0045]

[0046] Where E is the environmental state hypothesis and z is the observation data of each sensor.

[0047] 3.Beneficial effects:

[0048] The technical solution provided by the present invention has the following beneficial effects compared with the existing technology: obstacle avoidance control is performed through multi-sensor data fusion and environment modeling, dynamic obstacle trajectory prediction, deep reinforcement learning path planning, nonlinear model predictive control, and final planning algorithm summary. The method achieves high-precision real-time obstacle avoidance in complex waters through multimodal sensor fusion, LSTM trajectory prediction, deep reinforcement learning decision-making and nonlinear model predictive control. The entire method combines dynamic modeling and optimization control to significantly improve path safety, efficiency and energy utilization, and is suitable for fields such as marine exploration and intelligent shipping. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A method diagram of the present invention. DETAILED DESCRIPTION

[0050] To facilitate understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0051] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "page", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore should not be understood as limiting the present invention.

[0052] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0053] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connected," "fixed," "provided with," and the like should be understood in a broad sense. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0054] Example

[0055] See also Figure 1 ,

[0056] Example 1

[0057] The present invention provides a dynamic obstacle avoidance control method for an unmanned vessel, the method comprising the following steps:

[0058] S1. Multi-sensor data fusion and environmental modeling: Before navigation, the driving path is obtained, and on-site environmental data and PTZ weather data are obtained through multiple sensors. The global track information calculated by the planning algorithm is used to establish the environment and navigation model. The multi-sensor data fusion and environmental modeling include sensor configuration and data synchronization and environmental grid map construction. The sensor configuration and data synchronization are achieved by forming a sensing array of laser radar, millimeter wave radar, binocular vision and multi-beam sonar, and using timestamp alignment and spatial coordinate transformation to achieve spatiotemporal synchronization of multi-sensor data. The formula is:

[0059]

[0060] in, is the sensor extrinsic parameter Lie algebra, is the transformation matrix from the base coordinate system to the world coordinate system;

[0061] S2. Dynamic obstacle trajectory prediction: Perform obstacle prediction for autonomous navigation. Obstacle trajectory prediction based on LSTM network outputs the position distribution and confidence interval for the next 5 seconds. The presence of obstacles and obstacle avoidance solutions are determined based on the site navigation model. The dynamic obstacle trajectory prediction includes trajectory feature extraction and uncertainty quantification. Trajectory feature extraction is performed based on obstacle information transmitted by the perception system. The obstacle motion model is constructed based on LSTM long short-term memory network. The historical trajectory sequence is input. (T = 10 seconds), output the position prediction for the next τ = 5 seconds The network structure is as follows:

[0062] f t =σ(W f ·[h t-1 ,xt ]+b f );

[0063] i t =σ(W i ·[h t-1 ,x t ]+b i );c t =f t ☉c t-1 +i t ⊙tanh(W c ·[h t-1 ,x t ]+b c );o t =σ(W o ·[h t-1 ,x t ]+b o );h t =o t ☉tanh(c t )

[0064] Where σ is the Sigmoid function and ⊙ is the Hadamard product;

[0065] S3, deep reinforcement learning path planning; the reward function integrates path tracking, obstacle avoidance safety and energy consumption optimization goals. For collision situations, further plan calculation optimization is carried out, and the optimized plan is archived. The best several sets of plans are listed and uploaded to the staff for subdivision and consolidation. The deep reinforcement learning path planning includes state-action space modeling, multi-objective reward function and strategy optimization and deployment, and the state space is marked as Including the unmanned ship pose x = [x, y, ψ, u, v, r] T , target point g, obstacle prediction trajectory And environmental disturbance d=[w wind , w current ] T , the action space is marked as The heading angle increment Δψ∈[-30°, 30°], the propulsion force F∈[0, F max ]、Transverse thrust T lat ∈[-T max , T max ], the multi-objective optimization function is path tracking, obstacle avoidance, safety, and energy consumption, and the reward function formula is:

[0066]

[0067] Where λ1 = 0.5, λ2 = 1.2, λ3 = 0.3, σ = 5m; the strategy optimization and deployment is to use the PPO algorithm to train the strategy network π θ (a|s);

[0068] S4. Nonlinear Model Predictive Control: This system solves the optimal control input under the ship's dynamic constraints in real time, performs nonlinear prediction of the ship's travel path, establishes a nonlinear model, and predictively controls unknown travel paths. The nonlinear model predictive control includes ship dynamics modeling and rolling horizon optimization.

[0069] S5. Final planning algorithm summary: The model prediction and optimization solution algorithm obtained by the above steps are handed over to the propulsion control module of the unmanned ship, so that it controls the state of the ship to achieve the calculated desired velocity vector direction;

[0070] The target point of the initial condition: (x g ,y g )=(900m, 800m), the initial position of the unmanned ship is: [x0, y0, ψ0]=[0, 0, 45°], the speed is u=3m / s; the dynamic obstacle is approached by two ships at an oblique speed of 4m / s and 2m / s respectively;

[0071] Trajectory prediction: LSTM outputs the obstacle's position in the next 5 seconds, with the semi-major axis of the confidence interval ≤ 0.5m;

[0072] Path planning: Deep reinforcement learning generates an avoidance path, adjusts the heading to 60°, and increases the speed to 3.8m / s;

[0073] Control execution: NMPC solver outputs propulsion force F = 1200N, lateral thrust T lat =300N;

[0074] The deviation between the actual track and the theoretical path is less than 0.8m, there is no collision throughout the entire process, and energy consumption is reduced by 32% compared with traditional methods.

[0075] The sensor configuration and data synchronization of the laser radar in step S1 have a detection range of 200m and an accuracy of ±0.1m. The millimeter wave radar is a rain and fog interference-resistant radar, the binocular vision has a resolution of 3840×2160 and a frame rate of 30fps, and the multi-beam sonar has an underwater detection depth of 50m. The environmental grid map construction in step S1 is to fuse the perception data into a 2.5D grid map with a resolution of 0.2m×0.2m×0.1m. Each grid contains an obstacle probability P obs ∈[0, 1] and motion velocity vector The update formula is:

[0076]

[0077] Among them, α=0.7 is the attenuation factor, Detection indicator function for the current frame.

[0078] In step S2, the uncertainty quantification is performed by generating N=100 samples through Monte Carlo Dropout and calculating the confidence interval of the predicted position:

[0079]

[0080] Among them, z 0.95 =1.96 is the Z value at the 95% confidence level.

[0081] The strategy network π of the strategy optimization and deployment in step S3 θ (a|s) The objective function is:

[0082]

[0083] Among them, ∈ = 0.2 is the cutoff threshold and A is the advantage function.

[0084] In step S4, the ship dynamics modeling adopts a three-degree-of-freedom dynamics equation, and the three-degree-of-freedom dynamics equation formula is as follows:

[0085]

[0086] Where m is the mass, is the hydrodynamic derivative, and l is the thrust arm length.

[0087] The rolling time domain optimization in step S4 is performed in the prediction time domain T p = Solve the following optimization problem within 5 seconds, and the rolling time domain optimization formula is:

[0088]

[0089] Among them, Q=diag(10,10,5,1,1,1), R=diag(0.1,0.1).

[0090] The model constructed by multi-sensor data fusion and environmental modeling in step S1 is a spatiotemporal alignment model of lidar point cloud-visual semantics-hydrogeographic information. The multi-sensor data fusion and environmental modeling data fusion in step S1 adopt Bayesian network to realize multi-source data fusion, and the fusion formula is as follows:

[0091]

[0092] Where E is the environmental state hypothesis and z is the observation data of each sensor.

[0093] Example 2

[0094] The present invention provides a dynamic obstacle avoidance control method for an unmanned vessel, the method comprising the following steps:

[0095] S1. Multi-sensor data fusion and environmental modeling: Before navigation, the driving path is obtained, and on-site environmental data and PTZ weather data are obtained through multiple sensors. The global track information calculated by the planning algorithm is used to establish the environment and navigation model. The multi-sensor data fusion and environmental modeling include sensor configuration and data synchronization and environmental grid map construction. The sensor configuration and data synchronization are achieved by forming a sensing array of laser radar, millimeter wave radar, binocular vision and multi-beam sonar, and using timestamp alignment and spatial coordinate transformation to achieve spatiotemporal synchronization of multi-sensor data. The formula is:

[0096]

[0097] in, is the sensor extrinsic parameter Lie algebra, is the transformation matrix from the base coordinate system to the world coordinate system;

[0098] S2. Dynamic obstacle trajectory prediction: Perform obstacle prediction for autonomous navigation. Obstacle trajectory prediction based on LSTM network outputs the position distribution and confidence interval for the next 5 seconds. The presence of obstacles and obstacle avoidance solutions are determined based on the site navigation model. The dynamic obstacle trajectory prediction includes trajectory feature extraction and uncertainty quantification. Trajectory feature extraction is performed based on obstacle information transmitted by the perception system. The obstacle motion model is constructed based on LSTM long short-term memory network. The historical trajectory sequence is input. (T = 10 seconds), output the position prediction for the next τ = 5 seconds The network structure is as follows:

[0099] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0100] i t =σ(W i ·[h t-1 ,x t ]+b i );c t =f t ☉c t-1 +i t ☉tanh(W c ·[h t-1 ,x t ]+b c );o t =σ(W o ·[h t-1,x t ]+b o );h t =o t ☉tanh(c t )

[0101] Where σ is the Sigmoid function and ⊙ is the Hadamard product;

[0102] S3, deep reinforcement learning path planning; the reward function integrates path tracking, obstacle avoidance safety and energy consumption optimization goals. For collision situations, further plan calculation optimization is carried out, and the optimized plan is archived. The best several sets of plans are listed and uploaded to the staff for subdivision and consolidation. The deep reinforcement learning path planning includes state-action space modeling, multi-objective reward function and strategy optimization and deployment, and the state space is marked as Including the unmanned ship pose x = [x, y, ψ, u, v, r] T , target point g, obstacle prediction trajectory And environmental disturbance d=[w wind , w current ] T , the action space is marked as The heading angle increment Δψ∈[-30°, 30°], the propulsion force F∈[0, F max ]、Transverse thrust T lat ∈[-T max , T max ], the multi-objective optimization function is path tracking, obstacle avoidance, safety, and energy consumption, and the reward function formula is:

[0103]

[0104] Where λ1 = 0.5, λ2 = 1.2, λ3 = 0.3, σ = 5m; the strategy optimization and deployment is to use the PPO algorithm to train the strategy network π θ (a|s);

[0105] S4. Nonlinear Model Predictive Control: This system solves the optimal control input under the ship's dynamic constraints in real time, performs nonlinear prediction of the ship's travel path, establishes a nonlinear model, and predictively controls unknown travel paths. The nonlinear model predictive control includes ship dynamics modeling and rolling horizon optimization.

[0106] S5. Final planning algorithm summary: The model prediction and optimization solution algorithm obtained by the above steps are handed over to the propulsion control module of the unmanned ship, so that it controls the state of the ship to achieve the calculated desired velocity vector direction;

[0107] Sensor compensation: Millimeter-wave radar dominates perception, and the lidar point cloud is denoised through Kalman filtering;

[0108] Emergency obstacle avoidance: Detecting a sudden floating object (15m away) triggers NMPC emergency braking (F = -800N);

[0109] Energy consumption statistics: The total energy consumption is 2.1×10 6 J, a 28% reduction compared to the baseline algorithm.

[0110] The sensor configuration and data synchronization of the laser radar in step S1 have a detection range of 200m and an accuracy of ±0.1m. The millimeter wave radar is a rain and fog interference-resistant radar, the binocular vision has a resolution of 3840×2160 and a frame rate of 30fps, and the multi-beam sonar has an underwater detection depth of 50m. The environmental grid map construction in step S1 is to fuse the perception data into a 2.5D grid map with a resolution of 0.2m×0.2m×0.1m. Each grid contains an obstacle probability P obs ∈[0, 1] and motion velocity vector The update formula is:

[0111]

[0112] Among them, α=0.7 is the attenuation factor, Detection indicator function for the current frame.

[0113] In step S2, the uncertainty quantification is performed by generating N=100 samples through Monte Carlo Dropout and calculating the confidence interval of the predicted position:

[0114]

[0115] Among them, z 0.95 =1.96 is the Z value at the 95% confidence level.

[0116] The strategy network π of the strategy optimization and deployment in step S3 θ (a|s) The objective function is:

[0117]

[0118] Among them, ∈ = 0.2 is the cutoff threshold and A is the advantage function.

[0119] In step S4, the ship dynamics modeling adopts a three-degree-of-freedom dynamics equation, and the three-degree-of-freedom dynamics equation formula is as follows:

[0120]

[0121] Where m is the mass, is the hydrodynamic derivative, and l is the thrust arm length.

[0122] The rolling time domain optimization in step S4 is performed in the prediction time domain T p = Solve the following optimization problem within 5 seconds, and the rolling time domain optimization formula is:

[0123]

[0124] Among them, Q=diag(10,10,5,1,1,1), R=diag(0.1,0.1).

[0125] The model constructed by multi-sensor data fusion and environmental modeling in step S1 is a spatiotemporal alignment model of lidar point cloud-visual semantics-hydrogeographic information. The multi-sensor data fusion and environmental modeling data fusion in step S1 adopt Bayesian network to realize multi-source data fusion, and the fusion formula is as follows:

[0126]

[0127] Where E is the environmental state hypothesis and z is the observation data of each sensor.

[0128] In specific use, the present invention provides a dynamic obstacle avoidance control method for an unmanned ship, which performs obstacle avoidance control through multi-sensor data fusion and environment modeling, dynamic obstacle trajectory prediction, deep reinforcement learning path planning, nonlinear model predictive control, and final planning algorithm summary, thereby improving the unmanned ship's driving perception accuracy, improving the obstacle detection rate, optimizing the obstacle avoidance scheme, and reducing the obstacle avoidance response time. The fine route planning reduces energy consumption efficiency. The method achieves high-precision real-time obstacle avoidance in complex waters through multimodal sensor fusion, LSTM trajectory prediction, deep reinforcement learning decision-making and nonlinear model predictive control. The entire method combines dynamic modeling and optimization control to significantly improve path safety, efficiency and energy utilization, and is suitable for marine exploration, intelligent shipping and other fields.

[0129] The above-mentioned embodiments only express a certain implementation method of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the patent scope of the present invention. It should be pointed out that for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent of the present invention shall be based on the attached claims.

[0130] It should be noted that the above content belongs to the technical knowledge scope of the inventor. Since the technical content in this field is vast and too complicated, the above content of this application does not necessarily constitute prior art.

Claims

1. A dynamic obstacle avoidance control method for an unmanned vessel, characterized in that: The method comprises the following steps: S1. Multi-sensor data fusion and environmental modeling: Based on the driving path acquired before the flight, a sensing array consisting of lidar, millimeter-wave radar, binocular vision, and multi-beam sonar simultaneously collects on-site environmental data, pan-tilt meteorological data, and global track information generated by the planning algorithm to build an environmental and navigation model. Specifically, this includes: Sensor configuration and data synchronization, multi-sensor spatiotemporal synchronization is achieved through timestamp alignment and spatial coordinate transformation. The formula is: in, is the sensor extrinsic parameter Lie algebra, is the transformation matrix from the base coordinate system to the world coordinate system; Environmental grid map construction; S2. Dynamic obstacle trajectory prediction: Based on the LSTM network, obstacle trajectory prediction is performed, outputting the position distribution and confidence interval for the next 5 seconds. The presence of obstacles and obstacle avoidance solutions are determined based on the location navigation model. The dynamic obstacle trajectory prediction includes trajectory feature extraction and uncertainty quantification. According to the obstacle information transmitted by the perception system, trajectory features are extracted, and an obstacle motion model is constructed based on the LSTM long short-term memory network. The historical trajectory sequence is input. (T = 10 seconds), output the position prediction for the next τ = 5 seconds The network structure is as follows: f t =σ(W f ·[h t-1 ,x t ]+b f ); i t =σ(W i ·[h t-1 ,x t ]+b i );c t =f t ⊙c t-1 +i t ⊙tanh(W c ·[h t-1 ,x t ]+b c );o t =σ(W o ·[h t-1 ,x t ]+b o );h t =o t ⊙tanh(c t ) Where σ is the Sigmoid function and ⊙ is the Hadamard product; S3, Deep Reinforcement Learning Path Planning: including state-action space modeling, multi-objective reward function and strategy optimization and deployment, where the state space is marked as Including the unmanned ship pose x = [x, y, ψ, u, v, r] T , target point g, obstacle prediction trajectory And environmental disturbance d=[w wind , w current ] T , the action space is marked as The heading angle increment Δψ∈[-30°, 30°], the propulsion force F∈[0, F max ]、Transverse thrust T lat ∈[-T max , T max ], the multi-objective optimization function is path tracking, obstacle avoidance, safety, and energy consumption, and the reward function formula is: Where λ1 = 0.5, λ2 = 1.2, λ3 = 0.3, σ = 5m; the strategy optimization and deployment is to use the PPO algorithm to train the strategy network π θ (a|s); S4. Nonlinear Model Predictive Control: This system solves the optimal control input under the ship's dynamic constraints in real time, performs nonlinear prediction of the ship's travel path, establishes a nonlinear model, and predictively controls unknown travel paths. The nonlinear model predictive control includes ship dynamics modeling and rolling horizon optimization. S5. Final planning algorithm summary: The model prediction and optimization solution algorithm obtained by the above steps are handed over to the propulsion control module of the unmanned ship, so that it controls the state of the ship to achieve the calculated desired velocity vector direction.

2. The unmanned vessel dynamic obstacle avoidance control method according to claim 1, characterized in that: The sensor configuration in step S1 and the data-synchronized laser radar have a detection range of 200m and an accuracy of ±0.1m. The millimeter-wave radar is a rain-and-fog interference-resistant radar, the binocular vision has a resolution of 3840×2160 and a frame rate of 30fps, and the multi-beam sonar has an underwater detection depth of 50m.

3. The unmanned vessel dynamic obstacle avoidance control method according to claim 1, characterized in that: The construction of the environmental grid map in step S1 is to fuse the perception data into a 2.5D grid map with a resolution of 0.2m×0.2m×0.1m, and each grid contains an obstacle probability P obs ∈[0, 1] and motion velocity vector The update formula is: Among them, α=0.7 is the attenuation factor, Detection indicator function for the current frame.

4. The unmanned vessel dynamic obstacle avoidance control method according to claim 1, characterized in that: The uncertainty quantification in step S2 is to generate N=100 samples through Monte Carlo Dropout and calculate the confidence interval of the predicted position: Among them, z 0.95 =1.96 is the Z value at the 95% confidence level.

5. The unmanned vessel dynamic obstacle avoidance control method according to claim 1, characterized in that: The strategy network π of the strategy optimization and deployment in step S3 θ (a|s) The objective function is: Among them, ∈ = 0.2 is the cutoff threshold and A is the advantage function.

6. The unmanned vessel dynamic obstacle avoidance control method according to claim 1, characterized in that: The ship dynamics modeling in step S4 adopts a three-degree-of-freedom dynamics equation, and the three-degree-of-freedom dynamics equation formula is as follows: Where m is the mass, is the hydrodynamic derivative, and l is the thrust arm length.

7. The unmanned vessel dynamic obstacle avoidance control method according to claim 1, characterized in that: The rolling time domain optimization in step S4 is performed in the prediction time domain T p = Solve the following optimization problem within 5 seconds, and the rolling time domain optimization formula is: Among them, Q=diag(10,10,5,1,1,1), R=diag(0.1,0.1).

8. The unmanned vessel dynamic obstacle avoidance control method according to claim 1, characterized in that: The model constructed by multi-sensor data fusion and environmental modeling in step S1 is a spatiotemporal alignment model of lidar point cloud-visual semantics-hydrographic information.

9. The unmanned vessel dynamic obstacle avoidance control method according to claim 1, characterized in that: In step S1, the multi-sensor data fusion and the environmental modeling data fusion adopt the Bayesian network to realize multi-source data fusion, and the fusion formula is as follows: Where E is the environmental state hypothesis and z is the observation data of each sensor.

Images