38 results about "Collision model" patented technology

The invention discloses a visual perception and dynamic anti-collision intelligent closed-loop control method for a central control mechanical arm in a vehicle, and the method comprises the steps: S1, loading an improved YOLOv11 network detection model, and reading a video stream of a camera; s2, preprocessing each frame of image in the video stream, inputting the preprocessed image into the model, and performing target identification; s3, inputting an identification result into a dynamic safe area collision model, and performing coordinate conversion through a hand-eye calibration algorithm; calculating a relative position and a motion trend in combination with depth sensor data, and judging a risk level to generate a corresponding protection instruction; and S4, executing a hierarchical control strategy according to the risk level, updating the action track of the mechanical arm, and returning to the step S1 through a closed-loop feedback mechanism to adjust the safety threshold and the control parameters to form a complete'perception-decision-execution-optimization 'autonomous evolution closed loop. According to the method, the visual perception robustness and the target detection precision under the complex illumination condition are improved, and real-time dynamic anti-collision response and millisecond-level rapid obstacle avoidance are realized.

Provided are a planning and control method for a legged robot, an apparatus, a robot, and a storage medium. The method includes: determining, based on the movement state and the swing parameter of each leg at the next moment, a foot-end reference state of each leg, and planning, based on a foot-end dynamic model and a discrete collision model, an impact-aware swing leg trajectory for a leg that starts to swing at the next moment; and performing, based on the movement state, the reference state sequence, and the foot-end reference state, whole-body control on the legged robot, and performing, based on a whole-body dynamic model and the discrete collision model, impact-aware whole-body control for a leg that is a supporting leg in planning but does not touch the ground.

The invention relates to the technical field of aviation safety, in particular to a short-distance parallel runway collision risk assessment method based on influence of front and side wind, which comprises collision risk assessment, Bayesian parameter estimation, linear regression analysis, dynamic front and side wind fitting and Monte Carlo analog simulation. The problem of small sample parameter estimation is solved by introducing a Bayesian parameter estimation method, crosswind parameters are calculated by combining linear regression analysis, positive crosswind data are fitted by applying Weibull distribution, and an influence rule of positive crosswind and an initial interval on collision probability is quantified through Monte Carlo simulation. According to the short-distance parallel runway collision risk assessment method based on the influence of the positive and crosswind, on the basis of a traditional collision model, a re-flight condition is incorporated into the model, complex scenes such as crosswind and re-flight are considered, a dynamic interval adjustment model is fused, and a new collision risk model is established; the crosswind grading minimum interval standard is provided for airline companies and air traffic control operation management, and the runway safety and the airport operation capacity are improved.

The application discloses a simulation method of output characteristics of a three-state inertial switch, comprising the following steps: S1, establishing an inertial model based on the structure of an inertial system in the three-state inertial switch and performing simulation and solution to obtain the mechanical response of a mass block in the three-state inertial switch; S2, establishing a collision model of a collision process of the mass block from a front collision to an output electrode based on the contact relationship between the mass block and the output electrode and performing simulation and solution to obtain an invasion displacement time history; and S3, establishing an electromechanical conversion model and taking the invasion displacement time history of the mass block and the electrode as the input of the electromechanical conversion model, and performing simulation on the electromechanical conversion model to obtain the voltage time history of the output electrode as the output characteristics of the three-state inertial switch. The application can accurately simulate the output characteristics of the three-state inertial switch under different penetration environments.

The invention relates to the technical field of automobile safety performance development, in particular to a column collision sliding table test boundary parameter optimization method. Comprising the following steps: extracting related parts from a vehicle column collision passenger protection finite element collision model to obtain a simplified model; establishing a simulation column collision sliding table test finite element model; performing parameterization setting on key parameters in the model, and setting initial values and value ranges of the parameters; establishing a multi-objective optimization function and setting optimization constraint conditions; generating a DOE matrix of a plurality of groups of samples in a parameter value range, wherein each group of samples comprises a combination of all parameters; constructing a reduced order prediction model for predicting rib compression response under different parameter combinations; and with multi-objective function minimization as an objective, obtaining an optimal parameter combination meeting constraint conditions. According to the technical scheme, the calculation complexity can be reduced, the input boundary condition of the side column collision sliding table invasion equipment can be quickly determined, and the passenger injury condition under the whole vehicle column collision working condition can be highly reproduced.

The invention discloses a power plant safety intelligent early warning method based on a dynamic electronic fence, and the method comprises the steps: firstly analyzing the historical work ticket data of a power plant through a BERT-BiLSTM-CRF hybrid model, and recognizing an equipment entity, an operation type and a risk level; in combination with space coordinates of a power plant equipment three-dimensional model, an initial virtual safety boundary is automatically divided according to work ticket operation content, operation personnel operation actions, multi-person position relations and equipment states are fused in real time, and a buffer distance is dynamically adjusted to form a dynamic electronic fence; based on a dynamic electronic fence and high-precision positioning data, a group operation anti-collision model and a safe path planning strategy are established, a graded early warning mechanism is constructed, and corresponding grade early warning is triggered according to the real-time distance between an operator and the electronic fence through linkage of an intelligent wearable device and a power plant monitoring system. According to the invention, the intelligence and refinement level of power plant safety management can be significantly improved, and the occurrence rate of operation safety accidents is effectively reduced.

The invention discloses a method and system for monitoring a bubble contact angle in real time in a bubble impingement model, and belongs to the field of contact angle measurement. The problem that an existing contact angle monitoring method is not accurate is solved. According to the technical scheme, the method comprises the following steps: inversely fixing the surface of a to-be-tested sample in an experimental tank with a preset exposure area, and reserving a preset distance between the wall surface of the to-be-tested sample and a bubble generation channel at the bottom of the experimental tank so as to meet the space requirements of bubble generation and collision; the output end of the bubble generation channel outputs a single independent bubble, and a camera is started to record the collision dynamic process of the bubble and the wall surface of the to-be-detected sample; selecting a dynamic endpoint on the surface of the bubble in an image shot by a camera, calculating a dynamic angle in real time through a three-point method, and establishing a correlation model of the dynamic angle and a contact angle; the method is applied to the bubble collision model.

A method and a device for planning and controlling a leg robot, as well as a robot and a storage medium, wherein the method comprises: determining a foot-end reference state of each leg based on the motion state and swing parameter of the leg of a leg robot in the next instant; planning a collision-aware swing leg trajectory based on the foot-end dynamics model and the discrete collision model for the leg that will begin to swing in the next instant; performing whole-body control of the leg robot based on the motion state, the reference state sequence, and the foot-end reference state, wherein for the leg that was intended to be a support leg during planning and that did not touch the ground, collision-aware whole-body control is performed based on a whole-body dynamics model and a discrete collision model.This solves the problems of limited passive flexible buffering of collisions in leg robots in related technology, the large computational volume of the control algorithm, the low control accuracy, the high wear of the robot hardware, the short lifespan and the inability to meet the needs of practical applications.

This application relates to the field of automotive technology, and more particularly to a method, apparatus, device, and storage medium for adjusting a model. The method includes: obtaining a first collision simulation result based on a whole-vehicle collision model of a vehicle; the first collision simulation result includes at least the impact force of a first barrier and a protective beam in the vertical direction and a first acceleration of the non-deformed position of the protective beam in the vertical direction; determining the equivalent mass that generates the impact force based on the impact force and the first acceleration; controlling a second barrier included in the simplified collision model to impact the protective beam included in the simplified collision model with the impact force based on a simplified collision model of the vehicle and the equivalent mass, thereby obtaining a second collision simulation result; determining a first distance based on the first collision simulation result; determining a second distance based on the second collision simulation result; and adjusting the simplified collision model based on the difference between the first distance and the second distance.

The present invention concerns a system and a method for collision avoidance of a 3D robotic concrete printer, whereby the method comprises the steps of performing an application and / or manipulation process by executing instructions, such as based on G-Code, with a 3D robotic concrete printer by moving a tool in a path for applying and / or manipulating concrete material; and moving the tool in response to commands to a motion planner, said movement being assisted by a model-based collision prediction system, whereby the model-based collision-prediction system involves the steps of inferring and generating 3D geometries of concrete structures from applying and / or manipulating concrete structures by commanding the 3D robotic concrete printer, using the 3D geometries as a 3D collision model, which allow at any point in time to check if parts of the 3D robotic concrete printer would yield a collision when performing said movement; whereby the 3D collision model of the concrete structure is developed in parallel to the application and / or manipulation process and is updated in correspondence to the progress of the execution of the process instructions.

A method for solving the thermal-mechanical coupling multi-scale dynamics of flexible multi-body systems with clearance hinges, the technical field of space deployment mechanism dynamics modeling and numerical simulation. The existing solving method is difficult to accurately capture the change of the dynamics response of the space deployment mechanism under the thermal environment. According to the constructed temperature related hinge gap model, the temperature related material parameter model, the normal contact collision model, the partial slip tangential slip friction model, the tangential total friction force, the adhesion zone displacement and slip velocity model, the thermal viscous damping and tangential friction force model, the static friction growth model and the friction heat feedback transient heat conduction model, the thermal-mechanical coupling mechanical dynamics equation is constructed, which is decomposed into slow field equation and fast field equation, and the slow field equation and the fast field equation are solved through BDF2 and Newton-Raphson method and double parameter adaptive multi-scale implicit integration method, and the motion response of the hinge is obtained. The present application is used for obtaining the motion response of the hinge.

The application discloses a prefabricated wallboard adaptive segmentation method based on primary and secondary collision models and strong limb constraints, relates to the technical field of building information modeling (BIM) and prefabricated building (PC) deepening design, and comprises the following steps: global topology scanning and infinite extension line collision detection; node adaptive retreat based on a primary and secondary game model; hole feature extraction and strong limb shield safety checking; optimal equal division deduction based on a building module; physical connection blocking and gap instantiation reconstruction; and the like. The application can automatically identify the multi-wall body joint topology relationship in a BIM environment through a bottom algorithm, forcibly perform primary and secondary angle retreat, combine the door and window hole safety distance and the building module, and generate an independent prefabricated wallboard model with real physical joints and unified size and height, so as to provide an accurate geometric carrier for the automatic arrangement of subsequent steel plate and bolt connection nodes.

This disclosure relates to a trajectory planning method for a robotic arm, a mobile robot, a medium, and a product. The method includes: determining the initial pose of the end effector of the robotic arm in a working environment model and its target pose in a predetermined operation task; determining multiple workspaces based on the initial pose and the target pose; performing trajectory planning based on the working environment model, and during the trajectory planning process, acquiring multiple collision models with differences in geometric modeling accuracy and collision detection efficiency, and performing collision detection on trajectory points in the corresponding workspaces using the collision models; after the trajectory planning is completed, obtaining the target motion trajectory of the robotic arm under the predetermined operation task. This embodiment solves the problem of high computational complexity in global collision detection using traditional collision models, improves the trajectory planning efficiency of the robotic arm, and thus ensures the task response efficiency of the mobile robot in scenarios with dense equipment distribution.

This disclosure provides a motion optimization obstacle avoidance control method and system for vehicle-like robots, relating to the field of robot path planning and obstacle avoidance technology. The method includes: defining the state space of the vehicle-like robot; planning the global path of the vehicle-like robot; fitting the sparse state sequence using Hermite curves to obtain a time-space continuous trajectory; establishing a collision model between the vehicle-like robot and obstacles; transforming the geometric separation conditions between the robot and obstacles into a set of continuously differentiable nonlinear algebraic constraints based on duality theory; using the time-space continuous trajectory as a reference trajectory; and constructing a trajectory optimization objective function. This function can solve for the motion control quantities of the vehicle-like robot while satisfying its dynamic constraints, input constraints, and acceleration smoothing constraints, thus achieving obstacle avoidance planning and control of polygonal obstacles. This disclosure enables real-time, full-size robot planning and obstacle avoidance control.

The invention relates to a role avoidance control method and device, equipment and a medium. The method comprises the following steps: responding to an avoidance operation event triggered by a player role in a game scene, and obtaining a role avoidance path corresponding to the avoidance operation event; determining an avoidance space model correspondingly occupied by the player character on the character avoidance path in real time; and judging whether an attack collision model corresponding to an attack behavior in the current game scene collides with the avoidance space model or not, and when the collision is monitored, correspondingly triggering an avoidance interface special effect in the current game scene. By optimizing a role dodging control mechanism, the game running efficiency and judgment accuracy are improved, the operation experience and immersion of players are enhanced, and meanwhile, the compatibility and fluency of the game on multiple devices are ensured.

The application discloses a method, device, medium and equipment for simulating nuclear fusion based on a quantum circuit, and belongs to the field of quantum computing. The method comprises the following steps: constructing a plurality of energy models according to the energy types generated by nuclear fusion reactions; constructing a Hamiltonian H according to the plurality of energy models; calculating eigenvalues of the Hamiltonian H by using a quantum circuit under physical conditions set by the plurality of energy models; obtaining constraint conditions corresponding to electromagnetic energy terms, constraint conditions corresponding to the electromagnetic energy terms and physical conditions corresponding to kinetic energy terms under the eigenvalues of the Hamiltonian H; and inputting the constraint conditions corresponding to the electromagnetic energy terms, the physical conditions corresponding to the kinetic energy terms and the physical conditions corresponding to the kinetic energy terms into a collision model to simulate the nuclear fusion reactions, thereby improving the simulation accuracy and efficiency.

This invention relates to a simplified modeling method, system, storage medium, and electronic device for vehicle side impact collisions. The method includes: when a moving barrier collides with the side of a vehicle, acquiring a complete vehicle model and dividing the model into a collision-deformed portion and a non-collision-deformed portion; dividing the non-collision-deformed portion into a simplified vehicle portion and the tires on the non-collision side, and acquiring the center of mass position of the simplified vehicle portion; constructing a mass point at the center of mass position and connecting the mass point to the collision-deformed portion and the tires on the non-collision side, respectively, to obtain a simplified vehicle model; and generating a simplified vehicle side impact collision model based on the simplified vehicle model and the moving barrier. This invention simplifies the vehicle side impact collision model by simplifying the vehicle's structure and dividing the model into regions, thus improving modeling speed. The simplified model retains the deformation portion of the vehicle during the collision, reducing simulation time while accurately simulating the collision contact response process between the vehicle and the moving barrier.

The application discloses a method for analyzing the erosion characteristics of a steam turbine regulating valve based on a modified Oka erosion model, which comprises the following steps: S1, modifying the parameters of the Oka erosion model; S2, analyzing the flow field characteristics; and S3, analyzing the erosion characteristics. By constructing a single particle impact three-dimensional model and performing explicit dynamic finite element simulation, the particle size, impact velocity and impact angle parameters in the Oka erosion model are modified, so that the erosion prediction model is matched with the actual valve core material and high temperature and high pressure working conditions, and the accuracy of the erosion rate calculation is improved. By establishing a three-dimensional entity model of the regulating valve and extracting a fluid domain for flow field numerical simulation, the distribution of the velocity field and the pressure field in the valve is obtained, the modified Oka erosion model is introduced and coupled with the solid particle parameters, the particle motion equation and the particle-wall collision model are solved, and the distribution of the erosion and wear of the valve core surface is obtained, so that the positioning accuracy of the erosion area under the complex gas-solid two-phase flow is improved.

The application relates to the technical field of particle simulation, and discloses a particle collision method, device and equipment based on physical simulation and a storage medium, the method comprises the following steps: determining current parameters of each particle in a particle system based on initial parameters and external force parameters of each particle; detecting collision information between each particle and / or collision information between the particle and other objects in the particle system through a collision detection algorithm; and updating target collision parameters of each particle based on a collision model, the current parameters and the collision information. In the foregoing manner, the initial parameters and the external force parameters of the particles are considered to more accurately simulate the motion state of the particles under the action of force, a preset collision detection algorithm is used to detect the collision information between the particles and between the particles and other objects in the system, the simulation effect is more realistic, powerful technical support is provided for simulating complex physical phenomena, and the accuracy of the particle collision effect is improved.

The application provides a self-adaptive intelligent detection control system and method for automobile error-proof assembly, the system comprises information acquisition, point prediction, point correction, collision detection, path planning, speed planning, motion execution, motion judgment, information detection and deviation acquisition modules, and optionally comprises point re-prediction, path re-planning, speed re-planning and motion re-execution modules. The method obtains the position information of the vehicle workpiece to be detected through sensors and global cameras, establishes a multi-coordinate system conversion relationship and performs weight fusion, predicts and corrects the target point, plans the path and speed in combination with the collision model, controls the mechanical arm to complete the detection task in the meeting, tracking or reaching mode under the non-stop line state, and determines whether the workpiece is correctly assembled through the image collected by the depth camera. The application can realize full-automatic error-proof detection under the condition of mixed production of multiple vehicle models, does not require manual intervention, and has the advantages of high detection flexibility, high precision and high efficiency.

The invention belongs to the technical field of automatic driving, and particularly relates to an automatic driving automobile layered trajectory planning method and system for a dynamic city scene, and the method employs a layered architecture: an upper global planner generates an initial trajectory under a Frenet coordinate system, and actively screens a non-homotopy trajectory as an optimization starting point, so as to optimize the optimal trajectory; an optimal control problem is solved by combining a lane center line tracking cost function, and a global trajectory conforming to traffic rules is obtained; when a lower-layer local re-planner detects sudden change of an obstacle, safe starting and ending points are determined based on track segmentation diagnosis, a middle obstacle avoidance point is introduced to guide a detouring direction, and a local track is rapidly solved by using a double-disc simplified collision model; and finally, fusing the global track and the local track through space-time splicing to form a safe, smooth, dynamic and feasible final execution track. The system correspondingly comprises a global planning module, a local re-planning module and a trajectory fusion module. According to the method, the planning success rate and the driving safety and comfort in a complex dynamic scene are remarkably improved.

The application relates to the technical field of automobiles, in particular to a method and device for adjusting a model, equipment and a storage medium, which comprises the following steps: obtaining a first collision simulation result based on a whole vehicle collision model of a vehicle; the first collision simulation result at least comprises a collision force of a first barrier and a protective beam in a vertical direction and a first acceleration of a non-deformation position of the protective beam in the vertical direction; determining an equivalent mass generating the collision force based on the collision force and the first acceleration; controlling a second barrier included in a simplified collision model to collide with a protective beam included in the simplified collision model with the collision force based on the equivalent mass and the simplified collision model of the vehicle, so as to obtain a second collision simulation result; determining a first distance based on the first collision simulation result; determining a second distance based on the second collision simulation result; and adjusting the simplified collision model based on a difference between the first distance and the second distance.

The invention relates to a collision detection method and equipment. A collision model is generated according to geometric data of a robot CAD model; obtaining triangular mesh models of all obstacles in the working space, and combining to obtain an environment obstacle model and a spatial index structure thereof; shaft alignment bounding boxes of all the connecting rods are constructed according to the DH parameters and the joint limiting parameters of the robot, and the movement range of all the shaft alignment bounding boxes is obtained; sequentially judging whether the movement range of each axis alignment bounding box is intersected with the environmental obstacle model or not; if not, judging that there is no collision risk; if yes, surrounding balls corresponding to the axis alignment surrounding boxes are extracted, and whether the minimum distance between each surrounding ball and the environment obstacle model is smaller than a set collision early warning distance or not is judged; and if yes, determining that a collision risk exists. The problems that in the prior art, the collision detection efficiency cannot be remarkably improved on the premise that the detection accuracy is guaranteed, so that the overall motion planning process is accelerated, and the requirements of industrial application for real-time performance and reliability are met are solved.

The invention provides a robot space perception and motion planning method based on differential physics, and relates to the technical field of data processing, environment perception data and a robot motion state are fused, real-time vector grid cost with direction attributes is calculated for each grid, and direction association is carried out; establishing a continuous differentiable collision effect expansion function based on direction correlation, and representing collision intensity and spatial diffusion; attenuating and correcting the real-time vector grid cost to form a corrected vector cost field; and based on the corrected vector cost field, taking minimization of the path integral total cost as a target, carrying out gradient optimization by utilizing a full-link differentiable characteristic, and solving an optimal motion track in combination with vector direction information. According to the method, deep fusion of perception and motion states, continuous micronization of a collision model and integration of perception and planning are realized, and the motion planning efficiency and safety of the robot in a dynamic complex environment can be improved.

The application discloses a multi-axis collaborative robot dynamic path planning method based on deep reinforcement learning, constructs a kinematics model and an obstacle collision model of the robot, designs a deep reinforcement learning model of the robot, introduces irrelevant dynamic zero-mean Gaussian noise into the deep reinforcement learning model to improve the explorativeness of action selection, dynamically controls the size of the Gaussian noise variance, maintains the balance of the algorithm exploration process and the development process, then introduces an importance sample playback mechanism, takes the sample data error size as a standard for measuring the importance of the sample, performs importance sorting, increases the probability of extracting and learning the data sample ranking in the front, improves the convergence speed of the learning process, further constructs a state space, an action space and a reward function of the robot intelligent agent interaction process which fully considers the dynamic obstacle avoidance planning characteristics of the robot, promotes the robot intelligent agent to learn better, and finally the intelligent agent continuously iterates and updates strategies in a simulation environment to obtain an intelligent dynamic obstacle avoidance planning strategy.

The invention discloses a multi-source data fused tornado-caused high-speed rail foreign matter risk early warning method. The method comprises the following steps: firstly, acquiring and preprocessing multi-source data, further performing macro, medium and micro multi-scale vortex decomposition based on the horizontal diameter of a tornado vortex, and adaptively weighting and extracting a fusion feature vector by adopting a gating fusion mechanism; then five risk factors including public early warning, tornado probability, track invasion and overhead line system hanging are calculated, and a fractional order collision model is introduced to accurately describe the nonlinear rebound process after the foreign matter collides with the overhead line system; and finally, performing optimization solution on the constructed Hamiltonian containing the fitting item and the space-time smoothing item through a quantum annealing algorithm, and outputting a global early warning level which is smooth and optimal in time and space. According to the method, the problems of low precision and high false alarm rate of traditional early warning are solved, and the accuracy and reliability of high-speed rail operation safety early warning are remarkably improved.

The invention belongs to the technical field of physics and infrared radiation simulation, and particularly relates to a vacuum plume infrared radiation rapid calculation method which comprises the following steps: S1, generating a discrete dot matrix covering a computational domain according to input computational domain geometric parameters; s2, performing weighted average on the initial parameters of the components at the nozzle to obtain equivalent mixed gas parameters at the nozzle; s3, solving based on an incoming flow-plume collision model to obtain an equivalent particle source position under the influence of incoming flow; s4, solving by utilizing a vacuum plume engineering calculation model and combining a numerical integration algorithm to obtain micro flow field parameter distribution of each component gas on the discrete dot matrix; s5, vacuum plume overall flow field distribution of the whole computational domain is obtained; s6, solving the radiation transfer equation by adopting an apparent ray method to obtain spectral radiation intensity; the problems that an existing vacuum plume flow field calculation process is complex, consumed time is long, and the whole process from plume flow field generation to infrared radiation transmission cannot be covered are solved.

The invention provides a collision spray morphological feature prediction method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining collision spray simulation data obtained based on CFD simulation software simulation; constructing a correction function based on key parameters of a spray collision model, and introducing the influence of environmental factors on the spray form through the correction function so as to correct the original collision spray simulation data; wherein the key parameters at least comprise a collision Weber number, a collision eccentricity and a droplet radius ratio; and inputting the preprocessed correction function and the collision spray simulation data into a pre-constructed deep learning model so as to output and obtain corrected collision spray morphological feature prediction data. According to the method, the collision spray morphological characteristics obtained through simulation are corrected through the deep learning model, and compared with a traditional method only adopting CFD simulation software for calculation, the prediction precision is higher, and the environmental adaptability is higher.

The application discloses an aircraft cluster security situation awareness method in an autonomous operation mode, is oriented to a high-altitude civil aircraft operation scene and belongs to the field of aviation operation safety. First, in order to meet the requirement of risk management refinement among aircrafts in the autonomous operation mode, communication navigation surveillance factors influencing flight path uncertainty are comprehensively considered, errors caused by CNS performance are introduced into a traditional collision model, traditional aircraft edge connection judgment basis is optimized, and an aircraft cluster situation network is constructed. Second, based on the constructed situation network, combined with aircraft cluster space structure representation and traffic flow parameter information, a cluster risk evaluation index is designed, cluster risk modes are further divided, and an air traffic risk situation network is formed. Finally, based on the cluster structure in the risk situation network, the time sequence distribution, survival characteristics and space operation preference of aircraft clusters in each risk mode are discussed, and the space-time evolution characteristics of the risk situation network are mined. The application can realize refined determination of conflict risks among aircrafts and mine space-time evolution characteristics of aircraft clusters, thereby providing theoretical support for the autonomous operation mode.

This invention belongs to the field of ship automation control and provides a method, device, equipment, and storage medium for automatic collision avoidance of ships. The method includes: constructing a collision avoidance model based on deep reinforcement learning training to formulate avoidance decisions; when a grid sensor detects other ships, acquiring the state information of the other ships and calculating the collision risk area based on the state information; when the collision risk area overlaps with the sensing area of ​​the grid sensor, determining whether the ship is a avoidant or a hold-off ship; if it is a hold-off ship, maintaining its course and speed; if it is an avoidant ship, calculating an avoidance decision; after executing the avoidance decision, updating the state information of the ship and other ships in the environmental information, repeating the process until the distance between the ship and other ships is greater than a threshold or less than a threshold and there is no collision risk; this invention aims to solve the problems of traditional ship collision avoidance technology being unsuitable for complex sea conditions and having high computational complexity.