A dynamic navigation method and system based on multi-factor fusion and human-computer collaboration

Abstract

The application discloses a kind of dynamic navigation method and system based on multi-factor fusion and man-machine cooperation, it is related to vehicle navigation and artificial intelligence application technical field.The method includes real-time acquisition and fusion multi-source data.Fused multi-source data is input into reinforcement learning model, and the integrated generation value corresponding to one or more alternative paths and each alternative path at the current time is output.The dynamic cost function of reinforcement learning model at least includes time cost term, energy consumption cost term, comfort cost term and health risk cost term.For each alternative path, the confidence of the alternative path and the expected time saving relative to the current path are calculated respectively, and based on the preset three-level man-machine cooperative decision rule, the target path is selected from the alternative path and the corresponding man-machine cooperative response level is determined.The navigation operation corresponding to the selected target path and man-machine cooperative response level is executed.

Description

Technical Field

[0001] This invention relates to the fields of vehicle navigation and artificial intelligence application technology, and in particular to a dynamic navigation method and system based on multi-factor fusion and human-machine collaboration. Background Technology

[0002] As cities become increasingly dynamic and user needs become more diversified, traditional in-car navigation systems can no longer meet people's comprehensive expectations for travel efficiency, comfort, and safety.

[0003] Currently, mainstream in-vehicle navigation systems still heavily rely on static or periodically updated traffic flow data and historical road network models for their decision-making. This calculation model, based on fixed rules and limited data dimensions, often exhibits significant response lag and poor adaptability when faced with real-time interference such as sudden accidents, temporary traffic control, and severe weather. The optimal route planned may become ineffective in actual travel due to information distortion or environmental changes, not only failing to save travel time but also potentially exacerbating traffic congestion, increasing energy consumption, and affecting the consistency and predictability of the travel experience.

[0004] Furthermore, in terms of interaction, most existing technical solutions simply switch between two modes: fully automatic frequent replanning and passive one-way information prompts. The fully automatic frequent replanning mode is prone to distracting the driver due to low decision-making transparency and unstable reliability, damaging the user's trust in the system. The passive one-way information prompt mode, on the other hand, is too conservative and fails to provide timely and effective action suggestions when there are clear optimization benefits, causing the user to miss the optimal opportunity to switch routes. Summary of the Invention

[0005] To address issues such as limited data foundation, singular optimization objectives, and rigid human-computer interaction mechanisms, this invention proposes a dynamic navigation method and system based on multi-factor fusion and human-computer collaboration. Through multi-source data fusion, reinforcement learning, dynamic cost functions, and three-level human-computer collaborative decision-making rules, the method achieves a balance between the dynamism, adaptability, security, and user experience of navigation services, thereby improving the quality of navigation services.

[0006] The present invention achieves the above objectives through the following technical solutions:

[0007] A dynamic navigation method based on multi-factor fusion and human-machine collaboration includes:

[0008] Real-time acquisition and fusion of multi-source data;

[0009] The fused multi-source data is input into the reinforcement learning model, which outputs one or more alternative paths at the current moment and the comprehensive cost value corresponding to each alternative path; the dynamic cost function of the reinforcement learning model includes at least time cost, energy cost, comfort cost and health risk cost.

[0010] For each alternative path, calculate the confidence level of the alternative path and the expected time saving relative to the current path, and select the target path from the alternative paths and determine the corresponding human-machine collaborative response level based on the preset three-level human-machine collaborative decision-making rules.

[0011] Perform navigation operations corresponding to the selected target path and human-machine interaction response level.

[0012] As a preferred embodiment of the present invention, the multi-source data includes real-time traffic data, environmental data, and personalized health data of users.

[0013] As a preferred embodiment of the present invention, the calculation basis of the time cost item includes the intersection green light passage probability obtained based on the passage probability model; the passage probability model is a machine learning model, and the input includes at least the distance of the vehicle from the intersection, the current vehicle speed, the remaining time of the traffic light, and the historical data of the average green light interval of the intersection phase;

[0014] The health risk cost item is calculated based on environmental data and personalized health data.

[0015] As a preferred embodiment of the present invention, the reinforcement learning model introduces an attention mechanism and an improved experience replay mechanism;

[0016] The weight parameters of each cost item in the dynamic cost function can be dynamically adjusted; the dynamic adjustment is achieved through a combination of one or more of the following methods: weight adaptation based on the driving mode selected by the user, meta-learning optimization based on user feedback, and weight template switching based on real-time scene recognition.

[0017] As a preferred embodiment of the present invention, the preset three-level human-machine collaborative decision-making rules include:

[0018] The first-level human-machine collaborative decision-making rule is used to select an automatic execution level response when the confidence of any alternative path is higher than the first confidence threshold and the expected time saving is greater than the first time threshold.

[0019] The two-level human-machine collaborative decision-making rule is used to select and confirm the execution level response when the confidence level of any alternative path is between the first confidence threshold and the second confidence threshold, and the expected time saving is between the first time threshold and the second time threshold.

[0020] The three-level human-machine collaborative decision-making rule is used to select a suggestion-level response when the confidence level of any alternative path is lower than the second confidence threshold, or when the environment of the alternative path is determined to be a complex road condition.

[0021] As a preferred embodiment of the present invention, the confidence level is calculated based on a weighted average of data reliability, driving habit adaptability, and road condition complexity adaptability.

[0022] The expected time savings The calculation method is as follows: In the formula, The estimated time to travel the current route; The estimated time to travel the alternative route.

[0023] As a preferred embodiment of the present invention, selecting the target path from the alternative paths includes:

[0024] Based on confidence level, expected time savings, and preset three-level human-machine collaboration decision-making rules, the human-machine collaboration response level corresponding to each alternative path is determined.

[0025] Based on the priority of human-machine collaboration response levels, the alternative path with the highest priority is determined as the target path; among them, the automatic execution level response has the highest priority, the confirmation execution level response has the second highest priority, and the suggestion prompt level response has the lowest priority.

[0026] If two or more alternative paths have the same highest priority, the alternative path with the lowest overall cost will be selected as the target path.

[0027] As a preferred embodiment of the present invention, the navigation operation corresponding to the selected target path and the human-machine collaborative response level includes:

[0028] When the human-machine collaboration response level corresponding to the target path is automatic execution level, the navigation path will be automatically switched to the target path, and a status notification will be sent to the user.

[0029] When the human-machine collaboration response level corresponding to the target path is the confirmation execution level response, a voice prompt is issued to the user, and after obtaining the user's confirmation, the navigation path is switched to the target path;

[0030] When the human-machine interaction response level corresponding to the target path is the suggestion prompt level, the target path is visually prompted through the augmented reality head-up display, and the user's manual selection command is received. Based on the command, it is determined whether to switch the navigation path to the target path.

[0031] As a preferred embodiment of the present invention, the judgment parameters for complex road conditions include at least one of the following: frequency of lane number change, average speed difference between adjacent road segments, traffic event density, and traffic light intersection density.

[0032] The visual cues provided via augmented reality head-up display include lane-level blue light strips overlaid on the real road, dynamic green arrows at turns, and amber warning zones displayed on high-risk road sections.

[0033] A dynamic navigation system based on multi-factor fusion and human-machine collaboration includes:

[0034] The data fusion module is used to acquire and fuse multi-source data in real time;

[0035] The path planning module is used to input the fused multi-source data into the reinforcement learning model and output one or more alternative paths at the current time and the comprehensive cost value corresponding to each alternative path; the dynamic cost function of the reinforcement learning model includes at least time cost, energy cost, comfort cost and health risk cost.

[0036] The human-machine collaborative decision-making module is used to calculate the confidence level and expected time saving relative to the current path for each alternative path, and select the target path from the alternative paths and determine the corresponding human-machine collaborative response level based on the preset three-level human-machine collaborative decision-making rules.

[0037] The navigation execution module is used to perform navigation operations corresponding to the selected target path and the human-machine collaborative response level.

[0038] The beneficial effects of this invention are as follows: By acquiring and fusing multi-source data in real time, a comprehensive and accurate environmental perception capability is constructed, fundamentally overcoming the limitations of traditional navigation data sources being singular and lagging in updates. This provides real and multi-dimensional data support for path planning, ensuring that the planning starting point closely matches the real-time scenario, effectively avoiding initial path deviations caused by information fragmentation, and laying a reliable foundation for subsequent decisions. Based on this, a dynamic cost function incorporating time, energy consumption, comfort, and health risk costs is constructed and optimized by a reinforcement learning model, breaking through the limitations of traditional navigation that only focuses on the shortest time or distance as a single objective. This method can sustainably weigh multiple, even conflicting, objectives during travel, thereby dynamically planning the path with the best long-term comprehensive benefits. It not only simultaneously meets users' diverse expectations for efficiency, economy, experience, and safety, but also adapts to diverse user needs such as health-sensitive and energy-saving users, expanding application scenarios. Finally, by introducing a three-level human-machine collaborative decision-making rule based on confidence level and expected time savings, the differentiation and intelligence of navigation response are realized. This hierarchical response mechanism reduces interference from invalid information, lowers the risks that may be brought about by fully automated decision-making, improves travel efficiency, ensures safety and user trust, and builds an efficient and reliable human-machine collaborative relationship. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0040] Figure 1 A flowchart of a dynamic navigation method based on multi-factor fusion and human-machine collaboration provided by the present invention; Figure 2 This invention provides a modular structure diagram of a dynamic navigation system based on multi-factor fusion and human-machine collaboration. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0042] Traditional in-vehicle navigation systems suffer from significant limitations in both data foundation and interaction modes. At the data level, they largely rely on outdated static or periodic data, making it difficult to effectively respond to real-time dynamic changes such as sudden road conditions and temporary traffic restrictions. This often results in planned routes no longer being the optimal choice in actual driving, and may even increase travel time and energy consumption. At the interaction level, they often fall into a dilemma of being overly proactive or overly conservative. Frequent automatic replanning can easily disrupt driver focus due to insufficient decision-making transparency, thereby eroding user trust; while simple one-way information prompts are too passive, failing to provide clearly beneficial route switching suggestions at the optimal time, ultimately missing crucial opportunities for route optimization. To address these issues, this invention proposes a dynamic navigation method and system based on multi-factor fusion and human-machine collaboration.

[0043] like Figure 1 As shown, this is an embodiment of the present invention, which provides a dynamic navigation method based on multi-factor fusion and human-machine collaboration, including:

[0044] S1 acquires and integrates multi-source data in real time. This multi-source data includes real-time traffic data, environmental data, and personalized user health data.

[0045] Furthermore, real-time traffic data is collected using a low-latency data link built with V2X communication technology.

[0046] Specifically, high-frequency traffic flow data is acquired in real time from roadside units deployed at key intersections and road sections via V2I communication. This traffic flow data includes real-time average vehicle speed, traffic density, queue length, and current traffic light phase and precise remaining time from the traffic signal control system for the current and adjacent lanes. For example, the high-frequency is set to once per second. Simultaneously, access to the traffic information service center's cloud platform via 4G or 5G networks provides access to a wider range of macro-level road condition distribution and officially released traffic event information. Furthermore, real-time traffic data is highly time-sensitive; the data update cycle is set to no more than 2 seconds, and data that has not been updated within the specified time is automatically marked with a decreased confidence level to avoid invalid data input.

[0047] Environmental data is accessed through standardized application programming interfaces (APIs) to obtain basic data from authoritative external data sources such as environmental protection departments and meteorological bureaus. This data is then combined with vehicle location data to achieve path-level matching, providing support for health risk cost calculation.

[0048] Specifically, by accessing the air quality monitoring application interface of the environmental protection department, real-time pollutant concentration data from various monitoring stations, such as PM2.5, PM10, and nitrogen dioxide concentrations, are obtained. Combined with vehicle GPS locations, an air quality index profile along each alternative route is generated using a spatial interpolation algorithm. Simultaneously, by accessing the weather forecast application interface of the meteorological bureau, current and short-term precipitation type and probability, temperature, wind speed and direction, visibility, and other data are obtained and directly matched to each alternative route. For example, the short-term future is set to 30 minutes.

[0049] Personalized health data originates from simplified electronic medical records that are pre-authorized and locally encrypted, or from personal health cards created by the user within the in-vehicle health application. This data includes user-defined chronic disease history, pollutant allergy history, and other health information directly related to environmental exposure risks. Data access requires one-time user authorization and is processed locally on the device or through end-to-end encrypted transmission to ensure privacy and security.

[0050] Data fusion transforms the three types of multi-source data mentioned above into feature vectors in a unified format, providing standardized input for reinforcement learning models. The fusion steps include spatiotemporal alignment, feature association, and vector construction, as detailed below:

[0051] First, real-time traffic data, environmental data, and personalized user health data are mapped to the same baseline. Spatially, this is linked to a high-precision map layer; temporally, it is synchronized to a unified timestamp to eliminate the spatiotemporal heterogeneity of the data.

[0052] Next, based on the vehicle's current GPS location, each alternative route is associated with its traffic, environmental, and geographical features, and these associated features are integrated with the user's personalized health information to form a set of route-level features.

[0053] Finally, key features are extracted and vectorized from the feature set of each path to form a multi-dimensional feature vector containing traffic status, environmental quality, geographic information, and user health context.

[0054] S2 inputs the fused multi-source data into the reinforcement learning model and outputs one or more alternative paths at the current time and the comprehensive cost of each alternative path. The dynamic cost function of the reinforcement learning model includes at least time cost, energy cost, comfort cost, and health risk cost.

[0055] Furthermore, the reinforcement learning model recommends using a proximal policy optimization algorithm to optimize the navigation strategy through continuous interaction with the environment, thereby achieving a dynamic balance of multi-objective costs.

[0056] The reinforcement learning model employs a multi-stage training model combining offline pre-training and online fine-tuning. The offline dataset utilizes a large-scale historical driving trajectory dataset with multi-dimensional labels including time, energy consumption, comfort, and health events. The state vector is composed of the fused multi-source feature vector at the current moment and the vehicle state. The action space is designed as a discrete set of movement directions. The model hyperparameters are set as follows: learning rate 0.001, discount factor 0.95, and initial exploration rate ε of 0.1.

[0057] To achieve a dynamic balance of multi-objective costs, this model employs an exponential reward function based on the dynamic cost function A, i.e., R = -exp(λ * A), where R is the immediate reward of the current step, A is the dynamic total cost calculated by fusing multiple objectives such as time, energy consumption, comfort, and health events, and λ is a scaling factor greater than zero. This design strengthens the penalty for high-cost paths through exponential mapping, thereby improving the model's decision sensitivity. The scaling factor λ is a key parameter controlling the reward scale, and its initial value is determined based on the cost distribution of typical trajectories in offline data. The calculation formula is: In the formula, is the median cost of a typical trajectory in the offline dataset, representing the cost level of moderate driving behavior. The target reward value is usually set between -2 and -5 based on engineering experience, so that the reward value is within a reasonable range suitable for gradient optimization.

[0058] The reinforcement learning model incorporates an attention mechanism and improves the experience replay mechanism.

[0059] Furthermore, a multi-head self-attention mechanism is introduced into the value network of the proximal policy optimization algorithm. This mechanism allows the model to dynamically capture long-range dependencies and global correlations between different features in the state vector. When evaluating a path, the model can simultaneously consider global information such as the transmission effect of upstream intersection congestion on downstream road segments and the spatial distribution pattern of environmental risks in the road network, thereby making a more globally optimal decision and improving the model's policy generalization ability in complex road networks.

[0060] To address the problem of traditional priority-based experience replay often over-focusing on experiences with high temporal difference errors, this invention implements hierarchical storage and balanced replay in the experience replay buffer of the training loop. Specifically, historical transfer experiences are stored hierarchically according to decision scenario type and decision result. During training, experience batches are extracted from each scenario layer according to a preset ratio, and sampling is performed within each layer based on priority. This ensures both the efficiency of learning high-value, high-error key experiences and the balance of training data across diverse scenarios, thereby improving the robustness of the strategy and user acceptance.

[0061] The dynamic cost function of a reinforcement learning model includes at least time cost, energy cost, comfort cost, and health risk cost.

[0062] Furthermore, the reinforcement learning model's built-in dynamic cost function quantifies the long-term cost of the path from four dimensions and achieves multi-objective optimization through adaptive weights. Specifically:

[0063] The time cost component incorporates a lightweight intersection passage probability model based on XGBoost, deployed on the vehicle for real-time inference. This model takes as input historical data the vehicle's distance from the intersection, current speed, remaining traffic light time, and the average green light interval for that intersection phase, and outputs the probability of the vehicle passing through on a green light.

[0064] For the path in the first At a traffic light intersection, the expected travel time is... The calculation formula is:

[0065]

[0066] in, For the first At each traffic light intersection, the free passage time is specified. This represents the average red light duration. This is an empirical coefficient. For the vehicle to arrive at the The probability of passing through a traffic light intersection without stopping. In a preferred embodiment, an empirical coefficient... Setting it to 0.5 indicates the average waiting time for half a red light.

[0067] The total time cost of the route is the sum of the travel time of all road segments in the route and the expected passage time at intersections, and is normalized using the max-min normalization method.

[0068] The energy consumption cost is accurately estimated based on the vehicle's longitudinal dynamics model, using a maximum-minimum normalization method for normalization. The input data includes path gradient, radius of curvature, predicted speed curve, and real-time vehicle parameters.

[0069] The comfort cost is calculated based on the longitudinal acceleration derived from the lateral acceleration and the rate of change of the velocity curve, derived from the path curvature. It is obtained through weighted calculation and normalized using the maximum-minimum normalization method.

[0070] The health risk cost is calculated based on environmental and personalized health data. This cost enables the navigation system to provide differentiated health route planning for users with varying health conditions. The formula for calculating the health risk cost is:

[0071]

[0072] in, For each segment along the route The health risk costs. For each segment along the route The environmental exposure dose is calculated using real-time air quality index, concentrations of major pollutants, pollen concentration, and meteorological data, combined with the user's expected breathing rate and exposure time in that road section. For example, major pollutant concentrations include PM2.5 and NO2, and meteorological data includes temperature and humidity. To obtain an individual's health risk coefficient, a risk coefficient matrix generated based on the user's encrypted health record is obtained.

[0073] The total health risk cost is the sum of the health risk costs of all road segments, and is normalized using the max-min normalization method.

[0074] The weight parameters of each cost item in the dynamic cost function can be dynamically adjusted. This dynamic adjustment is achieved through a combination of one or more of the following methods: weight adaptation based on the user-selected driving mode, meta-learning optimization based on user feedback, and weight template switching based on real-time scene recognition.

[0075] Furthermore, to ensure that the navigation strategy accurately adapts to the user's immediate needs and changing scenarios, the weights of each cost item in the dynamic cost function can be adjusted collaboratively through one or more of the following methods to achieve precise strategy adaptation. Specifically:

[0076] Weighted adaptation is performed based on the user's selected driving mode. For each selected driving mode, a pre-defined weighted template is invoked for adaptation. This method is responsive, directly meets the user's specific needs, and requires no learning curve.

[0077] The pre-defined weighted templates are constructed based on large-scale user surveys and simulation optimizations. Each template corresponds to a core need and clarifies the weight priority of each cost item. For example, taking the extreme efficiency template, the weight of time cost is set to 0.6, the weight of energy cost is set to 0.15, the weight of comfort cost is set to 0.15, and the weight of health risk cost is set to 0.1.

[0078] Meta-learning optimization based on user feedback analyzes user behavior along the recommendation path using a meta-learner to optimize weight parameters in real time. This approach requires no manual user intervention and adaptively captures changes in user preferences over time.

[0079] The weight template switching based on real-time scene recognition uses a lightweight scene classifier to identify the current navigation scene in real time and automatically match a simulation-optimized weight template. This approach offers rapid response and adaptability to sudden scene changes.

[0080] The scene classifier is a lightweight neural network model deployed on the vehicle. Input features include time features, environmental features, road network features, and historical features. Road network features include the current road type and traffic flow density. Historical features include the user's historical navigation destinations during that time period.

[0081] After outputting the scene labels, the corresponding weight template will be automatically invoked. The template parameters are derived from large-scale simulation tests and user behavior statistics in this scene.

[0082] Taking the peak commuting scenario as an example, the weight of time cost is set to 0.5, the weight of comfort cost is set to 0.1, the weight of energy cost is set to 0.25, and the weight of health risk cost is set to 0.15, in order to meet the core need of being in a hurry.

[0083] After training, the reinforcement learning model executes an online decision-making and path generation process. Specifically:

[0084] The current fused state vector is input into the model. The policy network of the model outputs the probability distribution of each alternative turning action, and the value network outputs the baseline value of the current state. Furthermore, Monte Carlo tree search is used to perform forward inference of finite depth, thereby generating one or more alternative paths to the destination at the current time. The dynamic cost function is used to accurately calculate the comprehensive cost of each alternative path from the current position to the destination. Finally, one or more alternative paths, the comprehensive cost of each alternative path, and the path feature uncertainty evaluated internally by the model are output.

[0085] By employing reinforcement learning models and multi-factor dynamic cost functions, alternative route plans can be generated in real time. These plans not only aim for the shortest travel time but also finely balance energy efficiency, driving comfort, and personalized health risks. Furthermore, by combining model improvement mechanisms and dynamic weight adjustments, highly personalized, multi-objective optimized intelligent route planning capabilities are achieved in changing environments, providing a high-quality selection basis for human-machine collaborative decision-making.

[0086] S3. For each candidate path, calculate the confidence level of the candidate path and the expected time saving relative to the current path, and select the target path from the candidate paths and determine the corresponding human-machine collaborative response level based on the preset three-level human-machine collaborative decision-making rules.

[0087] The confidence level is calculated by weighting data reliability, driving habit fit, and road condition complexity fit.

[0088] Furthermore, the formula for calculating the confidence level C is:

[0089]

[0090] in, This refers to the completeness of real-time traffic data used to predict road conditions along this route, representing data reliability. The calculation formula is as follows: , The missing rate of core fields in real-time traffic data is calculated using the following formula: , The number of fields not retrieved. These are pre-defined core fields that are crucial for path time prediction. For example, core fields include average vehicle speed per road segment, congestion status, and event information.

[0091] To ensure compatibility with driving habits, the system determines routes based on the user's historical navigation choices. For example, if a user selects expressways more than 80% of the time, then expressway-type alternative routes will be provided. The value ranges from 0.85 to 1.0 for urban branch roads. It ranges from 0.4 to 0.6.

[0092] The road condition complexity adaptability is determined based on the degree of matching between the route and the user's preset road condition preferences. For example, if the user's preset preference is simple road conditions, then alternative routes without frequent lane changes or traffic events will be selected. The value ranges from 0.85 to 1.0 for complex road conditions. It ranges from 0.3 to 0.6.

[0093] These are weighting coefficients, which are personalized by the user based on their emphasis on data, habits, or road conditions, or dynamically adjusted according to driving modes, and must meet certain requirements. The default configuration is... =0.4, =0.3, =0.3.

[0094] Expected time savings The calculation method is as follows: .in, This represents the estimated time to travel the current route. The estimated time to travel the alternative route.

[0095] The pre-defined three-level human-machine collaborative decision-making rules include:

[0096] The first-level human-machine collaborative decision-making rule is used to select an automatic execution level response when the confidence level of any alternative path is higher than the first confidence threshold and the expected time saving is greater than the first time threshold.

[0097] Furthermore, the first confidence threshold is obtained through posterior probability analysis of historical successful path planning data to ensure that the triggering of automatic execution is extremely cautious, in order to establish and maintain user trust in the system's automatic capabilities. A recommended setting range is between 0.8 and 0.9. The first time threshold can be adjusted proportionally according to the total trip duration to ensure that automatic execution has clear user utility; a recommended value is 15% of the total trip duration.

[0098] The two-level human-machine collaborative decision-making rule is used to select and confirm the execution level response when the confidence level of any alternative path is between the first confidence threshold and the second confidence threshold, and the expected time saving is between the first time threshold and the second time threshold.

[0099] Furthermore, the second confidence threshold is related to the user's familiarity with and tolerance of the system, and can be configured by the user, with a recommended setting range of 0.5 to 0.7. The second time threshold is related to the complexity of the specific human-computer interaction interface design, with a recommended setting range of 2 to 5 minutes.

[0100] The three-level human-machine collaborative decision-making rule is used to select a suggestion-level response when the confidence level of any alternative path is lower than the second confidence threshold, or when the environment of the alternative path is determined to be complex road conditions. The judgment parameters for complex road conditions include at least one of the following: lane number change frequency, average speed difference between adjacent road segments, traffic event density, and traffic light intersection density.

[0101] Furthermore, complex road conditions are determined by a combination of one or more judgment parameters exceeding their corresponding thresholds. The criteria for setting each judgment parameter and its threshold are as follows:

[0102] When the frequency of lane number changes exceeds a preset threshold, the path is considered to have complex road conditions in this dimension. The preset threshold is determined by analyzing high-precision map data and historical vehicle trajectory data.

[0103] When the average speed difference between adjacent road segments exceeds a preset speed threshold, the route is determined to be in a complex road condition in this dimension. The preset speed threshold is set based on real-time traffic flow theory and accident statistics.

[0104] When the traffic incident density exceeds a preset incident occurrence threshold, the route is determined to be in a complex road condition in this dimension. The preset incident occurrence threshold is derived from real-time event pushes from the traffic management platform.

[0105] When the density of traffic light intersections exceeds a preset threshold, the route is considered to have a complex road condition in this dimension. The preset threshold is derived from the analysis of the road network structure in different areas of the city and is a key indicator for distinguishing between roads with high traffic efficiency and roads with high traffic control requirements.

[0106] In a preferred embodiment, the preset change threshold is set to 2 times per kilometer, the preset vehicle speed threshold is set to 20 kilometers per hour, the preset event occurrence threshold is set to 2 events per kilometer, and the preset traffic light intersection threshold is set to 3 intersections per kilometer.

[0107] The aforementioned confidence threshold and time threshold have dynamic adjustment capabilities to adapt to different users and scenarios:

[0108] Based on the user's response to historical recommendations, the system learns online to fine-tune the user's personal preference thresholds, making the collaborative strategy increasingly aligned with the user's habits.

[0109] If multiple paths simultaneously meet the same rule, the path with the highest time saving is selected first. If a path triggers both complex road conditions and high confidence, the response level is downgraded to Level 3 based on the complex road condition priority principle to ensure safety and user experience.

[0110] Selecting a target path from the alternative paths includes:

[0111] Based on confidence level, expected time savings, and preset three-level human-machine collaborative decision-making rules, the human-machine collaborative response level corresponding to each alternative path is determined.

[0112] Furthermore, for each alternative path, based on confidence level, expected time savings, and preset three-level human-machine collaboration decision-making rules, and in conjunction with preset thresholds, the corresponding human-machine collaboration response level is determined. The human-machine collaboration response levels include automatic execution level response, confirmed execution level response, or suggestion / prompt level response.

[0113] Based on the priority of human-machine collaboration response levels, the alternative path with the highest priority is determined as the target path. Among them, the automatic execution level response has the highest priority, followed by the confirmation execution level response, and the suggestion prompt level response has the lowest priority.

[0114] Furthermore, the path with the highest priority response level among all candidate paths is included in the final candidate set. If the final candidate set contains only one path, that path is directly designated as the target path, and the human-machine collaboration response level corresponding to the target path is that priority response level.

[0115] If two or more alternative paths have the same highest priority, the alternative path with the lowest overall cost will be selected as the target path.

[0116] Furthermore, if the final candidate set contains two or more paths, the path with the lowest overall cost will be selected as the final target path. The human-machine collaborative response level corresponding to this target path will remain the aforementioned highest priority level.

[0117] For example, suppose there are two alternative paths, X and Y. The confidence level and expected time savings of path X are calculated. Trigger confirmation execution level, confidence level of path Y, and expected time savings. Automatic execution level is triggered. Based on priority rules, path Y is determined as the target path, and an automatic execution level response is adopted. If both paths X and Y trigger automatic execution level, their overall cost values ​​are compared, and the path with the lower overall cost value is determined as the target path.

[0118] S4 executes navigation operations corresponding to the selected target path and human-machine interaction response level.

[0119] When the human-machine collaboration response level corresponding to the target path is automatic execution level, the navigation path will be automatically switched to the target path, and a status notification will be sent to the user.

[0120] Furthermore, this step allows the navigation interface to smoothly transition to the new route with an animation, ensuring visual continuity. Simultaneously, the system will provide a brief voice prompt or display a concise graphic notification in a fixed area of ​​the screen to inform the user that the change has been completed.

[0121] During this process, the user does not need to perform any operation. If the user wishes to undo the operation, they can immediately restore the original path using a shortcut. For example, the shortcut can be executed by saying "Cancel Change" or pressing the "Cancel" button on the steering wheel.

[0122] When the human-machine collaboration response level corresponding to the target path is the confirmation execution level response, a voice prompt is issued to the user, and after obtaining the user's confirmation, the navigation path is switched to the target path.

[0123] Furthermore, a structured suggestion message is read to the user via voice, containing information about the advantages of the target route. For example, if there is congestion ahead, it is suggested that you switch to the Yellow River section, which is expected to save 8 minutes and pass through approximately 3 traffic lights. Would you like to switch?

[0124] Simultaneously, a confirmation window will appear on the central control screen or instrument panel, providing a "Yes / No" option or a countdown timer. Users must confirm within a limited time by responding with voice, pressing the confirmation button on the steering wheel, or tapping the touchscreen. The route switch will only be executed after a confirmation command is clearly recognized.

[0125] When the human-machine interaction response level corresponding to the target path is the suggestion prompt level, the target path is visually prompted through the augmented reality head-up display, and the user's manual selection command is received. Based on the command, it is determined whether to switch the navigation path to the target path.

[0126] Visual cues provided through augmented reality head-up displays include lane-level blue light strips overlaid on real roads, dynamic green arrows at turns, and amber warning zones displayed on high-risk road sections.

[0127] Furthermore, through an augmented reality heads-up display, the target path is overlaid onto the real road as a lane-level blue light strip, with dynamic green arrows displayed at turns, and high-risk areas such as construction sites and accident zones highlighted with amber warning zones. Users can make entirely autonomous decisions based on the augmented information they see. If they decide to use the route, they must actively trigger a selection command, such as confirming through eye tracking, making a specific gesture, or tapping a screen button. Only after receiving a clear manual selection command from the user will the target path be set as the navigation route.

[0128] like Figure 2 As shown, another embodiment of the present invention provides a dynamic navigation system based on multi-factor fusion and human-machine collaboration. The system includes a data fusion module, a path planning module, a human-machine collaborative decision-making module, and a navigation execution module.

[0129] The data fusion module is used to acquire and fuse multi-source data in real time.

[0130] The path planning module is used to input the fused multi-source data into the reinforcement learning model and output one or more alternative paths at the current time and the comprehensive cost of each alternative path. The dynamic cost function of the reinforcement learning model includes at least time cost, energy cost, comfort cost, and health risk cost.

[0131] The human-machine collaborative decision-making module is used to calculate the confidence level of each alternative path and the expected time saving relative to the current path for each alternative path, and select the target path from the alternative paths and determine the corresponding human-machine collaborative response level based on the preset three-level human-machine collaborative decision-making rules.

[0132] The navigation execution module is used to perform navigation operations corresponding to the selected target path and the human-machine collaborative response level.

[0133] In summary, this invention proposes a dynamic navigation method and system based on multi-factor fusion and human-machine collaboration. Through multi-source data fusion, reinforcement learning, dynamic cost functions, and a three-level human-machine collaborative decision-making rule, it achieves a unified balance between the dynamism, adaptability, security, and user experience of navigation services, thereby improving the quality of navigation services. Specifically, firstly, based on real-time multi-source data fusion, it integrates multi-dimensional data such as traffic, environment, and personalized user health data, overcoming the shortcomings of traditional navigation data sources being singular and fragmented. This provides comprehensive and accurate real-time data support for route planning, effectively avoiding route deviations caused by information lag or missing information. Furthermore, a reinforcement learning model integrating attention mechanisms and improved experience replay mechanisms is employed to enhance the adaptability and learning ability to complex and dynamic scenarios, enabling continuous optimization of decisions based on real-time interactions. To further coordinate diverse travel goals, a dynamic cost function is constructed that includes time cost, energy cost, comfort cost, and health risk cost. The weights of each cost item can be dynamically adjusted based on driving mode, user feedback, and real-time scenarios. This design breaks through the traditional navigation's singular focus on travel efficiency. While meeting core travel speed requirements, it also considers energy conservation, user experience, and health risk avoidance. It is particularly suitable for health-sensitive, comfort-oriented, and energy-saving users, significantly expanding the applicable scenarios and user groups for navigation services. Simultaneously, it continuously evaluates and weighs multiple, even conflicting, goals during travel, dynamically seeking the path with the optimal long-term comprehensive benefits in complex urban road networks to simultaneously meet users' comprehensive expectations for efficiency, economy, comfort, and safety. Finally, relying on pre-set three-level human-machine collaborative decision-making rules, and combining the confidence level of alternative routes with the expected time savings, a tiered response is achieved. In scenarios with high confidence and high time savings, routes are automatically switched to reduce the user's operational burden and improve travel efficiency. In the medium confidence and time-saving range, voice prompts are provided and user confirmation is awaited, balancing efficiency and user autonomy. In low confidence or complex road conditions, visual prompts are provided through an augmented reality head-up display, preserving the user's right to manually select routes and avoiding the decision-making errors that may arise from fully automated navigation. This differentiated response mode reduces unnecessary interference, ensures travel safety, and builds an efficient and reliable human-machine collaborative relationship.

[0134] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dynamic navigation method based on multi-factor fusion and human-machine collaboration, characterized in that, include: Real-time acquisition and fusion of multi-source data; The fused multi-source data is input into the reinforcement learning model, which outputs one or more alternative paths at the current time and the comprehensive cost value of each alternative path. The dynamic cost function of the reinforcement learning model includes at least a time cost term, an energy cost term, a comfort cost term, and a health risk cost term. For each alternative path, calculate the confidence level of the alternative path and the expected time saving relative to the current path, and select the target path from the alternative paths and determine the corresponding human-machine collaborative response level based on the preset three-level human-machine collaborative decision-making rules. Perform navigation operations corresponding to the selected target path and human-machine interaction response level.

2. The dynamic navigation method based on multi-factor fusion and human-machine collaboration according to claim 1, characterized in that, The multi-source data includes real-time traffic data, environmental data, and personalized health data of users.

3. The dynamic navigation method based on multi-factor fusion and human-machine collaboration according to claim 1, characterized in that, The calculation of the time cost item is based on the intersection green light passage probability obtained from the passage probability model; the passage probability model is a machine learning model, and the input includes at least the distance of the vehicle from the intersection, the current vehicle speed, the remaining time of the traffic light, and the historical data of the average green light interval of the intersection phase; The health risk cost item is calculated based on environmental data and personalized health data.

4. The dynamic navigation method based on multi-factor fusion and human-machine collaboration according to claim 1, characterized in that, The reinforcement learning model incorporates an attention mechanism and an improved experience replay mechanism. The weight parameters of each cost item in the dynamic cost function can be dynamically adjusted; the dynamic adjustment is achieved through a combination of one or more of the following methods: weight adaptation based on the driving mode selected by the user, meta-learning optimization based on user feedback, and weight template switching based on real-time scene recognition.

5. The dynamic navigation method based on multi-factor fusion and human-machine collaboration according to claim 1, characterized in that, The preset three-level human-machine collaborative decision-making rules include: The first-level human-machine collaborative decision-making rule is used to select an automatic execution level response when the confidence of any alternative path is higher than the first confidence threshold and the expected time saving is greater than the first time threshold. The two-level human-machine collaborative decision-making rule is used to select and confirm the execution level response when the confidence level of any alternative path is between the first confidence threshold and the second confidence threshold, and the expected time saving is between the first time threshold and the second time threshold. The three-level human-machine collaborative decision-making rule is used to select a suggestion-level response when the confidence level of any alternative path is lower than the second confidence threshold, or when the environment of the alternative path is determined to be a complex road condition.

6. The dynamic navigation method based on multi-factor fusion and human-machine collaboration according to claim 5, characterized in that, The confidence level is calculated based on a weighted average of data reliability, driving habit fit, and road condition complexity fit. The expected time savings The calculation method is as follows: In the formula, The estimated time to travel the current route; The estimated time to travel the alternative route.

7. The dynamic navigation method based on multi-factor fusion and human-machine collaboration according to claim 5, characterized in that, The process of selecting a target path from alternative paths includes: Based on confidence level, expected time savings, and preset three-level human-machine collaboration decision-making rules, the human-machine collaboration response level corresponding to each alternative path is determined. Based on the priority of human-machine collaboration response levels, the alternative path with the highest priority is determined as the target path; among them, the automatic execution level response has the highest priority, the confirmation execution level response has the second highest priority, and the suggestion prompt level response has the lowest priority. If two or more alternative paths have the same highest priority, the alternative path with the lowest overall cost will be selected as the target path.

8. The dynamic navigation method based on multi-factor fusion and human-machine collaboration according to claim 5, characterized in that, The navigation operations corresponding to the selected target path and the human-machine collaborative response level include: When the human-machine collaboration response level corresponding to the target path is automatic execution level, the navigation path will be automatically switched to the target path, and a status notification will be sent to the user. When the human-machine collaboration response level corresponding to the target path is the confirmation execution level response, a voice prompt is issued to the user, and after obtaining the user's confirmation, the navigation path is switched to the target path; When the human-machine interaction response level corresponding to the target path is the suggestion prompt level, the target path is visually prompted through the augmented reality head-up display, and the user's manual selection command is received. Based on the command, it is determined whether to switch the navigation path to the target path.

9. A dynamic navigation method based on multi-factor fusion and human-machine collaboration according to claim 8, characterized in that, The parameters for judging complex road conditions include at least one of the following: frequency of lane number change, average speed difference between adjacent road segments, traffic event density, and traffic light intersection density. The visual cues provided via augmented reality head-up display include lane-level blue light strips overlaid on the real road, dynamic green arrows at turns, and amber warning zones displayed on high-risk road sections.

10. A dynamic navigation system based on multi-factor fusion and human-machine collaboration, based on the dynamic navigation method based on multi-factor fusion and human-machine collaboration as described in any one of claims 1 to 9, characterized in that, include: The data fusion module is used to acquire and fuse multi-source data in real time; The path planning module is used to input the fused multi-source data into the reinforcement learning model and output one or more alternative paths at the current time and the comprehensive cost value of each alternative path. The dynamic cost function of the reinforcement learning model includes at least a time cost term, an energy cost term, a comfort cost term, and a health risk cost term. The human-machine collaborative decision-making module is used to calculate the confidence level and expected time saving relative to the current path for each candidate path, and select the target path from the candidate paths and determine the corresponding human-machine collaborative response level based on the preset three-level human-machine collaborative decision-making rules; the navigation execution module is used to execute the navigation operation corresponding to the selected target path and the human-machine collaborative response level.

Images