A method, apparatus, vehicle, and storage medium for handling memory conflicts.

Abstract

This disclosure provides a method, apparatus, vehicle, and storage medium for handling memory conflicts. The method includes: determining associated memories in a hierarchical memory bank based on user instructions, user state data, and driving scenario data; the user state data is determined based on user identity data and user emotion data; performing semantic conflict detection, logical conflict detection, and temporal conflict detection on the user instructions and associated memories sequentially to obtain conflict detection results; when the conflict detection results indicate that there is a conflict between the user instructions and associated memories, performing conflict resolution on the user instructions and associated memories based on a conflict resolution strategy to obtain a processed memory; the conflict resolution strategy includes a conflict coverage strategy, a conflict fusion strategy, and a conflict marking strategy; and updating the weights of the processed memory based on a dynamic adjustment algorithm to obtain the updated weight parameters of the processed memory.

Description

Technical Field

[0001] This disclosure relates to the field of vehicle technology, and more particularly to a method, apparatus, vehicle, and storage medium for handling memory conflicts. Background Technology

[0002] With the development of intelligent vehicle technology, the user interaction functions of in-vehicle systems are becoming increasingly complex, which places higher demands on memory information management.

[0003] While related technologies can utilize large-scale language models to retrieve and fuse basic memories, their memory management dimensions are too simplistic. Specifically, they lack detailed technical solutions for memory conflicts arising from contradictory information. Therefore, a new approach is urgently needed to address these issues. Summary of the Invention

[0004] One objective of this disclosure is to provide a method for handling memory conflicts to resolve the memory conflict problem; another objective is to provide a method for handling memory conflicts; a third objective is to provide a vehicle; and a fourth objective is to provide a storage medium.

[0005] To achieve the above objectives, the technical solution adopted in this disclosure is as follows: This disclosure proposes a method for handling memory conflicts. The method includes: determining associated memories in a hierarchical memory bank based on user commands, user state data, and driving scenario data; user state data is determined based on user identity data and user emotion data; performing semantic conflict detection, logical conflict detection, and temporal conflict detection on user commands and associated memories sequentially to obtain conflict detection results; when the conflict detection results indicate that there is a conflict between user commands and associated memories, performing conflict resolution on user commands and associated memories based on conflict resolution strategies to obtain processed memories; the conflict resolution strategies include conflict coverage strategies, conflict fusion strategies, and conflict marking strategies; and updating the weight parameters of processed memories based on a dynamic adjustment algorithm.

[0006] Based on the aforementioned technical methods, relevant associated memories are retrieved from a hierarchical memory bank using user commands, user states, and driving scenario data. Subsequently, conflict detection is performed on user commands and associated memories at three levels: semantic, logical, and temporal. When conflicts are confirmed, appropriate conflict resolution strategies are selected, including conflict coverage, conflict fusion, or conflict marking strategies, to resolve the conflicts. Finally, the processed memories are updated with weights through a dynamic adjustment algorithm, thereby optimizing the memory priority ranking. This approach not only enables accurate matching of related memories through multi-dimensional data but also comprehensively identifies inconsistencies between user commands and associated memories through semantic, logical, and temporal conflict detection, and flexibly handles them using multiple conflict resolution strategies. Therefore, it ensures close alignment with the user's actual state and scenario needs while continuously optimizing the priority of memory data, significantly improving the accuracy and adaptability of interactive decisions.

[0007] Furthermore, based on user commands, user status data, and driving scenario data, associated memories are identified in the hierarchical memory bank, including: Acquire user commands, user identity data, user emotion data, and driving scenario data; User identity data and user emotion data are identified to obtain identified user status data; The user commands and driving scenario data are transformed to obtain the first structured vector; Based on the identified user state data and the first structured vector, associated memories are determined in the hierarchical memory bank.

[0008] Based on the aforementioned technical methods, multiple types of information data are first collected, including user commands, user identity data, user emotion data, and driving scenario data. By integrating user identity data and user emotion data, identified user state data reflecting the user's current state is generated. Next, user commands and driving scenario data are converted into a first structured vector to facilitate subsequent rapid comparison and calculation. Finally, based on the identified user state data and the first structured vector, corresponding associated memories are accurately matched from a hierarchical memory database. This not only more realistically recreates the user's real-time state but also significantly improves the matching accuracy and retrieval efficiency of memory retrieval.

[0009] Furthermore, based on the dynamic adjustment algorithm, the processed and memorized weight parameters are updated, including: The output value of the reward function is used as the input of the dynamic adjustment algorithm to process the post-processed memory and obtain the updated weight parameters of the post-processed memory. The reward function is determined based on retrieval hit rate, user satisfaction, and resource utilization. The input data for the dynamic adjustment algorithm also includes: user behavior sequences, memory attributes of processed memories, and driving scenario data; the output data of the dynamic adjustment algorithm also includes: memory level of processed memories and retrieval priority of processed memories; user behavior sequences include multiple behaviors related to processed memories; memory attributes include at least memory type, retrieval frequency, and storage duration; driving scenarios include at least road type, road condition, and driving time.

[0010] Based on the aforementioned technical means, a reward function is constructed by considering retrieval hit rate, user satisfaction, and resource utilization rate. Its output value is used as the input for dynamically adjusting the algorithm. At the same time, the memory weight, memory level, and retrieval priority are updated by combining user behavior sequences, memory attributes, and driving scenario data. This allows for the accurate differentiation of the importance of memories, the reasonable division of memory levels and retrieval order, the improvement of retrieval efficiency for high-frequency memories, and the dynamic optimization of memories to adapt to different driving scenarios.

[0011] Furthermore, semantic conflict detection, logical conflict detection, and temporal conflict detection are performed sequentially on user instructions and associated memories to obtain conflict detection results, including: Semantic conflict detection is performed on user commands and associated memories to obtain semantic conflict detection results; Logical conflict detection is performed on user commands and associated memories to obtain logical conflict detection results; Perform timing conflict detection on user commands and associated memory to obtain timing conflict detection results; The conflict detection results include semantic conflict detection results, logical conflict detection results, and temporal conflict detection results.

[0012] Based on the aforementioned technical methods, the system first determines whether there is a contradiction in meaning between the user's instruction and associated memory, obtaining a semantic conflict detection result; secondly, it analyzes whether there are illogical relationships at the causal or common-sense level, obtaining a logical conflict detection result; finally, it verifies whether there is a reversal or conflict in the time sequence, obtaining a temporal conflict detection result. This detection sequence, from simple to complex, allows the simpler semantic detection in the early stages to provide basic information for subsequent detections, helping to make the subsequent logical and temporal detections more targeted. This ensures comprehensiveness while improving overall detection efficiency and effectively controlling computational load.

[0013] Furthermore, semantic conflict detection is performed on user commands and associated memories to obtain semantic conflict detection results, including: The user commands and associated memory are encoded separately to obtain the user command encoding vector and the associated memory encoding vector; Determine the cosine similarity between the user instruction encoding vector and the associated memory encoding vector; When the cosine similarity is less than a preset threshold and the semantic recognition results of the user instruction and the associated memory are different, the semantic conflict detection result is determined to be that there is a semantic conflict between the user instruction and the associated memory.

[0014] Based on the aforementioned technical methods, user commands and associated memories are encoded into vectors, and their cosine similarity is calculated. The existence of semantic conflicts is then determined by combining a similarity threshold with the semantic recognition results. This approach, using vector comparison to quantify semantic differences while referencing semantic recognition results for verification, effectively reduces misjudgments caused by relying solely on similarity assessments. The judgment logic is simple and efficient, lowering the error rate in semantic conflict detection.

[0015] Furthermore, logical conflict detection is performed on user commands and associated memories to obtain logical conflict detection results, including: Based on the vehicle rule base, forward and reverse matching are performed on user commands and associated memories to obtain matching results; Based on the matching results, it is determined whether there is a logical conflict between the user's instructions and associated memories, and the logical conflict detection results are obtained.

[0016] Based on the aforementioned technical methods, a bidirectional matching process (both forward and reverse) is performed between user commands and associated memories using an in-vehicle rule base. The matching results are then used to determine if any logical conflicts exist. This avoids oversights that can occur with one-way matching, accurately identifies logical conflicts, and, moreover, allows for logical conflict detection to align with real-world in-vehicle application scenarios, resulting in higher accuracy.

[0017] Furthermore, temporal conflict detection is performed on user commands and associated memories to obtain temporal conflict detection results, including: The validity period of user commands and associated memories is checked to determine whether there are any time-series conflicts between user commands and associated memories, and the time-series conflict detection results are obtained.

[0018] Based on the aforementioned technical methods, by verifying the valid time range of user commands and associated memories, it is determined whether there is a conflict between the two at the temporal level and a detection result is generated. In this way, contradictions between outdated memories and current commands in the time dimension can be quickly identified, and temporal misalignment problems can be accurately identified. Moreover, this method is simple and intuitive, and can efficiently complete the detection of temporal conflicts.

[0019] Furthermore, the method also includes: If the memory value score of the processed memory is greater than that of the first data and the usage frequency is greater than that of the second data, the processed memory will be stored in the high-speed storage of the edge device. If the memory value score of the processed memory is greater than that of the first data and the usage frequency is less than that of the second data, the processed memory will be stored in the memory database of the edge device. If the memory value score of the processed memory is lower than that of the first data and the frequency of use is lower than that of the second data, the processed memory will be stored in the external storage of the edge device.

[0020] Based on the aforementioned technical methods, memories are categorized according to their value rating and usage frequency, and then stored in high-speed storage, in-memory databases, and external storage respectively. This allows for differentiated allocation of storage media, enabling faster response times for high-frequency, high-value memories.

[0021] Furthermore, the method also includes: Based on the timestamps of the processed memories, the processed memories are deleted due to their expiration date, resulting in an updated hierarchical memory bank.

[0022] Based on the aforementioned technical means, and according to the memory value score, the memory data can be deleted in a timely manner, which can effectively clean up invalid data, save storage space, and improve overall processing efficiency.

[0023] This disclosure provides a memory conflict processing apparatus, which includes a determination module, a conflict processing module, and an update module.

[0024] The determination module is used to determine associated memories in the hierarchical memory bank based on user commands, user status data, and driving scenario data; user status data is determined based on user identity data and user emotion data; and semantic conflict detection, logical conflict detection, and temporal conflict detection are performed on user commands and associated memories in sequence to obtain conflict detection results. The conflict resolution module is used to resolve conflicts between user commands and associated memories when the conflict detection results indicate that there is a conflict between the user commands and associated memories, based on the conflict resolution strategy, to obtain the processed memory; the conflict resolution strategy includes conflict coverage strategy, conflict fusion strategy, and conflict marking strategy. The update module is used to update the processed and memorized weight parameters based on a dynamic adjustment algorithm.

[0025] This disclosure provides a vehicle, including: a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When executed by the processor, the computer program implements the steps of the driving assistance method integrating driving and parking as described in any of the first aspects above.

[0026] This disclosure provides a storage medium storing one or more computer programs, which can be executed by one or more processors to implement some or all of the steps in the above method.

[0027] This disclosure provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of the aforementioned method.

[0028] The memory conflict handling method provided in this disclosure retrieves relevant associated memories from a hierarchical memory bank based on user commands, user states, and driving scenario data. Subsequently, conflict detection is performed on the user commands and associated memories at three levels: semantic, logical, and temporal. When a conflict is confirmed, an appropriate conflict resolution strategy is selected, including conflict coverage, conflict fusion, or conflict marking strategies, to resolve the conflict. Finally, the processed memories are updated with weights using a dynamic adjustment algorithm to optimize memory priority ranking. This approach not only enables accurate matching of related memories through multi-dimensional data but also comprehensively identifies inconsistencies between user commands and associated memories through semantic, logical, and temporal conflict detection, and flexibly handles conflicts using multiple conflict resolution strategies. Therefore, it ensures close alignment with the user's actual state and scenario needs while continuously optimizing memory data priority, significantly improving the accuracy and adaptability of interactive decisions. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating a memory conflict handling method provided in an embodiment of this disclosure. Figure 1 ; Figure 2 This is a flowchart illustrating a memory conflict handling method provided in an embodiment of this disclosure. Figure 2 ; Figure 3 This is a flowchart illustrating a memory conflict handling method provided in an embodiment of this disclosure. Figure 3 ; Figure 4 This is a flowchart illustrating a memory conflict handling method provided in an embodiment of this disclosure. Figure 4 ; Figure 5 This is a schematic diagram of the structure of a memory conflict processing device provided in an embodiment of the present disclosure.

[0030] Figure 6 This is a schematic diagram of the structure of a vehicle provided in an embodiment of the present disclosure.

[0031] It should be noted that the terms "first" and "second" mentioned above are only used to distinguish between different options and do not represent the degree of superiority or inferiority of the options or their priority in the implementation process. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this disclosure clearer, the disclosure will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this disclosure. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0033] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the specific technical solutions of the application will be further described in detail below with reference to the accompanying drawings of the embodiments of this disclosure. The following embodiments are used to illustrate this disclosure, but are not intended to limit the scope of this disclosure.

[0034] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0035] In the following description, the terms "first," "second," and "third" are used only to distinguish different objects and do not represent a specific order of objects or have any chronological limitation. It is understood that "first," "second," and "third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this disclosure described herein can be implemented in an order other than that illustrated or described herein.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used herein is for the purpose of describing embodiments of this disclosure only and is not intended to be limiting of this disclosure.

[0037] This disclosure provides a method, apparatus, vehicle, and storage medium for handling memory conflicts.

[0038] Figure 1 A flowchart illustrating a method for handling memory conflicts provided in this disclosure is shown below. Figure 1 As shown, this method for handling memory conflicts can be implemented through steps 101-104: Step 101: Based on user commands, user status data, and driving scenario data, identify associated memories in the hierarchical memory bank.

[0039] Here, user commands can be either voice commands or text commands. For example, a user command could be: "Um, play a rock song."

[0040] In some embodiments, user status data includes user identity data, user emotion data, and context data.

[0041] User identity data is obtained through facial recognition (FACEID) verification based on the camera data captured by the in-vehicle driver monitoring system (DMS); or through voiceprint recognition verification. By identifying users, memory isolation can be implemented for different users, allowing each user to have an independent memory bank, thereby ensuring that memory confusion will not occur in scenarios where multiple people share a vehicle.

[0042] User emotion data is derived by fusing and recognizing the user's voice tone and facial expression features. For example, voice tone features include speech rate and pitch, while facial expression features include frowning and smiling. After identifying the user's emotion data, the system can selectively access content from its memory bank. For instance, when the user's emotion data indicates a depressed mood, it will prioritize recalling memories associated with "soothing music" rather than "rock music."

[0043] Scene data includes at least vehicle speed, acceleration, road conditions, driving time, and road type. For example, road conditions include congestion or smooth traffic. Driving time includes daytime or nighttime. Road type includes highway or urban areas. After obtaining the scene data, precise recommendations can be made using it. For example, when a user makes food-related recommendations while on the highway, the system prioritizes recalling memories of "nearest service areas" rather than memories of "nearby restaurants."

[0044] Driving scenario data refers to driving data that can be directly obtained through vehicle sensors, such as vehicle speed, acceleration, steering wheel angle, yaw rate, brake pedal opening, accelerator pedal opening, gear status, and turn signal status.

[0045] Upon receiving user instructions, user status data, and driving scenario data, associated memories can be determined in the hierarchical memory bank based on these data.

[0046] Here, associative memory refers to memories that are matched based on user commands, user status data, and driving scenario data, and where the matching degree exceeds a first threshold. It can be understood that associative memory can include a single memory record or multiple memory records.

[0047] A hierarchical memory bank can include all memory data corresponding to the current user. Of course, a hierarchical memory bank can store memory data for multiple users, and the system can distinguish memory data from different users by the user's unique ID.

[0048] By using user commands, user status data, and driving scenario data as a basis, data comparison and filtering are performed in a hierarchical memory bank to identify matching memory data as associated memories. In this way, associated memories that are highly consistent with the current actual driving environment, the user's real needs, and real-time status can be accurately obtained, effectively avoiding interference from irrelevant memories.

[0049] Step 102: Semantic conflict detection, logical conflict detection, and temporal conflict detection are performed on the user instructions and associated memories in sequence to obtain the conflict detection results.

[0050] After identifying the associated memory, it is possible to determine whether there is a conflict between the user command and the associated memory. If there is a conflict, the user command and the associated memory can be further processed to achieve memory management.

[0051] In some embodiments, the process of determining whether there is a conflict between user instructions and associated memories may include: performing semantic conflict detection, logical conflict detection, and temporal conflict detection on user instructions and associated memories in sequence to obtain conflict detection results.

[0052] Here, semantic conflict detection is used to check for conflicts in semantic understanding between user instructions and associated memories. Logical conflict detection is used to check for conflicts in logical reasoning between user instructions and associated memories. Temporal conflict detection is used to check for conflicts in temporal order between user instructions and associated memories.

[0053] By performing semantic conflict detection, logical conflict detection, and temporal conflict detection on user instructions and associated memories in order of increasing difficulty, the accuracy of the detection results can be guaranteed, and the computational load of the entire process can be effectively controlled, avoiding resource shortage problems caused by excessive computation.

[0054] In some embodiments, after performing semantic conflict detection on the user instructions and associated memories in sequence, and determining that a semantic conflict exists, the user instructions and associated memories are then subjected to logical conflict detection in sequence. If a logical conflict exists, the user instructions and associated memories are then subjected to temporal conflict detection in sequence.

[0055] Alternatively, after performing semantic conflict detection on user instructions and associated memories in sequence, and determining that there is no semantic conflict, continue to perform logical conflict detection on user instructions and associated memories in sequence, and if no logical conflict is found, continue to perform temporal conflict detection on user instructions and associated memories in sequence.

[0056] As can be seen from the above, regardless of whether there is a semantic conflict between the user instruction and the associated memory, it is necessary to continue to judge the logical conflict; and regardless of whether there is a logical conflict between the user instruction and the associated memory, it is necessary to continue to judge the temporal conflict. In this way, the comprehensiveness and accuracy of conflict detection can be guaranteed.

[0057] Step 103: When the conflict detection result indicates that there is a conflict between the user instruction and the associated memory, the conflict is resolved based on the conflict resolution strategy to obtain the processed memory.

[0058] Here, conflict resolution strategies include conflict coverage strategies, conflict fusion strategies, and conflict labeling strategies.

[0059] In some embodiments, a conflict coverage strategy is applied when there is an explicit conflict between a user instruction (i.e., a new memory) and an associated memory (i.e., an old memory) (e.g., an address change, a contact information update, etc.).

[0060] A conflict-overlap strategy refers to completely replacing an old memory with a new one. For example, a conflict-overlap strategy can refer to replacing an associated memory with a user instruction; alternatively, a conflict-overlap strategy can also refer to replacing a user instruction with an associated memory.

[0061] When there is an implicit conflict between new and old memories (e.g., changes in dietary preferences, expansion of music styles, etc. (a user first liked Jay Chou, then liked Li Jian)), a conflict fusion strategy is applicable.

[0062] The conflict fusion strategy refers to establishing an association between user commands and associated memories when there are compatible conflicts, and setting a first flag. This first flag indicates that the user command and associated memory belong to the same group of preferences, and both are displayed simultaneously during recommendation. It's understandable that the weights corresponding to user commands and associated memories can be the same or different. For example, when the time corresponding to the user command is later than the time corresponding to the associated memory, the weight of the associated memory is greater than the weight of the user command.

[0063] The conflict marking strategy refers to setting a second flag for mutual exclusion conflicts when user commands and associated memories include such conflicts. The second flag is used to trigger a prompt pop-up to obtain user feedback on the mutual exclusion conflict.

[0064] Conflict marking strategies are suitable for high-value information conflicts, such as two versions of a home address.

[0065] The query pop-up can be displayed immediately upon detection of a conflict, or it can be displayed collectively within a fixed time window. When the query pop-up is displayed collectively within a fixed time window, it can initiate multiple mutually exclusive and conflicting queries at once, allowing users to respond to multiple results in a single operation.

[0066] In some embodiments, when there is a conflict between user instructions and associated memories, the conflict resolution strategy described above is used to process the conflict between user instructions and associated memories to obtain processed memories.

[0067] The post-processed memory may consist only of user instructions, or only of associated memories, or both user instructions and associated memories. The content of the post-processed memory can be determined based on the specific content of the user instructions and associated memories, as well as the conflict resolution strategy; this application does not limit this.

[0068] For example, when the conflict resolution strategy is a conflict coverage strategy, the post-processing memory consists of user instructions or associative memories. When the conflict resolution strategy is a conflict fusion strategy, the post-processing memory includes both user instructions and associative memories. When the conflict resolution strategy is a conflict marking strategy, the post-processing memory is determined based on user feedback.

[0069] Step 104: Update the processed and memorized weight parameters based on the dynamic adjustment algorithm.

[0070] In some embodiments, the processed and memorized weight parameters can be updated based on a dynamic adjustment algorithm.

[0071] Because the dynamic adjustment algorithm incorporates user behavior, it can flexibly adjust the weight of memories based on user actions, allowing memories to evolve with user behavior. This ensures the stability of the retrieval strategy while rapidly responding to changes in user behavior. For example, when a user selects a new address three times consecutively, the system will give a positive reward to the action of "increasing the weight of the new address," so that the new address will be prioritized when a similar conflict occurs next time.

[0072] In some embodiments, the method further includes: determining a memory value score for the processed memory.

[0073] For example, the process of determining the memory value score of a processed memory may include: calculating the memory value score of a processed memory by weighting it based on the frequency of use, importance, timeliness and user preference.

[0074] The weights of frequency of use, importance, timeliness, and user preference can be dynamically adjusted by the reinforcement learning model.

[0075] By calculating the memory value score of each memory in real time based on usage frequency, importance, timeliness, and user preferences, memories with higher scores are prioritized for storage on high-speed media. Consequently, memories with higher scores will also be ranked higher during retrieval. For example, if a user calls their "home address" every day, the memory value score of this memory will continuously increase, and the storage location and storage level of this memory will change accordingly, such as upgrading the memory level of "home address" from short-term memory to permanent memory.

[0076] In this embodiment, relevant associated memories are retrieved from a hierarchical memory bank based on user commands, user status, and driving scenario data. Subsequently, conflict detection is performed on the user commands and associated memories at three levels: semantic, logical, and temporal. When a conflict is confirmed, an appropriate conflict resolution strategy is selected, including conflict coverage, conflict fusion, or conflict marking strategies, to resolve the conflict. Finally, the processed memories are updated with weights using a dynamic adjustment algorithm, thereby optimizing the memory priority ranking. This approach not only enables accurate matching of related memories through multi-dimensional data but also comprehensively identifies inconsistencies between user commands and associated memories through semantic, logical, and temporal conflict detection, and flexibly handles conflicts using multiple conflict resolution strategies. Therefore, it ensures close alignment with the user's actual state and scenario needs while continuously optimizing the priority of memory data, significantly improving the accuracy and adaptability of interactive decisions.

[0077] The following section explains the process of determining associated memories in the hierarchical memory bank based on user instructions, user status data, and driving scenario data in step 101.

[0078] In one possible implementation, the process may include, but is not limited to, steps 201-204 below.

[0079] Step 201: Obtain user commands, user identity data, user emotion data, and driving scenario data.

[0080] In some embodiments, user commands, user identity data, user emotion data, and driving scenario data can be acquired through sensors or cameras on the vehicle.

[0081] The scene data and driving scene data may contain duplicate content, such as vehicle speed and acceleration.

[0082] In some embodiments, the memory conflict handling method is applied to a memory conflict handling architecture, which includes an interactive input layer, a preprocessing layer, a memory retrieval layer, a conflict handling layer, and a memory management layer.

[0083] The process of acquiring user instructions, user identity data, user emotion data, and driving scenario data may also include: the interaction input layer acquiring user instructions, user identity data, user emotion data, and driving scenario data, and inputting the user instructions, user identity data, user emotion data, and driving scenario data into the preprocessing layer.

[0084] Step 202: Identify user identity data and user emotion data to obtain identified user status data.

[0085] In some embodiments, the process of identifying user identity data and user emotion data to obtain identified user state data may include: identifying user identity data, user emotion data, and scenario data to obtain identified user state data.

[0086] As mentioned above, user identity data and user emotion data can be used to describe some explicit information about a user. By identifying user identity data, user emotion data, and scenario data, the identified user state data can be determined. For example, the identified user state data includes: in a nighttime highway driving scenario, user A is currently in a state of fatigue.

[0087] Step 203: Transform the user commands and driving scenario data to obtain the first structured vector.

[0088] In some embodiments, after obtaining user instructions and driving scenario data, the user instructions and driving scenario data can be transformed to obtain a first structured vector (also known as a first structured feature vector).

[0089] In some embodiments, the process of converting user instructions and driving scenario data to obtain a first structured vector may include: converting user instructions to obtain a structured feature vector of user instructions; converting driving scenario data to obtain a structured feature vector of driving; and concatenating the structured feature vector of user instructions and the structured feature vector of driving to obtain the first structured vector.

[0090] By transforming user commands and driving scenario data, a first structured vector is obtained. This first structured vector can be used to filter out invalid noise in user commands (e.g., filtering out invalid interjections such as "hey" or "that"), thereby ensuring that the key information of the user commands is completely preserved and improving the effectiveness of subsequent processing.

[0091] In some embodiments, user identity data and user emotion data are identified to obtain identified user state data.

[0092] In some embodiments, the process of identifying user identity data and user emotion data to obtain identified user state data, and transforming user instructions and driving scenario data to obtain a first structured vector, may further include: a preprocessing layer identifying user identity data, user emotion data, and scenario data to obtain identified user state data, and transforming user instructions and driving scenario data to obtain a first structured vector; and sending the identified user state data and the first structured vector to a memory retrieval layer.

[0093] Step 204: Based on the identified user state data and the first structured vector, determine the associated memory in the hierarchical memory bank.

[0094] After obtaining the identified user state data and the first structured vector, the associated memory can be determined in the hierarchical memory bank based on the identified user state data and the first structured vector.

[0095] In some embodiments, the process of determining associated memories in a hierarchical memory bank based on the identified user state data and the first structured vector may include: using the keywords in the identified user state data and the first structured vector as index keys, performing a retrieval operation in the hierarchical memory bank, thereby determining the associated memories that match the index keys.

[0096] In some embodiments, the process of determining associated memories in a hierarchical memory bank based on the identified user state data and the first structured vector may include: the memory retrieval layer determining associated memories in the hierarchical memory bank based on the identified user state data and the first structured vector.

[0097] In this embodiment, multiple types of information data are first collected, including user commands, user identity data, user emotion data, and driving scenario data. By integrating user identity data and user emotion data, identified user state data reflecting the user's current state is generated. Then, user commands and driving scenario data are converted into a first structured vector for subsequent rapid comparison and calculation. Finally, based on the identified user state data and the first structured vector, corresponding associated memories are accurately matched from a hierarchical memory database. This not only more realistically restores the user's real-time state but also significantly improves the matching accuracy and retrieval efficiency of memory retrieval.

[0098] The following explains the process of updating the processed and memorized weight parameters in step 104 based on the dynamic adjustment algorithm.

[0099] In one possible implementation, the process may include, but is not limited to, step 1041 below.

[0100] Step 1041: Use the output value of the reward function as the input of the dynamic adjustment algorithm to calculate the processed memory and obtain the updated weight parameters of the processed memory.

[0101] The reward function is determined based on retrieval hit rate, user satisfaction, and resource utilization.

[0102] In some embodiments, the process of using the output value of the reward function as the input of the dynamic adjustment algorithm to calculate the processed memory and obtain the updated weight parameters of the processed memory may include: the memory management layer using the output value of the reward function as the input of the dynamic adjustment algorithm to calculate the processed memory and obtain the updated weight parameters of the processed memory.

[0103] In some embodiments, the reward function satisfies the following expression: Rtotal=α Rhit+β Rsatisfy+γ Rresource formula (1) Where α+β+γ=1, Rhit is the search hit rate, Rsatisfy is the user satisfaction rate, and Rresource is the resource utilization rate. For example, α=0.4, β=0.4, γ=0.2.

[0104] Search hit rate refers to the proportion of searches that correctly retrieve the user's memory out of the total number of searches. Positive rewards are weighted higher than negative rewards to encourage accurate searches and prevent overly conservative search strategies due to excessive penalties.

[0105] User satisfaction includes explicit feedback data and implicit feedback data. Explicit feedback data is determined based on users' direct evaluations, and includes at least user voice evaluation data and facial expression data. Examples include users' direct voice evaluations (yes, no, etc.) and facial expression data (smiling, frowning, etc.).

[0106] Implicit feedback data is determined based on indirect evaluation information extracted from user actions. Implicit feedback data includes at least the duration of user interaction on the interface and actions taken to modify the system's output. For example, the duration of user interaction or whether search results were manually modified.

[0107] Resource utilization refers to the ratio of endpoint storage space occupied by the memory database to computing resources. Here, endpoint storage space includes at least endpoint cache (such as LPDDR5) and endpoint in-memory database (such as Redis).

[0108] Under the condition of adapting to the resource constraints of the cockpit, the above reward function can balance the search hit rate, user satisfaction and resource utilization, and expand the evaluation dimensions of user satisfaction.

[0109] In some embodiments, the input data to the dynamic adjustment algorithm further includes: user behavior sequences, post-processed memory attributes, and driving scenario data. The output data of the dynamic adjustment algorithm further includes: the post-processed memory level and the post-processed memory retrieval priority.

[0110] The user behavior sequence includes multiple post-processing memory-related behaviors; the memory attributes include at least memory type, retrieval frequency, and storage duration; the driving scenario includes at least road type, road condition, and driving time.

[0111] In some embodiments, the dynamic adjustment algorithm includes a state input layer, an action output layer, and a reward function layer.

[0112] The input data for the state input layer includes user behavior sequences, post-processed memory attributes, and driving scenario data. For example, if the post-processed memory is related to music playback, the user behavior sequence could include the last 10 music playback records. Or, if the post-processed memory is related to navigation addresses, the user behavior sequence could include the last 10 navigation address change records. Here, the memory type can be long-term memory, short-term memory, or transient memory. The retrieval frequency can be the frequency at which the post-processed memory is accessed, and the storage duration can be the duration for which the post-processed memory is stored.

[0113] For example, if the input data from the state input layer indicates that the user is more likely to listen to rock music on the highway, the system will record this scene characteristic and adjust the memory weights.

[0114] The output data of the action output layer includes the weight parameters updated after processing, the memory level of the processed memory, and the retrieval priority of the processed memory.

[0115] For example, the weight parameters after the post-processing memory update could be to increase the weight of new addresses (such as user instructions) and decrease the weight of old addresses (such as associative memory).

[0116] The memory level of processed memories can be changed; for example, "frequently visited coffee shops" can be changed from short-term memory to long-term memory. The retrieval priority of processed memories can also change; for example, "home address" can be set to the highest retrieval priority. In this way, through reinforcement learning-driven dynamic adjustment algorithms, the system can automatically adjust the weight parameters, memory level, and retrieval priority of processed memories, making the entire process more intelligent and requiring no human intervention.

[0117] In this embodiment, a reward function is constructed by retrieval hit rate, user satisfaction and resource utilization rate, and its output value is used as the input of the dynamic adjustment algorithm. At the same time, the memory weight, memory level and retrieval priority are updated by combining user behavior sequence, memory attributes and driving scenario data, so as to accurately distinguish the importance of memory, reasonably divide memory level and retrieval order, improve the retrieval efficiency of high-frequency memory, and adapt to different driving scenarios to realize dynamic optimization of memory.

[0118] The following describes the process of performing semantic conflict detection, logical conflict detection, and temporal conflict detection on user instructions and associated memories in step 102 to obtain conflict detection results.

[0119] In one possible implementation, the process may include, but is not limited to, steps 301-303 described below.

[0120] Step 301: Perform semantic conflict detection on user commands and associated memories to obtain semantic conflict detection results.

[0121] Step 302: Perform logical conflict detection on user commands and associated memories to obtain logical conflict detection results.

[0122] Step 303: Perform timing conflict detection on user instructions and associated memory to obtain timing conflict detection results.

[0123] The conflict detection results include semantic conflict detection results, logical conflict detection results, and temporal conflict detection results.

[0124] In some embodiments, when performing conflict detection on user instructions and associated memories, semantic conflict detection can be performed first, followed by logical conflict detection, and finally temporal conflict detection, in a sequence from simple to complex. This ensures the accuracy of conflict detection while controlling the computational load.

[0125] Of course, when performing conflict detection on user commands and associated memories, logical conflict detection can be performed first, followed by semantic conflict detection, and finally temporal conflict detection. The order of conflict detection can also be adjusted according to actual usage.

[0126] In some embodiments, the conflict detection results may include only semantic conflict detection results, or only logical conflict detection results, or a combination of semantic conflict detection results, logical conflict detection results, and temporal conflict detection results. This application does not limit the content of the conflict detection results; the actual use shall prevail.

[0127] In this embodiment, the system first determines whether there is a contradiction in meaning between the user instruction and the associated memory, obtaining a semantic conflict detection result; secondly, it analyzes whether there are unreasonable relationships at the causal or common-sense level, obtaining a logical conflict detection result; finally, it verifies whether there is a reversal or conflict in the time sequence, obtaining a temporal conflict detection result. This detection order from simple to complex allows the simpler semantic detection in the early stages to provide basic information for subsequent detections, helping subsequent logical and temporal detections to be more targeted, thereby improving overall detection efficiency while ensuring comprehensiveness and effectively controlling the amount of computation.

[0128] In some embodiments, the priority of various types of memories in the conflict resolution strategy is as follows: user-confirmed memory information > memory information with the shortest time interval to the current time > memory information with a confidence level greater than a first value > memory information subject to time constraints.

[0129] The following describes the process of performing semantic conflict detection on user instructions and associated memories in step 301 to obtain semantic conflict detection results.

[0130] In one possible implementation, the process may include, but is not limited to, steps 401-403 below.

[0131] Step 401: Encode the user instructions and the associated memory separately to obtain the user instruction encoding vector and the associated memory encoding vector.

[0132] In some embodiments, the process of encoding user instructions and associated memory separately to obtain user instruction encoding vectors and associated memory encoding vectors may include: using a Distilled Bidirectional Encoder Representations from Transformers (DistilBERT) model to encode user instructions and associated memory separately to obtain user instruction encoding vectors and associated memory encoding vectors.

[0133] Since DistilBERT's parameters are only 40% of those of the Bidirectional Encoder Representations from Transformers (BERT) model, it can guarantee a 60% improvement in overall inference speed.

[0134] Step 402: Determine the cosine similarity between the user instruction encoding vector and the associated memory encoding vector.

[0135] After determining the user instruction encoding vector and the associated memory encoding vector, the cosine similarity between the user instruction encoding vector and the associated memory encoding vector can be calculated, and then the semantic conflict between the user instruction and the associated memory can be further determined based on the cosine similarity.

[0136] Step 403: When the cosine similarity is less than the preset threshold and the semantic recognition results of the user instruction and the associated memory are different, the semantic conflict detection result is determined to be that there is a semantic conflict between the user instruction and the associated memory.

[0137] In some embodiments, when the cosine similarity is greater than a preset threshold, the semantic conflict detection result is determined to be that there is no semantic conflict between the user instruction and the associated memory.

[0138] In some embodiments, when the cosine similarity is lower than a preset threshold and the semantic recognition results of the user instruction and the associated memory are the same, the semantic conflict detection result is determined to be that there is no semantic conflict between the user instruction and the associated memory.

[0139] To further reduce inference time, the associative memory can be pre-encoded and the associative memory encoding vector can be cached, thus eliminating the need for real-time computation.

[0140] In this embodiment, user instructions and associated memories are encoded into vectors, their cosine similarity is calculated, and a dual determination of semantic conflict is made by combining a similarity threshold and semantic recognition results. This approach, using vector comparison to quantify semantic differences while referencing semantic recognition results for verification, effectively reduces misjudgments caused by a single similarity assessment. The determination logic is simple and efficient, lowering the error in semantic conflict detection.

[0141] The following describes the process of performing logical conflict detection on user instructions and associated memories in step 302 to obtain the logical conflict detection results.

[0142] In one possible implementation, the process may include, but is not limited to, the following steps 501-502.

[0143] Step 501: Based on the vehicle rule base, perform forward and reverse matching on user commands and associated memories to obtain matching results.

[0144] Step 502: Based on the matching results, determine whether there is a logical conflict between the user command and the associated memory, and obtain the logical conflict detection results.

[0145] As mentioned above, regardless of whether there is a semantic conflict between the user instruction and the associated memory, it is necessary to further determine whether there is a logical conflict between the user instruction and the associated memory.

[0146] In some embodiments, the process of determining whether there is a logical conflict between user commands and associated memories may include: performing forward and reverse matching on user commands and associated memories based on an in-vehicle rule base to obtain matching results, and determining whether there is a logical conflict between user commands and associated memories based on the matching results.

[0147] Based on the vehicle rule base, the process of performing forward and reverse matching on user commands and associated memories to obtain matching results may include: identifying and extracting target entities from user commands and associated memories, verifying the logical relationship between the extracted target entities, and determining whether they meet the constraints in the vehicle rule base.

[0148] Here, the in-vehicle rule base can cover core scenarios such as navigation, music, and communication. For example, the in-vehicle rule base includes rules such as: navigation commands cannot point to two destinations at the same time, music styles cannot be classical and rock at the same time, and call commands cannot request to dial two different contacts at the same time.

[0149] For example, based on the predefined entity types in the vehicle rule base (such as the starting point, destination, and waypoints in navigation scenarios; song names, artist names, and playlist names in music scenarios; and contact names and phone numbers in communication scenarios), the system scans and matches user commands and associated memories to complete the identification and extraction of target entities. Then, it checks whether "Destination A" and "Destination B" appear at the same time, thus violating navigation rules, or whether "classical music" and "rock music" are set at the same time, thus violating music rules, etc.

[0150] In the above process, key elements can be extracted from user commands and associated memories through forward matching, and then the logical relationships between these key elements are verified through reverse matching. By utilizing forward and reverse matching in tandem, a dual "identification-verification" mechanism can be formed to accurately determine the matching results between user commands and associated memories. In this way, the false positive rate of logical conflicts can be controlled below 3%, which can effectively improve the reliability of memory retrieval and command execution in in-vehicle scenarios.

[0151] In some embodiments, the vehicle rule base supports over-the-air (OTA) online updates, which can quickly adapt to new scenarios. For example, the updated content of the vehicle rule base includes charging pile preferences.

[0152] In this embodiment, a vehicle-mounted rule base is used to perform bidirectional matching of user commands and associated memories in both forward and reverse directions. The matching results are then used to determine whether logical conflicts exist. This avoids oversights that can occur with one-way matching, accurately identifies logical conflicts, and, based on the vehicle-mounted rule base, allows for logical conflict detection to align with real-world vehicle application scenarios, resulting in higher accuracy.

[0153] The following describes the process of performing time conflict detection on user instructions and associated memory in step 303 to obtain the time conflict detection result.

[0154] In one possible implementation, the process may include, but is not limited to, step 601 below.

[0155] Step 601: Verify the validity period of user instructions and associated memories to determine whether there are any temporal conflicts between user instructions and associated memories, and obtain the conflict detection results.

[0156] As mentioned above, regardless of whether there are semantic and logical conflicts between user instructions and associated memories, it is necessary to further determine whether there are temporal conflicts between user instructions and associated memories.

[0157] In some embodiments, the process of verifying the validity period of user instructions and associated memories to determine whether there is a temporal conflict between the user instructions and associated memories, and obtaining a conflict detection result, may include: checking the validity period of user instructions and associated memories; if the validity period of user instructions and associated memories overlap, then determining whether there is a temporal conflict between user instructions and associated memories based on the overlapping content; if the validity period of user instructions and associated memories do not overlap, but the content conflicts, then determining whether there is a temporal conflict between user instructions and associated memories; if user instructions and associated memories do not have a clear validity period, then determining whether there is a temporal conflict between user instructions and associated memories based on the context of the user memory.

[0158] The validity period of user instructions and associated memories includes explicit and implicit validity periods. For example, an explicit validity period is: "This week's traffic restriction rules." An implicit validity period is: "The default validity period for a 'temporary customer address' is 7 days."

[0159] If the validity period of the user instruction and the validity period of the associated memory do not overlap (e.g., "last year's company address" and "this year's company address"), then it is determined that there is a temporal conflict between the user instruction and the associated memory.

[0160] If user commands and associated memories do not have a clear expiration date, the system will infer the temporal relationship from the context (such as "the temporary address mentioned last week" and "the new address mentioned this week").

[0161] It is understandable that after performing semantic conflict detection, logical conflict detection, and temporal conflict detection on user instructions and associated memories in sequence, it can be determined that there is only one type of conflict between user instructions and associated memories, or it can be determined that there are two or more types of conflicts between user instructions and associated memories. This application does not limit this, and the specific determination shall be subject to the actual situation.

[0162] In this embodiment, by verifying the valid time range of the user instruction and the associated memory, it is determined whether there is a conflict between the two at the temporal level and a detection result is generated. In this way, contradictions between outdated memory and the current instruction in the time dimension can be quickly identified, and temporal misalignment problems can be accurately identified. Moreover, this method is simple and intuitive, and can efficiently complete the detection of temporal conflicts.

[0163] In some embodiments, the method further includes: Step 701: If the memory value score of the processed memory is greater than that of the first data and the usage frequency is greater than that of the second data, the processed memory is stored in the high-speed storage of the edge device.

[0164] Step 702: If the memory value score of the processed memory is greater than that of the first data and the usage frequency is less than that of the second data, the processed memory is stored in the memory database of the edge device.

[0165] Step 703: If the memory value score of the processed memory is less than that of the first data and the frequency of use is less than that of the second data, the processed memory is stored in the external storage of the edge device.

[0166] In some embodiments, after the processed memory is obtained, it can be stored in a hierarchical manner based on the memory value score and the frequency of use of the processed memory.

[0167] For example, if the memory value score of the processed memory is greater than that of the first data and the usage frequency is greater than that of the second data (i.e., the score is high and the usage is frequent), the processed memory is stored in the high-speed storage of the edge device, so that the read and write latency is ≤1ms, which is suitable for temporary memory of a single interaction.

[0168] If the memory value score of the processed memory is greater than that of the first data and the usage frequency is less than that of the second data (i.e., the score is high and the usage is infrequent), the processed memory will be stored in the memory database of the edge device. This way, the read / write latency is ≤5ms, which is suitable for memories that are frequently used recently.

[0169] If the memory value score of the processed memory is lower than that of the first data point and the usage frequency is lower than that of the second data point (i.e., the score is low and the usage is infrequent), the processed memory will be stored in the external storage of the edge device (e.g., UFS 3.1). Because external storage has a large storage capacity and low cost, it is suitable for low-frequency but high-value memories.

[0170] In some embodiments, memory data can be stored in JSON format, specifically including fields such as memory ID, user ID, memory content, timestamp, memory type, memory level, usage frequency, weight parameter, and confidence level. The weight parameter is dynamically assigned by a dynamic adjustment algorithm, and its value ranges from 0 to 1. A higher weight parameter value results in a higher retrieval priority. Multi-level index structures, such as primary key indexes, time indexes, type indexes, and user indexes, can also be established for the memory data to achieve multi-dimensional, fast retrieval and querying.

[0171] In this embodiment, memories are categorized based on their value rating and usage frequency, and then stored in high-speed storage, an in-memory database, and external storage, respectively. This allows for differentiated allocation of storage media, enabling faster response times for high-frequency, high-value memories.

[0172] In some embodiments, the method further includes: Step 801: Based on the timestamp of the processed memory, perform time-sensitive deletion on the processed memory to obtain the updated hierarchical memory bank.

[0173] In some embodiments, all memory data can be scored within a fixed time period to obtain a memory value score for each memory data.

[0174] In some embodiments, the memory value score satisfies the following expression: Score=α×F+β×I+γ×T+δ×P Formula (2) Where F represents usage frequency, I represents importance, T represents timeliness, P represents user preference, and α, β, γ, and δ are weighting coefficients.

[0175] When the memory value score of memory data falls below a set threshold, the memory data is added to the cleanup queue. Generally, memory cleanup is triggered when the memory data in the hierarchical memory bank occupies 80% of the edge storage. During memory cleanup, low-scoring temporary and short-term memories are deleted first to ensure smooth system operation.

[0176] For example, when adopting the "batch elimination" principle, the memory data deleted each time does not exceed 10% of the total memory data, to avoid deleting too much memory data at once and causing system abnormalities. At the same time, a preset number (e.g., 1000) of high-frequency memory data will be retained as a basic cache, so that even after cleaning, basic interactive functions can still be guaranteed to work normally, that is, the system will retain at least 1000 high-frequency memory data.

[0177] In this embodiment of the application, memory data is deleted based on its memory value score, which can effectively clean up invalid data, save storage space, and improve overall processing efficiency.

[0178] In some embodiments, the method further includes: Step 901: When user instructions and associated memory storage overlap, merge user instructions and associated memory according to the memory type of associated memory to obtain a second memory.

[0179] Among them, memory types include at least navigational memory, preference memory, and contact memory.

[0180] In some embodiments, after performing semantic conflict detection, logical conflict detection, and temporal conflict detection on the user instruction and associated memory in sequence, and obtaining the conflict detection result, the conflict detection result can also indicate that there is no conflict between the user instruction and the associated memory, but there is overlap.

[0181] In this case, based on the memory type of the associated memory, the user's instructions are merged with the associated memory to obtain the second memory.

[0182] For example, if the user command and associated memory are navigation-type memories, then the user command and associated memory are merged according to the merging strategy corresponding to the navigation-type memories to obtain the second memory. The merging strategy corresponding to the navigation-type memories is: merging based on the latest timestamp combined with usage frequency.

[0183] For example, the associated memory is: "Company is located at Technology XXX". The user instruction is: "Company moves to Quanyan XXX". First, the usage frequency of the addresses corresponding to the associated memory and the user instruction is checked. If the new address corresponding to the user instruction is used more than twice, and the old address corresponding to the associated memory has no usage record, then the user instruction overwrites the associated memory. If the new address corresponding to the user instruction is used more than twice, and the old address corresponding to the associated memory still has usage records (e.g., once a week), then the old address corresponding to the associated memory is marked as an "alternative address". The associated memory will still be displayed during retrieval, but its priority is lower than that of the user instruction.

[0184] For example, if the user instruction and associated memory are preference-based memories, then the user instruction and associated memory are merged according to the merging strategy corresponding to preference-based memories to obtain the second memory. The merging strategy corresponding to preference-based memories is: merging with high confidence.

[0185] For example, if the associated memory is: "The user likes Jay Chou," and the user instruction is: "The user likes Li Jian," then the weighted preferences corresponding to the associated memory and the user instruction are added together (e.g., Jay Chou 0.6 + Li Jian 0.4) to generate a comprehensive preference label, which is then stored in association. During recommendation, the content corresponding to both preferences is covered simultaneously.

[0186] For example, if the user instruction and associated memory are contact memories, then the user instruction and associated memory are merged according to the merging strategy corresponding to the contact memories to obtain the second memory. The merging strategy corresponding to the contact memories is: merging using a multi-distance cross-validation strategy.

[0187] For example, the associated memory might be: "Zhang San's phone number is 123456." The user's command might be: "Zhang San's phone number is 123456." The system will first cross-validate the associated memory and user command against the contacts and call logs. If the information doesn't exist, it will add it. If the information exists but doesn't match, it will be marked as "pending confirmation" to prevent incorrect information from overwriting correct information.

[0188] In this embodiment, when user commands and associated memory storage overlap, the overlapping content is merged and integrated to generate new memories according to memory categories such as navigation, preferences, and contacts. This avoids duplicate storage of similar memories, saves storage space, and improves query efficiency.

[0189] Figure 2 A flowchart illustrating a method for handling memory conflicts provided in this disclosure is shown below. Figure 2 As shown, this method for handling memory conflicts can be implemented through the following steps: Step 11: Interactive Input Layer: Receive user commands and synchronously collect driving scenario data.

[0190] Step 12: Preprocessing layer: Convert user instructions into structured feature vectors and filter out invalid noise.

[0191] For example, user commands and driving scenario data are transformed to obtain a first structured vector.

[0192] Step 13: Memory Retrieval Layer: Based on the identified user state data and the first structured vector, quickly locate the associated memory in the hierarchical memory bank.

[0193] Step 14: Conflict handling layer: Detects conflicts between user commands and associated memories, and executes the corresponding conflict resolution strategy.

[0194] Step 15: Memory Management Layer: Perform hierarchical storage, weight updates, and time-sensitive deletion on the processed memories.

[0195] Figure 3 A flowchart illustrating another method for handling memory conflicts provided in this disclosure is shown below. Figure 3 As shown, this method for handling memory conflicts can be implemented through the following steps: Step 21: User command input.

[0196] Step 22: Retrieve relevant historical memories.

[0197] For example, based on user instructions, user status data, and driving scenario data, associated memories (i.e., relevant historical memories) are identified in a hierarchical memory bank.

[0198] Next, semantic conflict detection, logical conflict detection, and temporal conflict detection are performed on the user instructions and associated memories in sequence to obtain the conflict detection results.

[0199] Step 23: Conflict type determination.

[0200] Conflict detection identifies the specific type of conflict between user commands and associated memories.

[0201] Step 24: Conflict resolution.

[0202] Conflict resolution is performed on user commands and associated memories according to conflict resolution strategies (conflict coverage strategy, conflict fusion strategy, and conflict marking strategy).

[0203] Step 25: Remember the output and update.

[0204] After resolving the conflict using the conflict resolution strategy, the post-processed memory is obtained. Next, the weights of the post-processed memory are updated to obtain the updated weight parameters.

[0205] Figure 4 A flowchart illustrating yet another memory conflict resolution method provided in this disclosure is shown below. Figure 4 As shown, this method for handling memory conflicts can be implemented through the following steps: Step 31: User input.

[0206] Here, user input refers to the input of user commands.

[0207] Step 32: Preprocessing.

[0208] Here, preprocessing refers to transforming user commands and driving scenario data to obtain the first structured vector.

[0209] Step 33: User status identification.

[0210] Here, user status recognition refers to the process of identifying user identity data and user emotion data to obtain the identified user status data.

[0211] Step 34: Memory retrieval.

[0212] For example, based on user instructions, user status data, and driving scenario data, associated memories (i.e., relevant historical memories) are identified in a hierarchical memory bank; this process is memory retrieval.

[0213] Conflict detection identifies the specific type of conflict between user commands and associated memories.

[0214] Step 35: Conflict detection.

[0215] Semantic conflict detection, logical conflict detection, and temporal conflict detection are performed sequentially on user instructions and associated memories to obtain conflict detection results.

[0216] Step 36: Conflict resolution.

[0217] Conflict resolution is performed on user commands and associated memories according to conflict resolution strategies (conflict coverage strategy, conflict fusion strategy, and conflict marking strategy).

[0218] Step 37: Memory storage.

[0219] After the conflict is resolved through conflict resolution strategies, the processed memory is obtained and stored according to its memory value rating and usage frequency.

[0220] Step 38: System response.

[0221] Here, system response refers to updating the weight parameters of the processed memory based on a dynamic adjustment algorithm, and responding again to the subsequent memory retrieval process according to the updated weight parameters of the processed memory.

[0222] The memory conflict handling method provided in this disclosure will be described in detail below through an embodiment, combined with a specific application scenario.

[0223] With the rapid development of intelligent vehicle technology, the user interaction functions of in-vehicle systems are becoming increasingly complex, placing higher demands on memory information management.

[0224] Current approaches to user interaction memory management in in-vehicle systems mainly fall into the following categories: Regarding memory-based interaction methods, a scheme has been disclosed that uses a large language model to determine the target memory associated with a query statement, including long-term memory, short-term memory, and memory summaries. This technology fuses the target memory and the query statement to obtain prompt words, which are then input into a trained large language model to generate a response statement. However, this scheme primarily focuses on the retrieval and fusion of memory information, lacking in-depth design for complex conflict handling, dynamic adjustment, and timeliness management mechanisms. In terms of memory control for in-vehicle functional components, a memory control technology has been disclosed that allows the vehicle to remember user habits, such as seat position, air conditioning temperature, and music preferences, restoring these settings upon the next vehicle entry. This technology mainly targets hardware parameter memory and has limited capabilities in managing interactive text information. Regarding cognitive large-model voice systems, an application of a cognitive large model in in-vehicle interaction systems has been disclosed, giving the system stronger semantic understanding and contextual memory capabilities. However, this technology lacks detailed technical solutions for specific memory management mechanisms, conflict handling strategies, and response time control.

[0225] This disclosure provides a method for handling memory conflicts, the details of which are as follows.

[0226] 1. The technical solution disclosed herein adopts a "fully closed-loop end-side" design, where all memory management operations are completed locally in the vehicle, without relying on cloud computing power. This ensures both response speed and avoids the risk of uploading user privacy data. The system architecture consists of five layers, forming a complete memory governance chain from input to output: First layer: Interactive input layer Receive user commands and synchronously collect driving scenario data (such as vehicle speed, road conditions, and driver status).

[0227] Second layer: Preprocessing layer User commands and driving scenario data are converted into first-order structured vectors, and invalid noise (such as interjections like "hey" or "that") is filtered out.

[0228] Third layer: Memory retrieval layer Based on the user's state and the first structured vector, relevant memories (i.e., associated memories) are quickly located in the hierarchical memory bank.

[0229] Fourth layer: Conflict resolution layer Detect conflicts between user commands and associated memories, and execute the corresponding conflict resolution strategies (conflict coverage strategy, conflict fusion strategy, and conflict marking strategy). Fifth layer: Memory management layer The processed memories are stored in a hierarchical manner, with weights updated and time-sensitive deletion performed. The core innovation of this architecture lies in the deep integration between the "memory management layer" and the "conflict handling layer". The result of conflict handling (i.e., the processed memory) will directly trigger the adjustment of memory weights - for example, if a user confirms a new address multiple times, the weight of that memory will automatically increase, and its retrieval priority will be higher than that of the old address; and the change in memory weight will in turn affect the conflict handling strategy - high-weight memories will be retained first in the event of a conflict.

[0230] 2. Three-layer collision detection algorithm This algorithm is a lightweight collision detection solution designed for in-vehicle scenarios. Its core logic is "layered detection and progressive priority"—detecting three types of collisions in order of difficulty, which ensures accuracy while controlling the amount of computation.

[0231] Semantic conflict detection employs a logic of "lightweight semantic encoding + similarity threshold filtering + polarity judgment"—first, DistilBERT (with only 40% of the parameters of BERT, resulting in a 60% improvement in inference speed) is used to encode new instructions and associated memories into semantic vectors, and cosine similarity is calculated; when the similarity is below a preset threshold and the core meanings contradict each other, a semantic conflict is determined. To further reduce inference time, the system pre-encodes and caches memories, eliminating the need for real-time computation.

[0232] Logical conflict detection: Based on a vehicle-specific rule base, a forward + reverse matching mechanism is used—the rule base covers core scenarios such as navigation, music, and communication (e.g., "navigation commands cannot point to two destinations simultaneously" and "music styles cannot be both classical and rock"). Forward matching identifies target entities, while reverse matching verifies the correlation between entities. This dual verification reduces the false positive rate of logical conflicts to below 3%. The rule base supports OTA online updates, allowing for rapid adaptation to new scenarios (e.g., "charging station preferences").

[0233] Temporal conflict detection: A mechanism of "validity period verification + contextual temporal association" is adopted. First, the explicit validity period of the memory (such as "this week's traffic restriction rules") or implicit validity period (such as "temporary customer address" with a default validity period of 7 days) is checked. If the time range of the new instruction does not overlap with that of the associated memory but the content conflicts (such as "last year's company address" vs. "this year's company address"), it is determined to be a temporal conflict. For memories without a clear validity period, the system will infer the temporal relationship through context (such as "the temporary address mentioned last week" vs. "the new address mentioned this week").

[0234] The conflict resolution strategy employs a "priority-based" logic, with priorities decreasing as follows: user-confirmed information > most recently entered information > high-confidence information > time-constrained information. Specific strategies fall into three categories: Conflict overriding strategy: Use user commands to replace associated memories, suitable for explicit conflicts such as address changes and contact information updates; Conflict Fusion Strategy: When user commands and associated memories contain compatible conflicts, an association is established between the user command and associated memory, and a first flag is set. The first flag indicates that the user command and associated memory belong to the same group of preferences, and both user commands and associated memories are displayed simultaneously during recommendations. The conflict fusion strategy is applicable to implicit conflicts such as changes in dietary preferences and expansion of music styles—for example, if a user first likes Jay Chou and then likes Li Jian, the system will mark both as "same group preferences" and cover them simultaneously during recommendations.

[0235] Conflict marking strategy: When user commands and associated memories involve mutual exclusion conflicts, a second flag is set for the mutual exclusion conflict; the second flag is used to trigger a prompt pop-up to obtain user feedback on the mutual exclusion conflict. The conflict marking strategy is suitable for conflicts of high-value information (such as "home address" having two versions).

[0236] 3. Four-layer memory hierarchy and transfer algorithm This algorithm divides memories into four levels, each with a clearly defined storage medium, lifecycle, and circulation rules. The core principle is "dynamic circulation and value priority"—making high-value memories easier to retrieve and automatically eliminating low-value memories. The algorithm's innovation lies in its "memory value scoring" mechanism (a weighted calculation based on four dimensions: usage frequency, importance, timeliness, and user preference, with weights dynamically adjusted by the algorithm). The system calculates the memory value score for each memory in real time based on user behavior. Memories with high value scores are prioritized for storage in high-speed media and are also ranked higher during retrieval. For example, a user's daily access to their "home address" will continuously increase its memory value score, eventually upgrading from short-term memory to permanent memory.

[0237] 4. Reinforcement Learning-Driven Dynamic Adjustment Algorithm The core of this algorithm is to "evolve the memory system according to user behavior"—dynamically adjusting memory weights, retrieval priorities, and conflict handling strategies through reinforcement learning, without the need for manual intervention.

[0238] The state input layer includes three core features: user behavior sequences (such as the last 10 music playback records and navigation address change records), memory attributes (such as memory type (long-term memory, short-term memory, and instantaneous memory), retrieval frequency, and storage duration), and driving scenarios (such as highway / city, congested / smooth traffic, day / night). For example, if a user is more inclined to listen to rock music on the highway, the system will record this scenario feature and adjust the memory weights accordingly.

[0239] Action output layer: includes three types of executable actions: memory weight adjustment (such as increasing the weight of the new address and decreasing the weight of the old address), memory hierarchy change (such as upgrading "frequently visited coffee shop" from short-term memory to long-term memory), and retrieval priority optimization (such as setting the retrieval priority of "home address" to the highest).

[0240] Reward function layer: The reward function is the core of the algorithm. It needs to balance the retrieval hit rate, user satisfaction, and resource utilization, while expanding the user satisfaction dimension and adapting to the resource limitations of the cockpit. For details of the reward function, please refer to the aforementioned formula (1).

[0241] In the reward function, the retrieval hit rate refers to the proportion of the number of times a user needs to be correctly recalled out of the total number of searches. Positive rewards are higher than negative rewards to encourage accurate retrieval and avoid overly conservative strategies due to excessive penalties.

[0242] User satisfaction includes explicit feedback data and implicit feedback data. Explicit feedback data is determined based on users' direct evaluation information. Explicit feedback data includes at least user voice evaluation data and facial expression data; for example, users directly express "yes" or "no", as well as "smile" or "frown".

[0243] Implicit feedback data is determined based on indirect evaluation information extracted from user actions. Implicit feedback data includes at least the duration of user interaction on the interface and actions taken to modify the system's output. For example, the duration of user interaction or whether search results were manually modified.

[0244] Resource utilization rate refers to the ratio of endpoint storage space occupied by the memory bank to computing resources.

[0245] The system employs the Proximal Policy Optimization (PPO) algorithm to train its memory management strategy. This allows for rapid response to changes in user behavior while maintaining policy stability. For example, if a user selects a new address three times consecutively, they receive a positive reward for "increasing the weight of the new address," and the system will prioritize retaining the new address in future similar conflicts.

[0246] 5. Semantic-aware Coverage Merging Algorithm The core of this algorithm is "preserving value and eliminating redundancy"—for the overlapping parts of user instructions and associated memories, a differentiated merging strategy is adopted according to the memory type to ensure the accuracy of memory while avoiding redundant storage.

[0247] Navigation-related memory: The system adopts a coverage strategy of "latest timestamp + usage frequency" - for example, if a user said "the company is in the technology sector" last week and "the company has moved to Quanyan" this week, the system will first check the usage frequency of the two addresses: if the new address has been used ≥2 times and the old address has no usage record at all, the old address will be completely overwritten; if the old address still has usage records (such as once a week), the old address will be marked as an "alternative address", and it will still be displayed during retrieval but with a lower priority than the new address.

[0248] Preference-based memory: A "high-confidence fusion" strategy is adopted. For example, if a user likes Jay Chou first and then Li Jian, the system will add the preference weights of the two (e.g., Jay Chou 0.6 + Li Jian 0.4) to generate a comprehensive preference tag, which will be stored in association. When recommending music, music related to both will be covered at the same time.

[0249] Contact memory: Employs a "multi-source cross-validation" strategy—for example, if a user says "Zhang San's phone number is 123456," the system will first cross-validate it with the address book and call records; if it does not exist, it will add it; if it exists but is inconsistent, it will be marked as "pending confirmation," thus avoiding incorrect information from overwriting correct information.

[0250] 6. Implementation details of key modules 6.1 User Status Recognition Module User state recognition is a prerequisite for memory management—only by accurately recognizing the user's state can the corresponding memory bank and processing strategy be invoked. This module adopts a multimodal fusion model.

[0251] User identity data recognition: Identity verification based on FACEID or voiceprint recognition from the in-vehicle DMS camera; the system supports multi-user memory isolation, with each user having an independent memory bank, so that memory confusion will not occur even if multiple people share the vehicle.

[0252] User emotion data recognition: This involves fusing voice tone features (such as speech rate and pitch) with facial expression features (such as frowning and smiling). For example, when a user is feeling down, the system will prioritize recalling "soothing music" rather than "rock music".

[0253] Scene data recognition: Recognition is achieved through data such as vehicle speed, road conditions (e.g., congestion / smooth traffic), time (e.g., day / night), and location (e.g., highway / city). For example, when a user is on the highway and the system makes food-related recommendations, it will prioritize recalling the memory of the "nearest service area" rather than the "nearby restaurants".

[0254] 6.2 Memory Storage Module The memory storage module adopts a hybrid storage architecture to balance storage capacity and retrieval speed, and the storage footprint on the edge is reduced compared to traditional solutions.

[0255] Storage media selection: High-scoring and frequently used data: uses edge-side high-speed cache (such as LPDDR5) for storage, with read / write latency ≤1ms, suitable for temporary data in single interactions; High-rated and infrequently used: Stored using an edge-side in-memory database (such as Redis), with read / write latency ≤5ms, suitable for memory used frequently in the near future; For low-rated and infrequently used data: use edge-side external storage (such as UFS 3.1), which has a large storage capacity and low cost, and is suitable for low-frequency but high-value memory.

[0256] Data structure design: The memory data is stored in JSON format, containing key fields such as memory ID, user ID, content, timestamp, type, level, usage frequency, importance, and confidence. The "importance" field is dynamically assigned a value by the reinforcement learning model, ranging from 0 to 1. The higher the value, the higher the retrieval priority.

[0257] Index optimization: Establish a multi-level index structure including primary key index, time index, type index, and user index to support fast multi-dimensional queries.

[0258] 6.3 Timeliness Management Module The timeliness management module adopts a mechanism of "timed evaluation + real-time triggering" to automatically clean up low-value memories and release end-side resources.

[0259] Regular assessment: Every day at dawn (when the vehicle is idle and resources are available), all memories are scored for their memory value. For the specific process of memory value scoring, please refer to the aforementioned formula (2).

[0260] Memories with scores below a set threshold will be added to a cleanup queue and deleted in batches to avoid consuming too many computing resources.

[0261] Real-time trigger: When the memory bank occupies ≥80% of the end-side storage, the system will immediately trigger emergency cleanup, prioritizing the deletion of the lowest-rated temporary memory and short-term memory to ensure smooth system operation.

[0262] Elimination strategy: The system adopts the principle of "phased elimination", deleting no more than 10% of the total memory at a time to avoid system abnormalities caused by deleting too many memories at once; at the same time, the system will retain the most recent 1,000 high-frequency memories as a basic cache, so that basic interactive functions can still be maintained even after cleanup.

[0263] The purpose of this disclosure is to provide a method and apparatus for handling memory conflicts, so as to solve the technical bottlenecks in response speed, conflict handling, memory hierarchical classification management and timeliness management.

[0264] Specifically, this disclosure can achieve the following technical objectives: Rapid response capability: Through optimized algorithm architecture and efficient retrieval mechanism, the processing of each memory is completed within tens of milliseconds, and the recall of related memories is controlled within tens to two hundred milliseconds, meeting the strict real-time requirements of the vehicle system.

[0265] Intelligent conflict resolution mechanism: A comprehensive conflict detection and resolution mechanism is built, which can quickly identify and process conflicts in associated memories based on the current user status or the addition of new memories, ensuring the consistency and accuracy of memory information.

[0266] Hierarchical and Categorized Memory Management: Establish a scientific hierarchical and categorized memory system to achieve differentiated management of different types of memory, thereby improving memory utilization efficiency and system performance.

[0267] Dynamic Adjustment and Overlay Merging: A flexible dynamic adjustment algorithm is designed to adaptively adjust the weight and priority of memory information based on changes in user behavior patterns and usage scenarios, and to achieve efficient overlay merging operations.

[0268] Timeliness management optimization: Establish an intelligent elimination mechanism based on time decay and usage frequency, which can remove some information according to timeliness, release system resources, and maintain the efficient operation of the memory bank.

[0269] Based on the above embodiments, this application also provides a memory conflict processing device, please refer to... Figure 5 , Figure 5 A block diagram of a memory conflict processing apparatus 500 is shown. The memory conflict processing apparatus 500 includes a determination module 501, a conflict processing module 502, and an update module 503, wherein: The determination module 501 is used to determine associated memories in the hierarchical memory bank based on user commands, user status data, and driving scenario data; user status data is determined based on user identity data and user emotion data; and semantic conflict detection, logical conflict detection, and temporal conflict detection are performed on user commands and associated memories in sequence to obtain conflict detection results. The conflict resolution module 502 is used to process the conflict between the user command and the associated memory based on the conflict resolution strategy when the conflict detection result indicates that there is a conflict between the user command and the associated memory, so as to obtain the processed memory. The conflict resolution strategy includes a conflict coverage strategy, a conflict fusion strategy and a conflict marking strategy. Update module 503 is used to update the processed and memorized weight parameters based on a dynamic adjustment algorithm.

[0270] In some embodiments, the determining module 501 is further configured to acquire user instructions, user identity data, user emotion data, and driving scenario data; identify the user identity data and user emotion data to obtain identified user state data; convert the user instructions and driving scenario data to obtain a first structured vector; and determine associated memories in the hierarchical memory bank based on the identified user state data and the first structured vector.

[0271] In some embodiments, the updating module 503 is further configured to use the output value of the reward function as input to the dynamic adjustment algorithm to process the processed memory and obtain the updated weight parameters of the processed memory; wherein, the reward function is determined based on the retrieval hit rate, user satisfaction, and resource utilization rate; the input data of the dynamic adjustment algorithm further includes: user behavior sequence, memory attributes of the processed memory, and driving scenario data; the output data of the dynamic adjustment algorithm further includes: memory level of the processed memory and retrieval priority of the processed memory; the user behavior sequence includes multiple behaviors related to the processed memory; the memory attributes include at least memory type, retrieval frequency, and storage duration; the driving scenario includes at least road type, road condition, and driving time.

[0272] In some embodiments, the determining module 501 is further configured to perform semantic conflict detection on the user instruction and associated memory to obtain a semantic conflict detection result; perform logical conflict detection on the user instruction and associated memory to obtain a logical conflict detection result; and perform temporal conflict detection on the user instruction and associated memory to obtain a temporal conflict detection result; wherein the conflict detection result includes a semantic conflict detection result, a logical conflict detection result, and a temporal conflict detection result.

[0273] In some embodiments, the determining module 501 is further configured to encode the user instruction and the associated memory respectively to obtain the user instruction encoding vector and the associated memory encoding vector; determine the cosine similarity between the user instruction encoding vector and the associated memory encoding vector; and determine that the semantic conflict detection result is that there is a semantic conflict between the user instruction and the associated memory when the cosine similarity is less than a preset threshold and the semantic recognition results of the user instruction and the associated memory are different.

[0274] In some embodiments, the determining module 501 is further configured to perform forward and reverse matching on user instructions and associated memories based on the vehicle rule base to obtain matching results; and determine whether there is a logical conflict between user instructions and associated memories based on the matching results to obtain logical conflict detection results.

[0275] In some embodiments, the determining module 501 is further configured to perform validity period verification on the user instruction and associated memory, determine whether there is a timing conflict between the user instruction and associated memory, and obtain a timing conflict detection result.

[0276] In some embodiments, the updating module 503 is further configured to: store the processed memory in the high-speed storage of the edge device when the memory value score of the processed memory is greater than the first data and the usage frequency is greater than the second data; store the processed memory in the memory database of the edge device when the memory value score of the processed memory is greater than the first data and the usage frequency is less than the second data; and store the processed memory in the external storage of the edge device when the memory value score of the processed memory is less than the first data and the usage frequency is less than the second data.

[0277] In some embodiments, the update module 503 is further configured to perform time-sensitive deletion on the processed memory based on the timestamp of the processed memory, thereby obtaining an updated hierarchical memory bank.

[0278] The descriptions of the above method embodiments are similar to those of the above device embodiments, and have similar beneficial effects. For technical details not disclosed in the device embodiments of this disclosure, please refer to the descriptions of the method embodiments of this disclosure for understanding.

[0279] It should be noted that, in the embodiments of this disclosure, if the above-described memory conflict handling method is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this disclosure, or the part that contributes to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several voice commands to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this disclosure are not limited to any specific hardware and software combination.

[0280] Please refer to Figure 6 , Figure 6 A schematic diagram of a vehicle's hardware structure is shown, including a processor, memory, and a communication bus. A memory conflict handling device may include a communication interface 601, a memory 602, and a processor 603; the various components are coupled together via a bus system 604. It is understood that the bus system 604 is used to implement communication between these components. In addition to a data bus, the bus system 604 also includes a power bus, a control bus, and a status signal bus. However, for clarity, [the following is omitted as it is not part of the schematic diagram]. Figure 6 The general designated all buses as Bus System 604.

[0281] In this embodiment of the disclosure, the communication interface 601 is used for receiving and sending signals during the process of sending and receiving information with other external devices; the memory 602 is used for storing computer programs that can run on the processor 603; the processor 603 is used to execute the steps of any of the above-mentioned parking-integrated assisted driving methods when running the computer program.

[0282] It is understood that the memory 602 in this embodiment of the present disclosure may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be read-only memory (Read-Only Memory). The volatile memory can be a ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or flash memory. It can also be random access memory (RAM) used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM). The memory 602 of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0283] The processor 603 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 603 or by instructions in software form. The processor 603 can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this disclosure can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 602, and the processor 603 reads the information in memory 602 and, in conjunction with its hardware, completes the steps of the above method.

[0284] It is also understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented on one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), or field-programmable gate arrays (FPGAs). Programmable Gate Array (FPGA), general-purpose processor, controller, microcontroller, microprocessor, other electronic unit or combination thereof for performing the functions of this disclosure.

[0285] For software implementation, the techniques described herein can be implemented through modules (e.g., procedures, functions, etc.) that perform the functions described herein. Software code can be stored in memory and executed by a processor. Memory can be implemented in the processor or externally. Wherein, if implemented as a software functional module and not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this disclosure, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of this embodiment. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0286] In yet another embodiment of this disclosure, a storage medium, namely a computer-readable storage medium, is provided, on which a computer program or voice instruction is stored, which, when executed by a processor, implements the steps of any of the driving assistance methods integrating driving and parking provided in the first aspect of the above embodiments.

[0287] In yet another embodiment of this disclosure, a computer program product is provided, which includes a computer program or voice instructions. When the computer program or voice instructions are executed by a processor, they implement the steps in any of the driving assistance methods integrating driving and parking provided in the first aspect of the above embodiments.

[0288] It should be noted that the descriptions of the above embodiments of storage media, devices, apparatuses, and program products are similar to the descriptions of the above method embodiments and have similar beneficial effects. For technical details not disclosed in the embodiments of storage media, devices, apparatuses, and program products of this disclosure, please refer to the descriptions of the method embodiments of this disclosure for understanding.

[0289] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this disclosure. Therefore, "in one embodiment" or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this disclosure, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this disclosure. The sequence numbers of the above-described embodiments are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0290] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0291] In the several embodiments provided in this disclosure, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components may be combined, or integrated into another system, or some features may be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0292] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0293] In addition, each functional unit in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0294] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program voice commands. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0295] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this disclosure, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several voice commands to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0296] The above are merely embodiments of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure.

[0297] It should be understood that if this disclosure references any user data and personal information (including but not limited to device information, behavioral data, location information, etc.) and before applying the technical solutions described in the embodiments of this disclosure, the relevant products or services should comply with the laws and regulations concerning the protection of user data and personal information, strictly process users' personal information and data in accordance with the provisions of applicable laws and regulations throughout the entire data processing lifecycle, follow the principles of legality, legitimacy, necessity, good faith, openness, and transparency, and adopt reasonable privacy design schemes and technical measures to ensure the security of user data and personal information, protect users' legitimate rights and interests, and prevent the risks of leakage, theft, or tampering of user data and personal information.

[0298] Specifically, the company must publish and display its privacy policy in a prominent position on the user interface, clearly informing users of the types, purposes, uses, and methods of processing personal information, as well as other matters that should be disclosed as required by laws and regulations; obtain users' prior informed consent or explicit authorization regarding data processing through user-initiated interaction (such as confirmation pop-ups); process or store user data securely within the legally required timeframe; adopt a series of security technologies and management measures, including but not limited to data encryption and access control; share and transfer user data within the scope permitted by law and in a legally required manner; and process user rights, including the rights to query, access, correct, delete, withdraw authorization and consent, cancel registration, and obtain copies of personal information, within the legally required timeframe.

Claims

1. A method for handling memory conflicts, characterized in that, The method includes: Based on user commands, user status data, and driving scenario data, associated memories are determined in a hierarchical memory bank; the user status data is determined based on user identity data and user emotion data. Semantic conflict detection, logical conflict detection, and temporal conflict detection are performed sequentially on the user instruction and the associated memory to obtain the conflict detection results; If the conflict detection result indicates that there is a conflict between the user instruction and the associated memory, the conflict is resolved based on the conflict resolution strategy to obtain the processed memory; the conflict resolution strategy includes a conflict coverage strategy, a conflict fusion strategy, and a conflict marking strategy. The weight parameters of the processed memory are updated based on a dynamic adjustment algorithm.

2. The method according to claim 1, characterized in that, The process of determining associated memories in a hierarchical memory bank based on user commands, user status data, and driving scenario data includes: Obtain the user instructions, the user identity data, the user emotion data, and the driving scenario data; The user identity data and the user emotion data are identified to obtain the identified user status data; The user instructions and the driving scenario data are converted to obtain a first structured vector; Based on the identified user state data and the first structured vector, the associated memory is determined in the hierarchical memory bank.

3. The method according to claim 1, characterized in that, The step of updating the processed and memorized weight parameters based on the dynamic adjustment algorithm includes: The output value of the reward function is used as the input of the dynamic adjustment algorithm to process the post-processed memory, thereby obtaining the updated weight parameters of the post-processed memory. The reward function is determined based on the retrieval hit rate, user satisfaction, and resource utilization rate. The input data of the dynamic adjustment algorithm also includes: user behavior sequence, memory attributes of the processed memory, and driving scenario data; the output data of the dynamic adjustment algorithm also includes: memory level of the processed memory and retrieval priority of the processed memory; the user behavior sequence includes multiple behaviors related to the processed memory; the memory attributes include at least memory type, retrieval frequency, and storage duration; the driving scenario includes at least road type, road condition, and driving time.

4. The method according to claim 1, characterized in that, The step of sequentially performing semantic conflict detection, logical conflict detection, and temporal conflict detection on the user instruction and the associated memory to obtain the conflict detection result includes: The semantic conflict detection is performed on the user instruction and the associated memory to obtain the semantic conflict detection result; The logical conflict detection is performed on the user instruction and the associated memory to obtain the logical conflict detection result; The timing conflict detection is performed on the user instruction and the associated memory to obtain the timing conflict detection result; The conflict detection results include the semantic conflict detection results, the logical conflict detection results, and the temporal conflict detection results.

5. The method according to claim 4, characterized in that, The step of performing semantic conflict detection on the user instruction and the associated memory to obtain a semantic conflict detection result includes: The user instruction and the associated memory are encoded respectively to obtain the user instruction encoding vector and the associated memory encoding vector; Determine the cosine similarity between the user instruction encoding vector and the associated memory encoding vector; When the cosine similarity is less than a preset threshold and the semantic recognition results of the user instruction and the associated memory are different, the semantic conflict detection result is determined to be that there is a semantic conflict between the user instruction and the associated memory.

6. The method according to claim 4, characterized in that, The step of performing logical conflict detection on the user instruction and the associated memory to obtain logical conflict detection results includes: Based on the vehicle rule base, forward and reverse matching are performed on the user commands and the associated memories to obtain matching results; Based on the matching results, it is determined whether there is a logical conflict between the user instruction and the associated memory, and the logical conflict detection result is obtained.

7. The method according to claim 4, characterized in that, The step of performing the timing conflict detection on the user instruction and the associated memory to obtain the timing conflict detection result includes: The validity period of the user instruction and the associated memory is checked to determine whether there is a time conflict between the user instruction and the associated memory, and the time conflict detection result is obtained.

8. The method according to any one of claims 1-7, characterized in that, The method further includes: If the memory value score of the processed memory is greater than the first data and the usage frequency is greater than the second data, the processed memory will be stored in the high-speed storage of the edge device. If the memory value score of the processed memory is greater than the first data and the usage frequency is less than the second data, the processed memory is stored in the memory database of the edge device. If the memory value score of the processed memory is less than the first data and the usage frequency is less than the second data, the processed memory will be stored in the external storage of the edge device.

9. The method according to any one of claims 1-7, characterized in that, The method further includes: Based on the timestamp of the processed memory, the processed memory is deleted due to its expiration date, resulting in an updated hierarchical memory bank.

10. A memory conflict processing device, characterized in that, The device includes: The determination module is used to determine associated memories in a hierarchical memory bank based on user instructions, user status data, and driving scenario data; the user status data is determined based on user identity data and user emotion data; and the module performs semantic conflict detection, logical conflict detection, and temporal conflict detection on the user instructions and the associated memories in sequence to obtain the conflict detection results. A conflict resolution module is used to perform conflict resolution on the user instruction and the associated memory based on a conflict resolution strategy when the conflict detection result indicates that there is a conflict between the user instruction and the associated memory, thereby obtaining the processed memory; the conflict resolution strategy includes a conflict coverage strategy, a conflict fusion strategy, and a conflict marking strategy; An update module is used to update the processed and memorized weight parameters based on a dynamic adjustment algorithm.

11. A vehicle, characterized in that, The vehicle includes: a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the memory conflict handling method as described in any one of claims 1 to 9.

12. A storage medium, characterized in that, The storage medium stores one or more computer programs, which can be executed by one or more processors to implement the memory conflict handling method as described in any one of claims 1 to 9.

Images