Labyrinth robot brain cognitive system based on improved emotion and memory mechanism

Abstract

The application discloses a kind of based on improved emotion and memory mechanism's maze robot brain cognitive system, belong to mobile robot technical field, the system includes perception module, emotion module, behavior decision module, memory module, learning module and action selection module.Emotion module is constructed in the "valence-waking degree" dynamic emotion mechanism, and emotion is expressed by two orthogonal dimensions of valence and waking degree;In the memory module, an exploration memory network algorithm is designed to enable the robot to explicitly remember and efficiently utilize the environment topology, and a short-term memory mechanism for multiple energy points is proposed, enabling the robot to flexibly select the nearest energy node based on its current location.The application significantly improves the exploration efficiency and autonomous decision-making ability of the robot in a complex multi-energy point maze environment.

Description

Technical Field

[0001] This invention relates to the field of mobile robot technology, and in particular to a brain-like cognitive system for maze robots based on improved emotion and memory mechanisms. Background Technology

[0002] From the perspective of life maintenance and danger avoidance, the fundamental reason why humans and animals can continuously improve their intelligence throughout their life cycle lies in their ability to ensure their own survival. For autonomously moving robots, when performing path recognition and autonomous planning in real-world scenarios, fully considering survival-related capabilities such as energy replenishment and hazard avoidance to achieve self-survival assurance will be of great significance for them to complete complex tasks, accumulate lifelong experience, and even achieve intelligent evolution. From a biomimetic perspective, survival is an intrinsic need of intelligent agents, involving the operation of the brain's intrinsic motivation. Therefore, it is necessary to learn from the working mechanisms of the brain, combining corresponding cognitive units and high-level cognitive functions, to design an advanced brain-like cognitive model for robots oriented towards survival assurance, thereby supporting them to achieve human-like embodied survival and learning, and further improve their intelligence level. Since the proposal of brain-like cognitive models, there have been numerous research results on the artificialization of learning and memory mechanisms. However, these models generally lack subjective experience and intrinsic motivation, making it difficult to achieve autonomous adaptation in open environments. As a biological organism's direct experience or reflection of its own "survival state," emotions have been shown in related studies to not only serve as intrinsic motivation to drive the exploratory behavior of intelligent agents, but also significantly improve the learning efficiency of robots under sparse reward conditions by reshaping rewards and dynamically adjusting learning parameters. Therefore, they are often used in the design of high-level cognitive mechanisms in brain-like cognitive models.

[0003] The applicant's patent application filed on October 13, 2025 (patent number: 202511456750.4) discloses a brain-like cognitive system for maze robots based on emotion and memory mechanisms, enabling the robot to avoid danger and ensure survival in single-energy-point maze environments. However, the decision-making required in real-world environments is far more complex. The robot must not only determine when to return to replenish energy but also the optimal path for doing so. Furthermore, relying solely on randomness in exploring unknown environments would significantly reduce search efficiency.

[0004] Against this backdrop, this invention designs a brain-like cognitive system for maze robots based on improved emotion and memory mechanisms, specifically for more complex multi-energy point maze environments. This system further enhances the robot's learning ability and autonomous decision-making capabilities. Summary of the Invention

[0005] To enable robots to perform complex intelligent behaviors in embodied survival environments, this invention, based on the existing published patent for a brain-like cognitive system for maze robots based on emotion and memory mechanisms (patent number: 202511456750.4), further proposes a brain-like cognitive system for maze robots based on an improved emotion and memory mechanism, comprising:

[0006] Perception module: Acquires the robot's environmental state;

[0007] Emotional module: Receives the environmental state from the perception module, constructs a dynamic emotion mechanism of "valence-arousal", and expresses emotions through two orthogonal dimensions of valence and arousal;

[0008] Behavioral decision module: Judges behavior based on the arousal level in the emotion module to obtain behavioral decision results;

[0009] Memory module: Stores and utilizes knowledge during the robot's learning process, including short-term memory units, working memory units, long-term memory units, exploratory memory network units, and emotional memory units;

[0010] Learning module: Used to help the robot learn and understand the unknown environment. It updates the target orientation table and the energy orientation table by exploring and learning the target nodes and energy nodes.

[0011] Action selection module: Based on the combined influence of the behavior decision module, the learning module, and the memory module on the action selection module, output the next action to be executed.

[0012] The perception module, the emotion module, the behavior decision module, the memory module, the learning module, and the action selection module are interconnected.

[0013] Optionally, the emotion module includes a valence dimension calculation unit and an arousal dimension calculation unit, wherein:

[0014] The valence dimension calculation unit is used for calculating the valence dimension, specifically:

[0015] The formula for calculating the design valence dimension is shown below:

[0016]

[0017] In the formula, valence As one of the dimensions of emotional state, it is used to characterize the robot's positive or negative evaluation of environmental stimuli; The initial valence is set to +1.0 or -1.0. The emotion mechanism activates the valence based on the objects encountered in the environment. Specific triggering conditions include: when the robot reaches the energy node, the valence is set to the maximum positive value. This indicates a positive emotional experience; when the robot encounters a trap node, the valence is set to the maximum negative value. This indicates a negative emotional experience; This represents the number of steps since the event was triggered. This is the valence decay coefficient;

[0018] The wakefulness dimension calculation unit is used for calculating the wakefulness dimension, specifically:

[0019] Arousal level is composed of three independent sub-components, which are weighted and summed to obtain the final arousal value, as shown in the following formula:

[0020]

[0021] In the formula, wakefulness As an activation dimension of emotional state, high arousal means that the robot is in a highly alert state, while low arousal corresponds to an energy-saving standby state. Indicates the urgency of energy use; Indicates the threat level of the trap; Indicates the return distance pressure; These are the respective weighting coefficients.

[0022] Optionally, the arousal level calculation unit includes an energy urgency calculation unit, a trap threat calculation unit, and a return distance pressure calculation unit, wherein,

[0023] The energy urgency calculation unit is used to calculate the energy urgency, and the specific formula is as follows:

[0024]

[0025] In the formula, energy urgency Used to quantify the impact of a robot's current energy reserves on its emotions; This is the energy sensitivity coefficient, used to adjust the steepness of the response curve; It is the hyperbolic tangent function; This represents the robot's remaining energy value. The safe energy threshold for a robot is the minimum energy value required for the robot to safely return to the energy point. When, that is, when energy is sufficient, A value close to -1 indicates low wake-up time; when When, that is, when energy is insufficient, A value close to +1 indicates high wakefulness;

[0026] The trap threat level calculation unit is used to calculate the trap threat level, and the specific formula is as follows:

[0027]

[0028] In the formula, the trap threat level This describes the impact of fear memories after a robot encounters a dangerous area on its current emotions. This part is activated when the robot passes through a trap node and gradually decays as the distance from the trap node increases. Represents the trap type coefficient; This represents the number of steps from the current node to the dangerous node. The threat attenuation coefficient controls the rate at which fear memories fade with distance;

[0029] The return distance pressure calculation unit is used to calculate the return distance pressure, and the specific formula is as follows:

[0030]

[0031] In the formula, the return distance pressure As the third arousal component, it directly reflects the impact of spatial distance pressure between the robot's current position and the nearest energy node on emotions; This indicates the number of steps required for the robot to return to the energy node, and the necessary information is obtained from short-term memory. This is a preset maximum return step threshold; the formula uses a linear mapping, and when the robot is close to the energy node... hour, When the robot moves away from the energy node ( )hour, .

[0032] Optionally, the behavior decision module is used to directly control the robot's switching between two behavior modes based on the arousal level, including a low arousal mode unit and a high arousal mode unit, specifically:

[0033] Low wake-up mode unit: used to implement when When the robot is in a low wake-up mode, it performs exploratory behavior, i.e. behavior=0, and selects an unvisited direction or the optimal action through a memory mechanism.

[0034] High wake-up mode unit: used to implement when When the robot enters a high arousal mode, the increased emotional arousal enhances its behavioral initiative, triggering a forced return strategy for energy replenishment, i.e., behavior=1.

[0035] Optionally, the short-term memory unit adopts an improved short-term memory mechanism, specifically a short-term memory mechanism oriented towards multiple energy points;

[0036] The working memory unit is used to record memory information originally intended for the exploration target node task;

[0037] The long-term memory unit is used to store the learned environmental information;

[0038] The exploration memory network unit is used to record memory information for environmental exploration tasks. This memory consists of part short-term memory and part long-term memory, and specifically adopts the exploration memory network algorithm.

[0039] The emotional memory unit is used to record the memory information generated by the emotional module, specifically employing an emotional memory formation mechanism.

[0040] Optionally, the memory module includes short-term memory units, working memory units, long-term memory units, exploratory memory network units, and emotional memory units; wherein,

[0041] The exploratory memory network unit is composed of an adjacency graph and boundary nodes. The exploratory memory network algorithm specifically includes:

[0042] The long-term memory unit Improved to triplet ,in, Guided by the final learned target node Network table, Guided by the energy nodes to be learned in the end The network table incorporates an adjacency graph into the long-term memory unit. The topology information used to store the environment is mathematically represented as:

[0043]

[0044] In the formula, This represents the set of nodes that have been explored. Let represent the set of directed edges. , Represents the set of directions. The direction is east. Represents the direction west. Represents the direction north. The direction is south, and each directed edge is defined as a triple. ,in , indicating from node Along direction reachable nodes ;

[0045] The bidirectional consistency of an adjacency graph is as follows:

[0046]

[0047] When recording a robot movement, the algorithm first checks if the starting node already exists in the adjacency graph; if not, it initializes the forward edge. Next, perform the same checks and initialization operations on the arriving nodes, while automatically adding reverse edges. In addition, general memories from environmental exploration Define a set of boundary nodes, where each boundary node is a visited node with at least one unexplored direction. The boundary update rule is the same as... Consistent;

[0048] The improved short-term memory mechanism in the short-term memory unit is specifically a short-term memory mechanism for multiple energy points. This mechanism maintains an independent set of return path sequences for each discovered energy node and dynamically selects the optimal return target by comparing the lengths of each sequence in real time.

[0049] The emotional memory formation mechanism in the emotional memory unit is as follows:

[0050] Changes in the valence dimension provide a coding basis for emotional memory—positive valence reinforces preference for corresponding paths, while negative valence fosters avoidance tendencies. Specifically, when a robot restarts its exploration from energy nodes or trap nodes, it will target energy node-oriented... Network table Each node it passes through will trigger the following Value update:

[0051]

[0052] As robots move away from energy nodes, valence Gradually decays, while moving in the opposite direction along the path. The process involves tracing back emotional rewards at each level, thereby generating emotional memories; the same applies to trap nodes.

[0053] Optionally, the function of the exploratory memory network algorithm is to search for the nearest boundary node from the current node along a specified direction. This exploratory memory network algorithm adopts a breadth-first search strategy, and the specific execution process is divided into three stages:

[0054] ① Validity verification phase: The algorithm first checks whether the starting node already exists in the adjacency graph, and whether there is a known connection in the specified initial direction;

[0055] ② Initial Advancement Phase: After passing the validity verification, the algorithm moves one step along the initial direction and reaches the first neighbor node; at this time, the algorithm immediately checks whether the node is a boundary node. If it is, there is no need to start the complete BFS process.

[0056] ③ Breadth-first search phase: The algorithm enters the BFS main loop and takes nodes out of the queue one by one for expansion.

[0057] Optionally, the improved short-term memory mechanism maintains an independent set of return path sequences for each discovered energy node, and dynamically selects the optimal return target by comparing the lengths of each sequence in real time, as follows:

[0058] 1) Let the set of energy nodes be... During the exploration process, the robot gradually discovers and records these energy nodes, forming a list of discovered energy nodes. Initially an empty set, whenever the robot first reaches a certain energy node... Add it to ;

[0059] To support parallel memory across multiple energy nodes, the system provides... Each energy node in Maintain an independent set of node information sequences:

[0060]

[0061] in It still represents node information, including status. With action , Energy node The node information; the sequences of each energy node are independent of each other and do not interfere with each other;

[0062] 2) Each time the robot performs a move, that is, "reaches a new node"... Execute actions "The system" The sequences of all energy nodes are updated synchronously.

[0063] 3) Define the sequence length To return to the energy node from the current location The required number of steps; when the behavioral decision-making mechanism determines that it needs to return to the energy node, the system starts from... Choose the energy node with the shortest return path as the target:

[0064]

[0065] That is, select the energy node with the smallest current sequence length. As the return target;

[0066] 4) Determine the target energy node Subsequently, the reflection path generation process is consistent with the single-energy-point case, based on the sequence. Execution; ultimately generating the path from the current point to the energy node. The shortest return path.

[0067] Optionally, the robot moves in one step (to reach a new node). Execute actions ), the system for The sequence of all energy nodes in the middle is updated synchronously; for each The following two situations will be handled in detail:

[0068] ①If Not appeared In the set of states, that is Then, the new node will be appended. Append to the end of the sequence:

[0069]

[0070] ②If The presence of the path in the sequence indicates a loop; at this point, a short-term memory triggering and reshaping process is executed, which involves deleting the first arrival. The node information between the two arrivals is used to add the new node information to the sequence.

[0071] Optionally, with the support of the exploratory memory network algorithm and learning mechanism, the "exploration and utilization" part of the action selection module specifically includes:

[0072] The exploratory memory network algorithm first checks the exploration state; if the boundary set... If the boundary set is not fully explored, then the exploration is considered incomplete. In this mode, the direction selection is entirely dominated by the exploration memory network algorithm; if the boundary set is not fully explored... This indicates that full exploration has been achieved, and the system has switched to exploitation mode: querying long-term memory units for each optional direction. ,choose The direction with the largest value;

[0073] The action selection module includes a return unit and a continue exploration unit, wherein...

[0074] The return unit is used to implement the behavior of returning to the energy node when the forced return strategy for energy replenishment is triggered in the behavior decision mechanism. First, it selects the optimal energy node according to the improved short-term memory mechanism and generates a path from the current point to the energy node. The shortest return path;

[0075] The continued exploration unit is used to integrate the memory information in the exploration memory network unit when the behavior selection mechanism determines that the robot needs to perform an exploration behavior. It adopts a three-level priority mechanism, which is as follows: firstly, it selects the unexplored direction of the node for exploration; if there is no unexplored direction of the node, it selects the direction leading to the boundary node; if neither of the above two priority levels can determine the direction, it randomly selects one from the available directions.

[0076] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0077] This invention provides a brain-like cognitive system for maze robots based on improved emotion and memory mechanisms. Through improvements to previously disclosed inventions, the system enables the robot to more rationally select task priorities based on various emotional states, ensuring timely return to energy nodes to replenish energy and minimizing excessive energy surplus. Simultaneously, the improved emotion and memory modules significantly enhance learning efficiency. The robot can not only find the optimal path back to the nearest energy node but also explore directly from boundary nodes, further improving efficiency.

[0078] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0079] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0080] Figure 1 This is a schematic diagram of the system architecture provided by the present invention;

[0081] Figure 2 The system flow architecture diagram provided by this invention;

[0082] Figure 3 A schematic diagram of the cognitive model framework provided by this invention;

[0083] Figure 4 The maze environment diagram provided by this invention;

[0084] Figure 5 This is a diagram illustrating the robot's first learning process as provided in an embodiment of the present invention.

[0085] Figure 6 The second learning result diagram of the robot provided in the embodiment of the present invention;

[0086] Figure 7The image shows the result of the robot's third learning iteration, as provided in an embodiment of the present invention. Detailed Implementation

[0087] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0088] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0089] like Figures 1 to 7 As shown, this invention provides a brain-like cognitive system for maze robots based on improved emotion and memory mechanisms, comprising the following modules:

[0090] 1. Sensing Module

[0091] It is used to obtain the robot's position in the environment, the target point position, the accessibility of the environment, ordinary nodes, energy nodes and trap nodes, etc., to obtain the environmental state.

[0092] 2. Emotional Module

[0093] The single state of fear is improved into a two-dimensional emotion model composed of "valence-arousal". Arousal directly affects the behavior choice in the behavior decision module, while valence affects the learning module in the form of intrinsic reward. At the same time, the valence dimension generates emotional memory, which changes the information in the memory module and further affects the action selection module.

[0094] 2.1 Valence Dimension Calculation Unit

[0095] Used for inclusion in the valence dimension calculation.

[0096] Specifically, valence, as one of the dimensions of emotional state, is used to characterize the robot's positive or negative evaluation of environmental stimuli. For a robot maze autonomous learning task considering survival, the valence dimension calculation formula is designed as follows:

[0097]

[0098] in, The initial valence (+1.0 or -1.0) is used. The emotion mechanism activates the valence based on the objects encountered in the environment. Specific triggering conditions include: when the robot reaches an energy node, the valence is set to the maximum positive value. = +1.0, representing a positive emotional experience; when the robot encounters a trap node, the valence is set to the maximum negative value. = -1.0, indicating a negative emotional experience; This represents the number of steps since the event was triggered. This is the valence decay coefficient. The valence after triggering does not remain constant, but decays exponentially with the number of steps the robot takes, until it decreases to 0, at which point the emotion tends to be neutral.

[0099] 2.2 Wakefulness Dimension Calculation Unit

[0100] Specifically used for calculating the wakefulness dimension:

[0101] Arousal, as an activation dimension of emotional state, indicates that the robot is in a highly alert state, while low arousal corresponds to an energy-saving standby state. In this task, it reflects the robot's level of agency in the current state, and its value range is normalized to the interval [-1, +1]. In this model, arousal consists of three independent sub-components: energy urgency, etc. Trap Threat Level and return distance pressure , The three factors are weighted coefficients, and the final arousal value is obtained by weighted summation.

[0102]

[0103] Furthermore, the wakefulness dimension calculation unit includes an energy urgency calculation unit, a trap threat calculation unit, and a return distance pressure calculation unit.

[0104] 2.21 Energy Urgency Calculation Unit

[0105] Specifically, for the calculation of the energy urgency:

[0106] Energy urgency is used to quantify the impact of a robot's current energy reserves on its emotions. The formula is as follows:

[0107]

[0108] Among them, parameters This is the energy sensitivity coefficient, used to adjust the steepness of the response curve. Hyperbolic tangent function. The introduction of this ensures that the output value is limited to the range of [-1, +1], while also having smooth nonlinear characteristics; This represents the robot's remaining energy value. This represents the robot's safe energy threshold, specifically the minimum energy value required for the robot to safely return to the energy point. When (energy is sufficient). A value close to -1 indicates low wake-up time; when When (insufficient energy). A value close to +1 indicates high wake-up rate.

[0109] 2.22 Trap Threat Calculation Unit

[0110] Used for calculating trap threat level, specifically:

[0111] Trap threat level describes the impact of fear memories after a robot encounters a dangerous area on its current emotional state. This component is activated as the robot passes a trap node and gradually diminishes with increasing distance from the trap node. The calculation formula is as follows:

[0112]

[0113] in, Represents the trap type coefficient; This represents the number of steps from the current node to the dangerous node. The threat attenuation coefficient controls the rate at which fear memories fade with distance.

[0114] 2.23 Return Distance Pressure Calculation Unit

[0115] The calculation of the return distance pressure is as follows:

[0116] Return distance pressure, as the third arousal component, directly reflects the impact of spatial distance pressure between the robot's current position and the nearest energy node on emotion. The formula is as follows:

[0117]

[0118] in, This indicates the number of steps required for the robot to return to the energy node, and the necessary information is obtained from short-term memory. This is a preset maximum return step threshold; the formula uses a linear mapping, and when the robot is close to the energy node... hour, When the robot moves away from the energy node ( )hour, .

[0119] 3. Behavioral Decision Module

[0120] In the improved behavioral decision-making mechanism, the level of emotional arousal directly regulates the robot's switching between two behavioral modes. The behavioral decision-making module includes a low-arousal mode unit and a high-arousal mode unit.

[0121] 3.1 Low wake-up mode unit

[0122] when When in a low-arousal mode, the robot primarily performs exploratory behaviors (behavior=0), selecting unexplored directions or optimal actions through memory mechanisms. In this mode, due to its calm and stable emotional state, the robot exhibits strong curiosity-driven characteristics and tends to expand its understanding of the environment.

[0123] 3.2 High Wake-up Mode Unit

[0124] when When the robot enters a high-arousal mode, the increased emotional arousal enhances its behavioral initiative, triggering a forced return strategy for energy replenishment (behavior=1). In this mode, the robot ceases exploration and instead utilizes an action selection mechanism influenced by memory to return to the nearest energy node along the shortest path. This improved behavioral decision-making mechanism allows the robot to explore the dynamic balance between efficiency and survival safety directly through emotional regulation, without relying on other external indicators. This aligns better with brain-like emotion-driven design theory, resulting in more natural and flexible behavioral regulation.

[0125]

[0126] 4. Memory Module

[0127] The knowledge learned during the robot's learning process is stored and utilized, including short-term memory units, working memory units, long-term memory units, exploratory memory network units, and emotional memory units.

[0128] This invention, based on the original three components of short-term memory, working memory, and long-term memory, additionally adds an exploratory memory network and an emotional memory component. Specifically, the short-term memory mechanism is improved to a multi-energy-point-oriented short-term memory mechanism; working memory is the memory information originally intended for exploring target nodes; the exploratory memory network is a memory for environmental exploration tasks, composed of a portion of short-term memory and a portion of long-term memory; emotional memory is the memory generated by the emotional module; and long-term memory stores learned environmental information. Working memory and emotional memory jointly influence the formation of long-term memory. The entire memory module is not only interconnected with the learning module but also directly affects the specific choices made in the action selection module.

[0129] 4.1 Exploratory Memory Network Algorithm in Exploratory Memory Network Units

[0130] In order to efficiently utilize the learned environmental information, long-term memory units Improved to triplet ,in, Guided by the final learned target node Network table, Guided by the energy nodes to be learned in the end The network table incorporates an adjacency graph into the long-term memory unit. It can store the topological information of the environment, and its mathematical representation can be defined as:

[0131]

[0132] in, This represents the set of nodes that have been explored. Let represent the set of directed edges. , Represents the set of directions. The direction is east. Represents the direction west. Represents the direction north. The direction is south, and each directed edge is defined as a triple. ,in , indicating from node Along direction reachable nodes An important characteristic of adjacency graphs is their bidirectional consistency, as shown below:

[0133]

[0134] When recording a robot movement, the algorithm first checks if the starting node already exists in the adjacency graph; if not, it initializes the forward edge. Next, the same checks and initialization operations are performed on the arriving nodes, while automatically adding reverse edges. This design ensures that the robot can travel in both directions on any explored path, providing a fundamental guarantee for return navigation. Additionally, it incorporates general memories from environmental exploration. Define a set of boundary nodes, where each boundary node is a visited node with at least one unexplored direction. The boundary update rule is the same as... Consistent. The adjacency graph and boundary nodes together constitute the exploration memory network.

[0135] Building upon this foundation, the main function of the Exploratory Memory Network algorithm is to start from the current node and search for the nearest boundary node along a specified direction, thereby rapidly improving exploration efficiency. This algorithm employs a breadth-first search (BFS) strategy to ensure that the found boundary node is the closest one.

[0136] The algorithm's execution process consists of three stages:

[0137] ① Validity verification phase: The algorithm first checks whether the starting node already exists in the adjacency graph and whether there is a known connection in the specified initial direction.

[0138] ② Initial Progression Phase: After passing the validity verification, the algorithm moves one step along the initial direction and reaches the first neighbor node. At this point, the algorithm immediately checks whether the node is a boundary node. If it is, there is no need to start the complete BFS process.

[0139] ③ Breadth-first search phase: The algorithm enters the BFS main loop and takes nodes out of the queue one by one for expansion.

[0140] 4.2 Improved short-term memory mechanisms in short-term memory units

[0141] When multiple energy nodes exist in the environment, the robot needs intelligent localization, navigation, and path planning capabilities to return to the optimal energy node before its energy is depleted. However, in the original system, the short-term reflective memory mechanism only applies to the case of a single energy node. To address this, this section proposes a short-term memory mechanism for multiple energy nodes. This mechanism maintains an independent set of return path sequences for each discovered energy node and dynamically selects the optimal return target by comparing the lengths of each sequence in real time. Specifically:

[0142] 1) Let the set of energy nodes be... During the exploration process, the robot gradually discovers and records these energy nodes, forming a list of discovered energy nodes. Initially an empty set, whenever the robot first reaches a certain energy node... Add it to .

[0143] To support parallel memory across multiple energy nodes, the system provides... Each energy node in Maintain an independent set of node information sequences:

[0144]

[0145] in It still represents node information, including status. With action , Energy node The node information. The sequences of each energy node are independent and do not interfere with each other.

[0146] 2) The robot moves one step at a time (to reach a new node). Execute actions ), the system for The sequences of all energy nodes are updated synchronously. For each The following two situations should be handled:

[0147] ①If Not appeared In the set of states, that is Then, the new node will be appended. Append to the end of the sequence:

[0148]

[0149] ②If The presence of this information in the sequence indicates that the exploration path has formed a loop. At this point, the short-term memory triggering and reshaping process is executed, consistent with the single-energy-point processing method in the original system: the first arrival is deleted. The node information between the two arrivals is used to add the new node information to the sequence. This ensures the sequence... It does not contain repeating states and always records the state from the energy node. A loopless path to the current location.

[0150] 3) Define the sequence length To return to the energy node from the current location The number of steps required. When the behavioral decision-making mechanism determines that it needs to return to the energy node, the system... Choose the energy node with the shortest return path as the target:

[0151]

[0152] That is, select the energy node with the smallest current sequence length. As the return target.

[0153] 4) Determine the target energy node Subsequently, the reflection path generation process is consistent with the single-energy-point case, based on the sequence. Execution. This will ultimately generate the path from the current point to the energy nodes. The shortest return path.

[0154] 4.3 The formation mechanism of emotional memory in emotional memory units

[0155] Because the "beneficial" or "harmful" nature of the environment affects a robot's emotions, and this effect is primarily manifested in the valence dimension, when a robot experiences events with significant valence changes during exploration (such as energy replenishment or depletion), these valence changes form emotional memories associated with specific environmental states. Therefore, changes in the valence dimension provide the encoding basis for emotional memories—positive valence reinforces preference for corresponding paths, while negative valence fosters avoidance tendencies. Specifically, when a robot restarts its exploration from an energy node or a trap node, its emotional memory is oriented towards the energy node. Network table Each node it passes through will trigger the following Value update:

[0156]

[0157] As robots move away from energy nodes, valence Gradually decays, while moving in the opposite direction along the path. The algorithm iteratively traces back emotional rewards, thereby generating emotional memories. The same applies to trap nodes. In this way, the algorithm not only achieves efficient danger avoidance but also implicitly learns the shortest return path, enabling the robot to more quickly locate the direction to return to the energy node in subsequent decisions.

[0158] 5. Learning Module

[0159] It is used to help robots learn and understand unknown environments by exploring and learning target nodes and energy nodes, and updating the target orientation table and the energy orientation table.

[0160] 6. Action Selection Module

[0161] The improved action selection module is not only guided by the behavior decision-making module, but also influenced by the learning module. At the same time, the memory module also plays a direct role in action selection.

[0162] Furthermore, the action selection module first improves the "exploration and utilization" part with the support of the exploration memory network algorithm and learning mechanism. The exploration memory network algorithm and reinforcement learning complement each other, achieving automatic switching between exploration and utilization strategies by judging the exploration status of the environment. The algorithm first checks the exploration state; if the boundary set... If the boundary set is not fully explored, then the exploration is considered incomplete. In this mode, the direction selection is entirely dominated by the exploration memory network algorithm; if the boundary set is not fully explored... This indicates that full exploration has been achieved, and the system has switched to exploitation mode: querying long-term memory units for each optional direction. ,choose The direction with the largest value.

[0163] 6.1 Return Unit

[0164] When the energy replenishment forced return strategy (behavior=1) is triggered in the behavioral decision-making mechanism, for the behavior of returning to the energy node, the optimal energy node is first selected according to the improved short-term memory mechanism, and the path from the current point to the energy node is generated. The shortest return path. Specifically, the robot needs to continuously check the long-term memory unit during the return journey. The memory information, when the node is found to exist When the value is greater than or equal to 0, it indicates that there is an emotional memory that returns to the energy node along that path segment, so that node is selected. The direction with the highest value is the optimal direction, until the energy node is reached. This mechanism combines short-term memory and long-term memory, which can effectively overcome the shortcomings of relying solely on short-term memory mechanisms in complex maze environments containing loops, where it is easy to get stuck in repeated loops and produce "detour" behavior. It enables real-time correction of path selection during the return process, and finally converges to the shortest path.

[0165] 6.2 Continued Exploration Unit

[0166] When the behavior selection mechanism determines that the robot needs to perform an exploration behavior (behavior=0), it integrates memory information from the exploration memory network and adopts a three-level priority mechanism. First, it prioritizes exploring directions that have not yet been explored at that node. If no unexplored directions exist for that node, it selects the direction leading to the boundary node. If neither of the above two priority levels can determine a direction, it randomly selects one from the available directions.

[0167] Furthermore, the present invention includes specific experiments.

[0168] Example 1:

[0169] This experiment was conducted in a virtual maze environment built with CoppeliaSim, using Python 3.11 as the program runtime environment. The maze was a 4 m × 4 m space, as shown below. Figure 4 As shown, the robot used is the dr20 mobile robot integrated with CoppeliaSim. Nodes 2, 26, 29, 37, and 47 represent trap nodes, and node 18 represents an energy node. The robot's position in the diagram is the starting node, and the target node is the opening on the right. The experimental environment contains multiple energy points, and the node connections are more complex, with a large number of loops. The environment contains a total of 54 nodes, of which nodes 4, 22, 26, 50, and 53 are trap nodes, and nodes 13, 23, and 47 are energy nodes. The robot's starting node is node 7, and the target node is node 48. It is assumed that the robot's initial energy value is 15, and it consumes 1 energy point per movement. Encountering a trap node will further reduce its energy by 2 points, and reaching an energy node will replenish its energy to 25.

[0170] This experiment aims to verify that the designed cognitive model can guide the robot to return to the energy node in time to replenish energy when it is low on energy, and at the same time, learn the shortest path to the target node while ensuring survival.

[0171] According to the brain-like cognitive model of maze robots based on improved emotion and memory mechanisms proposed in this invention, the robot's first learning process in a complex maze environment is as follows: Figure 5As shown. The robot begins learning at t=0s, at which point it knows nothing about the environment, its valence and arousal are both 0, and it is in an excited emotional state. At t=8s, the robot first explores an energy node, replenishing its energy to 25. Simultaneously, the robot perceives the environment as "beneficial," and its valence reaches its maximum value of 1.0. The robot then explores further. From t=10s to t=34s, due to sufficient energy and the absence of other dangers, its arousal remains less than 0, and its valence decreases with increasing distance from the energy node, but remains greater than 0, indicating a relaxed emotional state. At t=34s, the robot encounters a trap node located at a normal node. The danger of this type of node triggers the robot's "alertness," and its arousal reaches 0.22. It then returns to the energy node, and its valence drops to -1, putting it in a tense emotional state. At t=53s, the robot returns to the energy node, its emotional state returns to relaxed, and it resumes its exploration.

[0172] The robot restarts its exploration from the energy node, prioritizing unexplored directions at each node. At t=128s, the robot is at node 28, having explored all possible directions. According to the exploration memory network algorithm, the distance to the nearest boundary node 34 is 1, so the robot heads east towards node 34. At t=140s, the robot encounters a trap node in a dead end. Since such trap nodes are relatively low-risk, the arousal level becomes negative again as the robot leaves, placing it in a sad emotional state. Until t=154s, the robot explores a second energy node in the environment, and its emotional state returns to relaxation. At t=168s, the robot again explores a trap node in a normal node, and the arousal level is again greater than 0. Having explored two energy nodes, based on the improved short-term memory mechanism, the robot chooses to return to the nearest energy node, node 47, and successfully returns at t=183s. This decision significantly improves efficiency while ensuring safety.

[0173] The robot then continued its exploration. Based on the exploration memory network algorithm, after completing a large-scale exploration of a local area, the robot proceeded to node 19 to explore unknown areas (such as...). Figure 5 (j)). When t=275s, the robot explored the third energy node in the environment. The remaining process was similar to the above. Finally, the robot reached the target node when t=325s, completing the first learning.

[0174] After completing its first learning phase, the robot began its second learning phase, part of which is illustrated in the diagram below. Figure 6 As shown, by Figure 6As can be seen, the robot, based on the long-term memory units formed after the first learning, prioritized exploring the unexplored upper-left node region under the guidance of the exploration memory network algorithm. At t=111s, the robot reached node 31 again. At this point, the arousal level was greater than 0, and the robot was in an excited emotional state. It then took the action of returning to the energy node. Since the robot had already explored energy node 23 during the first learning, based on the emotional memory generated by the valence of the current path segment, the robot directly headed north to the nearest energy node 23 to replenish energy, and successfully reached node 23 at t=118s (as shown in the image). Figure 6 (as shown in (b)). This allows the robot to choose the shortest path to replenish energy, greatly improving exploration efficiency. Finally, at t=150s, the robot finds the target node again, completing the second learning iteration.

[0175] Then, the robot began its third learning session. Figure 7 This is the result of the robot's third learning iteration. As shown in the figure, the robot successfully learned the shortest path to the target node, completing the entire learning process.

[0176] Therefore, this invention, through improvements to previously disclosed inventions, enables robots to more rationally select task priorities based on various emotional states, ensuring timely return to energy nodes to replenish energy and minimizing excessive energy surplus. Simultaneously, improved emotion and memory modules significantly enhance learning efficiency; the robot can not only find the optimal path back to the nearest energy node but also explore directly from boundary nodes, further improving efficiency.

[0177] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A brain-like cognitive system for maze robots based on improved emotion and memory mechanisms, characterized in that, include: Perception module: Acquires the robot's environmental state; Emotional module: Receives the environmental state from the perception module, constructs a dynamic emotion mechanism of "valence-arousal", and expresses emotions through two orthogonal dimensions of valence and arousal; Behavioral decision module: Judges behavior based on the arousal level in the emotion module to obtain behavioral decision results; Memory module: Stores and utilizes knowledge during the robot's learning process, including short-term memory units, working memory units, long-term memory units, exploratory memory network units, and emotional memory units; Learning module: Used to help the robot learn and understand the unknown environment. It updates the target orientation table and the energy orientation table by exploring and learning the target nodes and energy nodes. Action selection module: Based on the combined influence of the behavior decision module, the learning module, and the memory module on the action selection module, output the next action to be executed. The perception module, the emotion module, the behavior decision module, the memory module, the learning module, and the action selection module are interconnected.

2. The maze robot brain-like cognitive system based on improved emotion and memory mechanisms according to claim 1, characterized in that, The emotion module includes a valence dimension calculation unit and an arousal dimension calculation unit, wherein: The valence dimension calculation unit is used for calculating the valence dimension, specifically: The formula for calculating the design valence dimension is shown below: In the formula, valence As one of the dimensions of emotional state, it is used to characterize the robot's positive or negative evaluation of environmental stimuli; The initial valence is +1.0 or -1.0, and the emotional mechanism activates its valence value based on the objects encountered in the environment. This represents the number of steps since the event was triggered. This is the valence decay coefficient; The wakefulness dimension calculation unit is used for calculating the wakefulness dimension, specifically: Arousal level is composed of three independent sub-components, which are weighted and summed to obtain the final arousal value, as shown in the following formula: In the formula, wakefulness As an activation dimension of emotional state, high arousal means that the robot is in a highly alert state, while low arousal corresponds to an energy-saving standby state. Indicates the urgency of energy use; Indicates the threat level of the trap; Indicates the return distance pressure; These are the respective weighting coefficients.

3. The maze robot brain-like cognitive system based on improved emotion and memory mechanisms according to claim 2, characterized in that, The wakefulness dimension calculation unit includes an energy urgency calculation unit, a trap threat calculation unit, and a return distance pressure calculation unit, wherein... The energy urgency calculation unit is used to calculate the energy urgency, and the specific formula is as follows: In the formula, energy urgency Used to quantify the impact of a robot's current energy reserves on its emotions; This is the energy sensitivity coefficient, used to adjust the steepness of the response curve; It is the hyperbolic tangent function; This represents the robot's remaining energy value. The safe energy threshold for a robot is the minimum energy value at which the robot can safely return to the energy point. The trap threat level calculation unit is used to calculate the trap threat level, and the specific formula is as follows: In the formula, the trap threat level This describes the impact of fear memories after a robot encounters a dangerous area on its current emotions. This part is activated when the robot passes through a trap node and gradually decays as the distance from the trap node increases. Represents the trap type coefficient; This represents the number of steps from the current node to the dangerous node. The threat attenuation coefficient controls the rate at which fear memories fade with distance; The return distance pressure calculation unit is used to calculate the return distance pressure, and the specific formula is as follows: In the formula, the return distance pressure As the third arousal component, it directly reflects the impact of spatial distance pressure between the robot's current position and the nearest energy node on emotions; This indicates the number of steps required for the robot to return to the energy node, and the necessary information is obtained from short-term memory. This is the preset maximum number of return steps threshold.

4. The maze robot brain-like cognitive system based on improved emotion and memory mechanisms according to claim 2, characterized in that, The behavior decision module is used to directly control the robot's switching between two behavior modes based on the arousal level, including a low arousal mode unit and a high arousal mode unit, specifically: Low wake-up mode unit: used to implement when When the robot is in a low-wake-up mode, it performs exploratory behavior and selects unvisited directions or optimal actions through a memory mechanism. High wake-up mode unit: used to implement when When the robot enters a high arousal mode, the increased emotional arousal enhances its behavioral initiative, triggering a forced return strategy for energy replenishment.

5. The maze robot brain-like cognitive system based on improved emotion and memory mechanisms according to claim 1, characterized in that, The short-term memory unit employs an improved short-term memory mechanism, specifically a short-term memory mechanism oriented towards multiple energy points; The working memory unit is used to record memory information originally intended for the exploration target node task; The long-term memory unit is used to store the learned environmental information; The exploration memory network unit is used to record memory information for environmental exploration tasks. This memory consists of part short-term memory and part long-term memory, and specifically adopts the exploration memory network algorithm. The emotional memory unit is used to record the memory information generated by the emotional module, specifically employing an emotional memory formation mechanism.

6. The maze robot brain-like cognitive system based on improved emotion and memory mechanisms according to claim 5, characterized in that, The memory module includes short-term memory units, working memory units, long-term memory units, exploratory memory network units, and emotional memory units; among which, The exploratory memory network unit is composed of an adjacency graph and boundary nodes. The exploratory memory network algorithm specifically includes: The long-term memory unit Improved to triplet ,in, Guided by the final learned target node Network table, Guided by the energy nodes to be learned in the end The network table incorporates an adjacency graph into the long-term memory unit. The topology information used to store the environment is mathematically represented as: In the formula, Indicates the set of nodes that have been explored. Let represent the set of directed edges. , Represents the set of directions. The direction is east. Represents the direction west. Represents the direction north. The direction is south, and each directed edge is defined as a triple. ,in , indicating from node Along direction reachable nodes ; The bidirectional consistency of an adjacency graph is as follows: When recording a robot movement, the algorithm first checks if the starting node already exists in the adjacency graph; if not, it initializes the forward edge. Next, perform the same checks and initialization operations on the arriving nodes, while automatically adding reverse edges. In addition, general memories from environmental exploration Define a set of boundary nodes, where each boundary node is a visited node with at least one unexplored direction. The boundary update rule is the same as... Consistent; The improved short-term memory mechanism in the short-term memory unit is specifically a short-term memory mechanism for multiple energy points. This mechanism maintains an independent set of return path sequences for each discovered energy node and dynamically selects the optimal return target by comparing the lengths of each sequence in real time. The emotional memory formation mechanism in the emotional memory unit is as follows: Changes in the valence dimension provide a coding basis for emotional memory—positive valence reinforces preference for corresponding paths, while negative valence fosters avoidance tendencies. Specifically, when a robot restarts its exploration from energy nodes or trap nodes, it will target energy node-oriented... Network table Each node it passes through will trigger the following Value update: As robots move away from energy nodes, valence Gradually decays, while moving in the opposite direction along the path. The process involves tracing back emotional rewards at each level, thereby generating emotional memories; the same applies to trap nodes.

7. The maze robot brain-like cognitive system based on improved emotion and memory mechanisms according to claim 6, characterized in that, The exploratory memory network algorithm is designed to search for the nearest boundary node starting from the current node and moving in a specified direction. This algorithm employs a breadth-first search strategy, and its execution process consists of three stages: ① Validity verification phase: The algorithm first checks whether the starting node already exists in the adjacency graph, and whether there is a known connection in the specified initial direction; ② Initial Advancement Phase: After passing the validity verification, the algorithm moves one step along the initial direction and reaches the first neighbor node; at this time, the algorithm immediately checks whether the node is a boundary node. If it is, there is no need to start the complete BFS process. ③ Breadth-first search phase: The algorithm enters the BFS main loop and takes nodes out of the queue one by one for expansion.

8. The maze robot brain-like cognitive system based on improved emotion and memory mechanisms according to claim 6, characterized in that, The improved short-term memory mechanism maintains an independent set of return path sequences for each discovered energy node, and dynamically selects the optimal return target by comparing the lengths of each sequence in real time, as follows: 1) Let the set of energy nodes be... During the exploration process, the robot gradually discovers and records these energy nodes, forming a list of discovered energy nodes. Initially an empty set, whenever the robot first reaches a certain energy node... Add it to ; The system is Each energy node in Maintain an independent set of node information sequences: in It still represents node information, including status. With action , Energy node The node information; the sequences of each energy node are independent of each other and do not interfere with each other; 2) Each time the robot performs a move, that is, "reaches a new node"... Execute actions "The system" The sequence of all energy nodes in the system is updated synchronously; 3) Define the sequence length To return to the energy node from the current location The required number of steps; when the behavioral decision-making mechanism determines that it needs to return to the energy node, the system starts from... Choose the energy node with the shortest return path as the target: That is, select the energy node with the smallest current sequence length. As the return target; 4) Determine the target energy node Subsequently, the reflection path generation process is consistent with the single-energy-point case, based on the sequence. Execution; ultimately generating the path from the current point to the energy node. The shortest return path.

9. The maze robot brain-like cognitive system based on improved emotion and memory mechanisms according to claim 8, characterized in that, Each time the robot performs a move, that is, "reaches a new node"... Execute actions "The system" The sequence of all energy nodes in the middle is updated synchronously; for each The following two situations will be handled in detail: ①If Not appeared In the set of states, that is Then, the new node will be appended. Append to the end of the sequence: ②If The presence of the path in the sequence indicates a loop; at this point, a short-term memory triggering and reshaping process is executed, which involves deleting the first arrival. The node information between the two arrivals is used to add the new node information to the sequence.

10. The maze robot brain-like cognitive system based on improved emotion and memory mechanisms according to claim 9, characterized in that, The action selection module, supported by the exploratory memory network algorithm and learning mechanism, specifically includes the "exploration and utilization" part as follows: The exploratory memory network algorithm first checks the exploration state; if the boundary set... If the boundary set is not fully explored, then the exploration is considered incomplete. In this mode, the direction selection is entirely dominated by the exploration memory network algorithm; if the boundary set is not fully explored... This indicates that full exploration has been achieved, and the system has switched to exploitation mode: querying long-term memory units for each optional direction. ,choose The direction with the largest value; The action selection module includes a return unit and a continue exploration unit, wherein... The return unit is used to implement the behavior of returning to the energy node when the forced return strategy for energy replenishment is triggered in the behavior decision mechanism. First, it selects the optimal energy node according to the improved short-term memory mechanism and generates a path from the current point to the energy node. The shortest return path; The continued exploration unit is used to integrate the memory information in the exploration memory network unit and adopt a three-level priority mechanism when the behavior selection mechanism determines that the robot needs to perform an exploration behavior.

Images