Robot control method based on pulse neural network

Abstract

The invention relates to the technical field of bionic control, and discloses a robot control method based on a pulse neural network, and the method comprises the steps: carrying out the spatial-temporal feature coding of the environment perception information and body state information of a robot, and obtaining a pulse feature sequence; performing pulse dependence fusion on the pulse characteristic sequence and the feedback pulse signal to obtain a bionic information processing process, and performing membrane potential integration on the bionic information processing process to obtain an integrated output quantity; performing pulse distribution processing on the integrated output quantity to obtain a driving pulse signal; performing path fitting on the driving pulse signal to obtain a continuous motion track; based on the continuous motion track and the bionic information processing process, performing joint space mapping on the continuous motion track to obtain an execution control instruction; the application executes the control instruction, synchronously collects a pulse response signal and feeds back the pulse response signal in real time so as to obtain an optimization control instruction of the robot; the pulse neural network-based robot control efficiency can be improved.

Description

Technical Field

[0001] This invention relates to the field of biomimetic control technology, and in particular to a robot control method based on spiking neural networks. Background Technology

[0002] Existing robot control methods lack efficient spatiotemporal feature fusion and encoding mechanisms when processing environmental perception information and body state information. This makes it difficult to fully capture the dynamic correlation and deep patterns between the two types of information, resulting in insufficient accuracy of the generated basic feature data. Furthermore, the failure to effectively simulate the impulse-dependent evolution and membrane potential integration characteristics of biological neurons during information processing leads to a lack of coherence and biomimicry in the process, consequently affecting the quality of subsequent drive signals and control commands.

[0003] Existing technologies lack scientific quantitative evaluation and competitive arbitration mechanisms in the pulse delivery decision-making process, making it difficult to accurately select the optimal delivery point and resulting in insufficient targeting and rationality of the drive pulse signals. Furthermore, in the path fitting and joint space mapping processes, dynamic adjustments are not fully integrated with task requirements and biomimetic characteristics, making it difficult to simultaneously achieve smoothness of the motion trajectory and joint coordination. Moreover, the lack of a real-time and effective feedback adjustment mechanism prevents timely optimization of the control strategy based on the robot's actual motion state. Therefore, improving the accuracy of robot control and the adaptability of control strategies has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a robot control method based on spiking neural networks to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a robot control method based on a spiking neural network, comprising: S1. Spatiotemporal feature encoding is performed on the robot's environmental perception information and body state information to obtain the robot's pulse feature sequence; S2. Perform pulse-dependent fusion of the pulse feature sequence and the feedback pulse signal of the robot to obtain the biomimetic information processing process of the robot, and perform membrane potential integration on the biomimetic information processing process to obtain the integrated output of the robot. S3. Perform pulse output processing on the integrated output to obtain the drive pulse signal of the robot; S4. Perform path fitting on the drive pulse signal to obtain the continuous motion trajectory of the robot; S5. Based on the continuous motion trajectory and the biomimetic information processing process, perform joint space mapping on the continuous motion trajectory to obtain the robot's execution control commands; S6. Apply the execution control command, synchronously collect the robot's pulse response signal, and feed back the optimized pulse response signal to S2 to obtain the robot's optimized control command.

[0006] In a preferred embodiment, the step of performing spatiotemporal feature encoding on the robot's environmental perception information and body state information to obtain the robot's pulse feature sequence includes: Asynchronous event-driven sampling is performed on the robot to obtain the robot's environmental perception information; The environmental perception information is represented by event-based sparse representation to obtain the spatiotemporal event stream of the robot; The robot's pose is captured to obtain the robot's body state information; Joint coordination analysis is performed on the body state information to obtain the body motion flow of the robot; Based on the spatiotemporal event flow and the body motion flow, pulse coding is performed on the robot to obtain the robot's pulse feature sequence.

[0007] In a preferred embodiment, the step of pulse-dependent fusion of the pulse feature sequence and the robot's feedback pulse signal to obtain the robot's biomimetic information processing process, and the integration of the biomimetic information processing process with membrane potential to obtain the robot's integrated output, includes: The pulse feature sequence and the feedback pulse signal of the robot are biomimeticly time-aligned to obtain the synchronization pulse group of the robot; Based on the synchronization pulse group, the pulse feature sequence is subjected to dependency evolution to obtain the connection strength of the pulse feature sequence; Based on the connection strength, the firing behavior of the synchronization pulse group is morphologically shaped to obtain the biomimetic information processing process of the robot; The local potential changes during the biomimetic information processing are spatiotemporally accumulated to obtain the accumulated membrane potential field of the biomimetic information processing process. Field domain filtering is performed on the accumulated membrane potential field to obtain the gated potential field of the accumulated membrane potential field; The gated potential field is subjected to transmembrane phase convergence to obtain the integrated output of the robot.

[0008] In a preferred embodiment, the step of pulse-emitting processing the integrated output to obtain the robot's drive pulse signal includes: A membrane potential field is constructed on the integrated output quantity to obtain the spatial membrane potential field of the integrated output quantity. The spatial membrane potential field is biomimetically evolved to obtain the updated membrane potential field of the spatial membrane potential field; Based on the updated membrane potential field, extreme region detection is performed on the robot to obtain the candidate release points of the robot; Competitive arbitration is conducted on the candidate distribution points to obtain the priority distribution points for the candidate distribution points; Based on the priority distribution point, the pulse distribution of the robot is driven and output to obtain the driving pulse signal of the robot.

[0009] In a preferred embodiment, the step of competitively arbitrating the candidate distribution points to obtain the priority distribution points includes: Based on the updated membrane potential field, the potential amplitude of the candidate firing points is extracted to obtain the instantaneous membrane potential amplitude of the candidate firing points; Based on the robot's biomimetic information processing process, the historical distribution time of the candidate distribution points is retrieved to obtain the historical distribution time of the candidate distribution points. The instantaneous membrane potential amplitude and the historical emission time are weighted and evaluated to obtain the emission priority score of the candidate emission point; Based on the priority score, the candidate distribution points are ranked globally to obtain the competitive dominance relationship of the candidate distribution points; Based on the aforementioned competitive dominance relationship, conflict analysis is performed on the candidate distribution points to obtain the priority distribution points of the candidate distribution points.

[0010] In a preferred embodiment, the formula for calculating the priority score is as follows: ; in, For the first The priority score for each of the candidate distribution points. For the first The instantaneous membrane potential amplitude, For the historical release time, For the current moment, The preset weighting coefficients, The preset weighting coefficients, This is the preset time decay coefficient.

[0011] In a preferred embodiment, the step of path fitting of the drive pulse signal to obtain the continuous motion trajectory of the robot includes: Spatiotemporal pose mapping is performed on the timing of the drive pulse signal to obtain the discrete spatial path of the robot's end effector; The discrete spatial path of the end effector is subjected to curvature continuity processing to obtain a smooth path segment of the robot; Based on the smooth path segment, the robot is embedded into a manifold to obtain the smooth motion manifold of the robot; The smooth motion manifold is fitted point by point to obtain the continuous motion trajectory of the robot.

[0012] In a preferred embodiment, the step of performing joint space mapping on the continuous motion trajectory based on the continuous motion trajectory and the biomimetic information processing to obtain the robot's execution control commands includes: Task-related feature decoding is performed on the biomimetic information processing process to obtain the biomimetic context information of the biomimetic information processing process; Based on the biomimetic context information, the continuous motion trajectory is segmented into task semantics to obtain a set of trajectory segments of the continuous motion trajectory; The importance of the trajectory segment set is marked to obtain a hierarchical trajectory segment sequence of the continuous motion trajectory.

[0013] Based on the hierarchical trajectory segment sequence, the trajectory segment sequence is projected into joint space to obtain the joint space path of the robot; Based on the biomimetic context information, the joint space path is biomimetically adjusted to obtain the robot's execution control commands.

[0014] In a preferred embodiment, the step of biomimeticly adjusting the joint space path based on the biomimetic context information to obtain the robot's execution control commands includes: The biomimetic context information is analyzed in a spatiotemporal manner to obtain the motion rhythm synchronization signal of the robot. The biomimetic context information is subjected to joint co-compilation to obtain the inter-joint co-pulse code of the robot; Based on the motion rhythm synchronization signal, the joint space path is dynamically scaled on the time axis to obtain the pulse firing synchronization path of the robot; Based on the inter-joint cooperative pulse coding, the relative amplitude of joint motion in the pulse firing synchronization path is adjusted by cooperative gain to obtain the biomimetic coupling path of the robot. The biomimetic coupling path is mapped to the robot's execution control commands.

[0015] In a preferred embodiment, the application of the execution control command synchronously acquires the robot's impulse response signal and feeds back the optimized impulse response signal to S2 to obtain the robot's optimized control command, including: Servo tracking is performed on the executed control commands to obtain the actual joint movements of the robot; The execution control commands are subjected to task space feedforward analysis to obtain the end-effector pose changes of the robot; Based on the actual joint motion and the end-effector pose change, the robot's motion state is transformed into a micro-dynamic event-based state to obtain the robot's original impulse response stream; The amplitude of the original impulse response stream is normalized to obtain the standardized impulse response signal of the robot; The standardized impulse response signal is subjected to impulse sparsification to obtain the optimized feedback impulse signal of the robot; The optimized feedback pulse signal is fed back in real time to obtain the optimized control command for the robot.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This control method fully explores the spatiotemporal correlation and inherent laws between robot environmental perception information and body state information by encoding their spatiotemporal features. Combined with pulse-dependent fusion and membrane potential integration mechanisms, it achieves a biomimetic information processing process, significantly improving the completeness and accuracy of information processing. Simultaneously, through quantitative calculation of priority scoring and a competitive arbitration mechanism, it accurately selects priority distribution points, making the generation of drive pulse signals more targeted and rational. This effectively improves the precision of robot motion control and the smoothness of motion trajectory, ensuring the robot can accurately respond to changes in the environment and its own state.

[0017] 2. Utilizing a real-time feedback adjustment mechanism, this method synchronously acquires the robot's impulse response signals, processes them through standardization and sparsification, and then feeds them back to the information fusion stage. This constructs a closed-loop optimized control system, allowing control commands to be dynamically adjusted according to the robot's actual motion state, significantly improving the adaptability and robustness of the control. Furthermore, through joint space mapping and biomimetic adjustment, the continuous motion trajectory is deeply integrated with task-related biomimetic context information, making the execution of control commands more aligned with the robot's joint cooperative motion characteristics. This further improves the fit between the motion trajectory and task requirements, enhancing the overall efficiency and reliability of robot control. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a robot control method based on a spiking neural network according to an embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] This application provides a robot control method based on spiking neural networks. The executing entity of the spiking neural network-based robot control method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the spiking neural network-based robot control method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0021] Reference Figure 1 The diagram shown is a flowchart illustrating a robot control method based on a spiking neural network according to an embodiment of the present invention. In this embodiment, the robot control method based on a spiking neural network includes: S1. Spatiotemporal feature encoding is performed on the robot's environmental perception information and body state information to obtain the robot's pulse feature sequence; In this embodiment of the invention, the step of performing spatiotemporal feature encoding on the robot's environmental perception information and body state information to obtain the robot's pulse feature sequence includes: Asynchronous event-driven sampling is performed on the robot to obtain the robot's environmental perception information; The environmental perception information is represented by event-based sparse representation to obtain the spatiotemporal event stream of the robot; The robot's pose is captured to obtain the robot's body state information; Joint coordination analysis is performed on the body state information to obtain the body motion flow of the robot; Based on the spatiotemporal event flow and the body motion flow, pulse coding is performed on the robot to obtain the robot's pulse feature sequence.

[0022] The robot uses various sensing devices such as vision sensors, tactile sensors, and distance sensors to sample in an asynchronous event-driven manner. Instead of following a fixed time interval, when a perceptible event occurs in the environment, such as a change in brightness, contact with an object, or a change in distance, the sensors immediately respond and capture the specific information corresponding to the event, including the spatial location of the event, the type of event, and related physical parameter data. All the captured information related to the environment constitutes the robot's environmental perception information.

[0023] The collected environmental perception information is filtered and processed to remove duplicate, redundant, and meaningless invalid information, retaining only the core event data that can accurately reflect key environmental changes. Then, according to the chronological order of each core event, combined with the spatial location information of the event, it is systematically organized and arranged in an orderly manner to form a continuous data stream that contains key information in both time and space dimensions. This data stream is the spatiotemporal event stream.

[0024] By using the robot's own gyroscope, accelerometer, joint angle sensors and other devices, the robot's position coordinate data in three-dimensional space is captured in real time to determine its specific orientation in space. At the same time, the robot's attitude information, including pitch angle, roll angle, yaw angle, etc., is detected, and the current angle values ​​of each joint of the robot are recorded in a comprehensive manner. All this data on the robot's position, attitude and joint angles are integrated and summarized to obtain body state information that can fully reflect the robot's current status.

[0025] By conducting in-depth analysis of the joint angle data contained in the body state information, we can explore the cooperative relationship between different joints during the movement process, clarify the logical sequence of joint movement, the degree of synchronization of angle changes, and the law of mutual influence between joints. Based on these analysis results, we can systematically sort out the continuous change process of the coordinated operation of each joint in the order of time. This continuous change process is the body motion flow.

[0026] By associating key environmental change information contained in the spatiotemporal event stream with joint coordinated motion information embodied in the ontological motion stream, the inherent logical connection between the two types of information is deeply explored. Based on the signal transmission characteristics of the spiking neural network, this associated information is transformed into discrete pulse signals. Each pulse signal precisely corresponds to a specific environmental event or joint motion state. These discrete pulse signals are then arranged in an orderly manner according to time sequence and logical association to form an ordered pulse set, which is the pulse feature sequence.

[0027] The beneficial effects are as follows: Asynchronous event-driven sampling accurately captures key environmental changes, and spatiotemporal event flows are formed through event-based sparse representation, ensuring the effectiveness and orderliness of environmental information; pose acquisition obtains comprehensive body state information of the robot, and body motion flow is obtained by combining joint coordination analysis, accurately reflecting the robot's own motion characteristics; finally, pulse coding transforms the two types of information into pulse feature sequences, fully exploring the spatiotemporal correlation and inherent laws of environmental and body information, providing high-quality and high-precision basic data for subsequent pulse-dependent fusion, membrane potential integration and other links, laying a solid foundation for improving the accuracy of robot control.

[0028] S2. Perform pulse-dependent fusion of the pulse feature sequence and the feedback pulse signal of the robot to obtain the biomimetic information processing process of the robot, and perform membrane potential integration on the biomimetic information processing process to obtain the integrated output of the robot. In this embodiment of the invention, the step of pulse-dependent fusion of the pulse feature sequence and the feedback pulse signal of the robot to obtain the biomimetic information processing process of the robot, and the integration of the biomimetic information processing process with membrane potential to obtain the integrated output of the robot, includes: The pulse feature sequence and the feedback pulse signal of the robot are biomimeticly time-aligned to obtain the synchronization pulse group of the robot; Based on the synchronization pulse group, the pulse feature sequence is subjected to dependency evolution to obtain the connection strength of the pulse feature sequence; Based on the connection strength, the firing behavior of the synchronization pulse group is morphologically shaped to obtain the biomimetic information processing process of the robot; The local potential changes during the biomimetic information processing are spatiotemporally accumulated to obtain the accumulated membrane potential field of the biomimetic information processing process. Field domain filtering is performed on the accumulated membrane potential field to obtain the gated potential field of the accumulated membrane potential field; The gated potential field is subjected to transmembrane phase convergence to obtain the integrated output of the robot.

[0029] Feedback pulse signals are a comprehensive pulsed representation of the robot's actual joint motion state and the pose deviation of the end effector. Referring to the temporal coordination characteristics of signal transmission in biological neural systems, the pulse feature sequence and the feedback pulse signal are compared to identify the pulse units that correspond to each other in time and are logically related in the two types of signals. These matching pulse units are grouped together to ensure that the pulses in each group are synchronized in the time dimension, and finally a well-structured and time-consistent synchronous pulse group is formed.

[0030] By deeply analyzing the frequency of co-occurrence of each pulse in the synchronization pulse group at different times, as well as the correlation between pulses inducing and interacting with each other, a systematic evaluation is conducted based on the tightness of these correlations. The correlation between each pair of pulses is assigned a specific characterization result, which reflects the tightness of the correlation between pulses and is called the connection strength of the pulse characteristic sequence.

[0031] Based on connection strength, the pulse firing order in the synchronization pulse group is adjusted so that pulses with high connection strength participate in the firing first. At the same time, the pulse firing interval is reasonably set so that the pulse firing rhythm matches the distribution of connection strength, and the duration of pulse firing is standardized. This simulates the differentiated firing pattern of biological nerves based on differences in connection strength. This complete and biomimetic pulse processing flow is the biomimetic information processing process.

[0032] The potential changes in each local area during the entire biomimetic information processing process are tracked. The potential changes generated by each local area at different times are accumulated one by one in chronological order. At the same time, the spatial location information of each local area is recorded. These potential data, which include the time accumulation results and spatial location information, are integrated to form a potential distribution covering the spatiotemporal dimensions, namely the cumulative membrane potential field.

[0033] A clear criterion for determining the effectiveness of potentials is established. Based on this criterion, all potential regions in the cumulative membrane potential field are screened one by one. Regions whose potential values ​​meet the criterion and can play a positive role in subsequent signal processing are retained, while invalid regions whose potential values ​​do not meet the criterion and do not have practical functions are eliminated. The potential distribution regions that remain after screening are the gated potential fields.

[0034] By meticulously analyzing the phase attributes of potentials in different regions of the gated potential field, potential signals with consistent phase characteristics and potential for synergistic effects are selected. These potential signals with phase synergy are then centrally summarized and fused to resolve potential signals with conflicting phases. The fused potential signals are then transformed into unified, coherent output data with a clear functional orientation, which is the integrated output quantity.

[0035] The beneficial effects are that the timing consistency of the pulse signal is ensured by biomimetic timing alignment, the correlation characteristics between pulses are accurately captured by dependency evolution, the morphological shaping realizes a biomimetic information processing mode, and the potential signal is gradually optimized by spatiotemporal accumulation, field screening and transmembrane phase convergence. The progressive processing process makes the final integrated output both highly accurate and highly coherent, providing high-quality data support for the generation of subsequent driving pulse signals, and effectively improving the biomimetic nature of robot control and the reliability of control decisions.

[0036] S3. Perform pulse output processing on the integrated output to obtain the drive pulse signal of the robot; In this embodiment of the invention, the step of pulse-emitting processing of the integrated output to obtain the drive pulse signal of the robot includes: A membrane potential field is constructed on the integrated output quantity to obtain the spatial membrane potential field of the integrated output quantity. The spatial membrane potential field is biomimetically evolved to obtain the updated membrane potential field of the spatial membrane potential field; Based on the updated membrane potential field, extreme region detection is performed on the robot to obtain the candidate release points of the robot; Competitive arbitration is conducted on the candidate distribution points to obtain the priority distribution points for the candidate distribution points; Based on the priority distribution point, the pulse distribution of the robot is driven and output to obtain the driving pulse signal of the robot.

[0037] The competitive arbitration of the candidate distribution points to obtain the priority distribution points includes: Based on the updated membrane potential field, the potential amplitude of the candidate firing points is extracted to obtain the instantaneous membrane potential amplitude of the candidate firing points; Based on the robot's biomimetic information processing process, the historical distribution time of the candidate distribution points is retrieved to obtain the historical distribution time of the candidate distribution points. The instantaneous membrane potential amplitude and the historical emission time are weighted and evaluated to obtain the emission priority score of the candidate emission point; Based on the priority score, the candidate distribution points are ranked globally to obtain the competitive dominance relationship of the candidate distribution points; Based on the aforementioned competitive dominance relationship, conflict analysis is performed on the candidate distribution points to obtain the priority distribution points of the candidate distribution points.

[0038] The formula for calculating the priority score is as follows: ; in, For the first The priority score for each of the candidate distribution points. For the first The instantaneous membrane potential amplitude, For the historical release time, For the current moment, The preset weighting coefficients, The preset weighting coefficients, This is the preset time decay coefficient.

[0039] The potential data contained in the integrated output are systematically divided according to the spatial dimension of robot motion control. The relationship between each spatial position and the corresponding potential value is clarified. The scattered potential information is organized into an overall field that covers the spatial range related to robot motion and can intuitively reflect the potential distribution. This field is the spatial membrane potential field of the integrated output.

[0040] By simulating the dynamic characteristics of membrane potential changes with time and environment in biological nervous systems and referring to the conduction, diffusion and interaction laws of biological nerve potentials, the potential values ​​at various spatial locations in the spatial membrane potential field are dynamically adjusted so that the changes in the potential field conform to the response pattern of biological nerves. The new membrane potential field formed after dynamic adjustment is the updated membrane potential field.

[0041] The robot comprehensively traverses and updates all spatial locations in the membrane potential field, compares the potential values ​​of each location with those of its neighboring regions, and identifies regions where the potential values ​​are significantly higher than those of the surrounding regions, forming local peaks. These regions where local peaks are located are the extreme value regions, and the core location of each extreme value region is the candidate launch point for the robot.

[0042] For each candidate firing point, the specific potential value corresponding to it in the updated membrane potential field is directly read. This value can reflect the current potential intensity of the candidate firing point in real time. The read potential value is the instantaneous membrane potential amplitude of the candidate firing point.

[0043] By reviewing all pulse emission-related data archives recorded during the robot's bionic information processing, and tracing the specific time points at which pulse emission occurred during past information processing for each candidate emission point, this time point is accurately extracted as the historical emission time of the candidate emission point.

[0044] Based on the influence of instantaneous membrane potential amplitude and historical release time on pulse release, each is assigned a fixed importance weight. The larger the instantaneous membrane potential amplitude, the greater its contribution to release priority. The closer the historical release time is to the current time, the higher the bonus to release priority. The comprehensive score of each candidate release point is calculated by combining the influence of these two factors. This comprehensive score is the release priority score.

[0045] The instantaneous membrane potential amplitude is obtained by extracting the potential amplitude of candidate firing points based on the updated membrane potential field. This extraction process directly filters the potential values ​​corresponding to candidate firing points from the updated membrane potential field. The historical firing time is obtained by retrieving past firing records of candidate firing points based on the robot's biomimetic information processing process. The retrieval process strictly follows the firing time sequence data retained during biomimetic information processing. The current time is the real-time time data collected during the robot's operation. The preset weighting coefficient and preset time decay coefficient are fixed values ​​set in advance according to control requirements and robot operating characteristics before the robot control program starts, used to adjust the degree of influence of different factors on the final result.

[0046] The priority score for firing is obtained by weighting the instantaneous membrane potential amplitude with historical firing times. The weighting process involves multiplying the instantaneous membrane potential amplitude by a preset weighting coefficient, then exponentially multiplying the time difference between historical firing times and the current time by the same preset weighting coefficient, and finally adding the two scores to obtain the final score. This score is used for global competitive ranking of all candidate firing points. The ranking process arranges candidate firing points in descending order of score, thereby obtaining the competitive dominance relationship between candidate firing points. Based on this competitive dominance relationship, conflict resolution is performed on candidate firing points with firing conflicts, ultimately determining the priority firing point.

[0047] The larger the instantaneous membrane potential amplitude, the larger the corresponding weighted result, and the higher the issuance priority score. The larger the time interval between the historical issuance time and the current time, the smaller the result after exponential calculation, the smaller the corresponding weighted result, and the lower the issuance priority score. The larger the preset weighting coefficient, the stronger the influence of the instantaneous membrane potential amplitude or exponential calculation result on the issuance priority score. The larger the preset time decay coefficient, the more significant the weakening effect of the time difference between the historical issuance time and the current time on the exponential calculation result, and thus the more significant the impact on the issuance priority score.

[0048] Collect priority scores for all candidate distribution points and rank them from highest to lowest score. This determines the priority of each candidate distribution point among all points. Higher-scoring candidate distribution points have a competitive advantage and can dominate lower-scoring points. This dominance is based on priority scores. After global competitive ranking, higher-scoring points either have priority in occupying the pulse distribution time window, or, in the event of a distribution conflict within the same time window, are retained due to their stronger necessity and rationality. Lower-scoring points, however, are suppressed due to insufficient priority. This relationship of time resource allocation and priority in distribution during conflicts constitutes the competitive dominance of candidate distribution points.

[0049] For candidate distribution points that have priority conflicts after sorting and may issue pulses simultaneously, the candidate distribution points with the dominant position are retained according to the competitive dominance relationship, and the dominated candidate distribution points are eliminated to resolve the distribution conflicts between points. The distribution point with no conflict and the highest priority is finally determined as the priority distribution point of the candidate distribution points.

[0050] According to the priority order of the priority distribution points, the pulse distribution function of each priority distribution point is activated in sequence. The pulse signal of each point is sorted and transmitted in a format that can be recognized by the robot drive system, forming a signal that can directly drive the movement of each execution part of the robot. This signal is the robot's drive pulse signal.

[0051] The beneficial effects are as follows: by constructing a spatial membrane potential field and biomimetic evolution, the dynamic rationality and biomimeticity of the potential information are ensured; extreme value region detection accurately locates candidate release points with release potential; competitive arbitration selects the optimal priority release point by comprehensively evaluating the instantaneous membrane potential amplitude and historical release time in multiple dimensions, effectively avoiding release conflicts; finally, the driving pulse signal generated based on the priority release point is highly targeted and logically rigorous, which can provide accurate and reliable driving support for robot movement, significantly improving the accuracy and response efficiency of robot control.

[0052] S4. Perform path fitting on the drive pulse signal to obtain the continuous motion trajectory of the robot; In this embodiment of the invention, the step of performing path fitting on the driving pulse signal to obtain the continuous motion trajectory of the robot includes: Spatiotemporal pose mapping is performed on the timing of the drive pulse signal to obtain the discrete spatial path of the robot's end effector; The discrete spatial path of the end effector is subjected to curvature continuity processing to obtain a smooth path segment of the robot; Based on the smooth path segment, the robot is embedded into a manifold to obtain the smooth motion manifold of the robot; The smooth motion manifold is fitted point by point to obtain the continuous motion trajectory of the robot.

[0053] By analyzing each firing moment of the drive pulse signal, a one-to-one correspondence is established between the pulse signal intensity at each moment and the spatial position and attitude parameters of the robot's end effector. The specific coordinates of the end effector in three-dimensional space and attitude information such as pitch, roll, and yaw at each firing moment are accurately determined. According to the order of firing time, these discrete position and attitude data points are arranged in an orderly manner to form a path composed of a series of discrete data points. This path is the discrete spatial path of the end effector.

[0054] By analyzing the curvature of line segments between adjacent data points on the discrete spatial path of the end effector one by one, line segments with abrupt curvature changes and sharp turning angles are identified. An appropriate number of transition data points are inserted between adjacent data points of these abrupt line segments, and the spatial position of the transition data points is adjusted so that the curvature change between adjacent line segments gradually transitions without abrupt turns. This ensures that the curvature of any adjacent line segments on the entire path can be smoothly connected. The path segment without abrupt changes and sharp angles formed after adjustment is the smooth path segment.

[0055] Based on smooth path segments, and combined with the robot's joint range of motion, kinematic constraints, and spatial limitations of task execution, a low-dimensional space that conforms to the robot's motion characteristics is constructed. All smooth path segments are then fully embedded into this low-dimensional space according to their positional relationships and motion logic in three-dimensional space. This allows the embedded paths to naturally follow the robot's motion constraints, forming a continuous, smooth, and conflict-free motion space representation. This representation is the robot's smooth motion manifold.

[0056] Along the extension direction of the smooth motion manifold, a large number of continuous path points are selected evenly according to the preset accuracy standard. The distribution density of these points is sufficient to ensure that the connected trajectory has no obvious discontinuities. Then, continuous curves are used to connect these selected points in sequence to ensure that the curves can closely fit the contour of the smooth motion manifold and not deviate from its motion trend. The final seamless, smooth and continuous curve is the robot's continuous motion trajectory.

[0057] The beneficial effects are as follows: by converting the driving pulse signal into a discrete spatial path through spatiotemporal pose mapping, the pulse delivery and the motion state of the end effector are accurately correlated; curvature continuity processing eliminates abrupt changes and sharp corners in the path, improving the smoothness of the path; manifold embedding makes the path fit the robot's motion constraints, ensuring the feasibility of the motion; point-by-point fitting finally forms a continuous motion trajectory. The entire process is optimized layer by layer, ensuring the coherence, smoothness and rationality of the motion trajectory, providing a high-quality trajectory foundation for the robot's subsequent joint space mapping and precise execution of control commands, effectively improving the stability of robot motion and the accuracy of task execution.

[0058] S5. Based on the continuous motion trajectory and the biomimetic information processing process, perform joint space mapping on the continuous motion trajectory to obtain the robot's execution control commands; In this embodiment of the invention, the step of performing joint space mapping on the continuous motion trajectory based on the continuous motion trajectory and the biomimetic information processing to obtain the robot's execution control commands includes: Task-related feature decoding is performed on the biomimetic information processing process to obtain the biomimetic context information of the biomimetic information processing process; Based on the biomimetic context information, the continuous motion trajectory is segmented into task semantics to obtain a set of trajectory segments of the continuous motion trajectory; The importance of the trajectory segment set is marked to obtain a hierarchical trajectory segment sequence of the continuous motion trajectory.

[0059] Based on the hierarchical trajectory segment sequence, the trajectory segment sequence is projected into joint space to obtain the joint space path of the robot; Based on the biomimetic context information, the joint space path is biomimetically adjusted to obtain the robot's execution control commands.

[0060] The step of biomimeticly adjusting the joint space path based on the biomimetic context information to obtain the robot's execution control commands includes: The biomimetic context information is analyzed in a spatiotemporal manner to obtain the motion rhythm synchronization signal of the robot. The biomimetic context information is subjected to joint co-compilation to obtain the inter-joint co-pulse code of the robot; Based on the motion rhythm synchronization signal, the joint space path is dynamically scaled on the time axis to obtain the pulse firing synchronization path of the robot; Based on the inter-joint cooperative pulse coding, the relative amplitude of joint motion in the pulse firing synchronization path is adjusted by cooperative gain to obtain the biomimetic coupling path of the robot. The biomimetic coupling path is mapped to the robot's execution control commands.

[0061] By deeply analyzing the pulse firing characteristics, membrane potential change patterns, and inter-pulse correlation patterns contained in the biomimetic information processing process, key features directly related to the current task are selected, including core elements such as the action type, execution accuracy requirements, and motion rhythm corresponding to the task. These key features are systematically integrated and interpreted to form comprehensive information that can clearly reflect the task background, execution requirements, and motion correlation. This information is the biomimetic context information.

[0062] Based on the clear task phase divisions and action function definitions in the biomimetic context information, the trajectory segments corresponding to each task stage in the continuous motion trajectory are identified. According to the logical order of task execution and the semantic boundaries of actions, the entire continuous motion trajectory is divided into multiple trajectory segments with independent functional meanings. The set of these trajectory segments is the trajectory segment set of the continuous motion trajectory.

[0063] Based on the core objectives and execution standards of the current task, the impact of each trajectory segment in the trajectory segment set on the completion of the task is evaluated one by one. The functional weight of each trajectory segment is clarified. The core trajectory segments that directly determine the success or failure of the task are marked as high importance, the trajectory segments that assist in completing the task are marked as medium importance, and the trajectory segments with less impact are marked as low importance. All trajectory segments are arranged in order of importance from high to low to form a hierarchical trajectory segment sequence of continuous motion trajectory.

[0064] Based on the hierarchical trajectory segment sequence, the position and attitude information of each trajectory segment in Cartesian space are transformed into motion parameters such as angles and angular velocities corresponding to each joint of the robot through coordinate transformation and inverse kinematics calculation. At the same time, computing resources are allocated according to the importance level of the trajectory segments to ensure higher accuracy of joint parameter transformation for high-importance trajectory segments. The joint motion parameters corresponding to all trajectory segments are concatenated in chronological order to form a joint motion parameter sequence covering the entire task process. This sequence is the joint space path of the robot.

[0065] By meticulously analyzing the pulse firing time interval, firing frequency, and spatial distribution characteristics in the biomimetic context information, we can extract the rhythmic patterns that reflect biological motion coordination, clarify the synchronization relationship between pulse firing and joint movement in the time dimension, and transform this rhythmic pattern into a signal with a clear time reference. This signal can guide joint movement to maintain a consistent rhythm with pulse firing, which is the robot's motion rhythm synchronization signal.

[0066] This study delves into the coordination logic of joint movements within the biomimetic context information, including the sequence of joint movements, the proportional relationship of movement amplitude, and the timing of movement state switching. These coordination characteristics are then transformed into standardized pulse sequence codes, with each code corresponding to a set of joint coordination movement rules. Through this code, the coordination constraints of each joint during movement can be clearly defined. This code is the inter-joint coordination pulse code of the robot.

[0067] Using the motion rhythm synchronization signal as a time reference, the time axis of the joint space path is dynamically adjusted. When the synchronization signal indicates that the motion rhythm is accelerating, the time length of the corresponding path segment is shortened proportionally. When the synchronization signal indicates that the motion rhythm is slowing down, the time length of the corresponding path segment is extended proportionally. This ensures that the time evolution of the joint space path is completely matched with the pulse firing rhythm. The adjusted joint space path is the robot's pulse firing synchronization path.

[0068] Based on the joint coordination rules specified in the joint coordination pulse coding, the motion amplitude of each joint in the pulse firing synchronization path is proportionally adjusted to enhance the motion amplitude of the key joints specified in the coordination coding and to appropriately adjust the motion amplitude of the auxiliary joints so that the relative motion amplitude of each joint meets the coordination constraint requirements, forming a path in which the motion of each joint is highly coordinated and mutually cooperates. This path is the biomimetic coupling path of the robot.

[0069] The parameters such as joint angles, angular velocities, and motion times contained in the biomimetic coupling path are converted according to the control protocol and signal format requirements of each joint of the robot. The path information is then transformed into command signals that can directly drive the robot's joint actuators. These command signals can precisely control the motion state of each joint, which are the robot's execution control commands.

[0070] The beneficial effects are as follows: by decoding task-related features and semantic segmentation, the motion trajectory is accurately matched with the task requirements; importance marking and joint space projection ensure the motion accuracy of the core trajectory segment; the bionic adjustment link, through rhythm synchronization and joint co-encoding, makes the joint movement conform to the biological movement characteristics, and the co-gain adjustment improves the coordination of the joint movement; finally, the bionic coupling path is mapped into the execution control command. The whole process fully integrates bionic information and task requirements, so that the control command has both accuracy, coordination and bionic characteristics, effectively improving the coordination of robot joint movement and the reliability of task execution, and ensuring that the robot's actions are more in line with the needs of actual application scenarios.

[0071] S6. Apply the execution control command, synchronously collect the robot's pulse response signal, and feed back the optimized pulse response signal to S2 to obtain the robot's optimized control command.

[0072] In this embodiment of the invention, the application of the execution control command, synchronously acquiring the robot's impulse response signal, and feeding back the optimized impulse response signal to S2 to obtain the robot's optimized control command includes: Servo tracking is performed on the executed control commands to obtain the actual joint movements of the robot; The execution control commands are subjected to task space feedforward analysis to obtain the end-effector pose changes of the robot; Based on the actual joint motion and the end-effector pose change, the robot's motion state is transformed into a micro-dynamic event-based state to obtain the robot's original impulse response stream; The amplitude of the original impulse response stream is normalized to obtain the standardized impulse response signal of the robot; The standardized impulse response signal is subjected to impulse sparsification to obtain the optimized feedback impulse signal of the robot; The optimized feedback pulse signal is fed back in real time to obtain the optimized control command for the robot.

[0073] After receiving and executing control commands, the robot's servo system monitors the rotation angle, speed, acceleration, and other motion parameters of each joint in real time. It continuously compares these monitoring data with the target motion parameters set in the commands, and fully records the actual motion trajectory, state changes, and motion details of each joint during the execution of commands. This data set, which comprehensively reflects the actual motion of the joints, is the robot's actual joint motion.

[0074] Based on the joint motion targets, task execution paths, and accuracy requirements specified in the execution control commands, and combined with the robot's kinematic model, the expected position coordinates and attitude parameters of the end effector in the task space are predicted in advance. At the same time, the actual position and attitude information of the end effector are collected in real time through the positioning and attitude sensors on the end effector, the deviation between the expected value and the actual value is calculated, and the dynamic changes in the position and attitude of the end effector throughout the entire task execution are fully recorded. These dynamic changes are the robot's end-effector pose changes.

[0075] By comprehensively integrating detailed data such as the subtle adjustments of joint angles, the minute fluctuations in motion speed, and the instantaneous changes in acceleration during actual joint movements, as well as information such as the minute displacements and attitude deflections in the end-effector pose changes, each of these independent minute dynamic changes is defined as a micro-dynamic event. A corresponding pulse signal feature is matched to each micro-dynamic event, and all the corresponding pulse signals are sequentially connected in order according to the chronological order of the occurrence of the micro-dynamic events to form a continuous and complete pulse data stream, which is the robot's original pulse response stream.

[0076] The amplitude data of all pulse signals in the original impulse response stream are statistically analyzed to determine the maximum and minimum amplitudes. A fixed and uniform standard amplitude range is set, and the conversion ratio between the original pulse amplitude and the standard range is calculated. The amplitude of each pulse signal is adjusted proportionally according to this ratio so that the amplitude of all pulse signals is normalized to the set standard range. The set of pulse signals that maintains the original time sequence and logical relationship after adjustment is the standardized impulse response signal.

[0077] Functional analysis is performed on each pulse in the standardized pulse response signal to identify the core pulse signals that can critically reflect the deviation of the robot's motion state and have important reference value for subsequent control strategy optimization. Invalid pulse signals that are repetitive, redundant, or do not contribute to control decisions are eliminated. While ensuring that the core feedback information is completely preserved, the number and complexity of pulse signals are simplified. The set of pulse signals after screening and optimization is the robot's optimized feedback pulse signal.

[0078] Through the internal signal transmission channel of the robot, the optimized feedback pulse signal is transmitted to the pulse-dependent fusion stage in real time. This feedback signal is then re-integrated with the original pulse feature sequence. Based on the motion state deviation reflected in the feedback signal, key parameters such as the connection strength of the pulse-dependent fusion and the screening criteria for membrane potential integration are dynamically adjusted. This enables subsequent integrated output and drive pulse signals to specifically correct motion errors, ultimately generating new control instructions that can accurately compensate for deviations and improve control precision. These new control instructions are the robot's optimized control instructions.

[0079] The beneficial effects are as follows: by using servo tracking and task space feedforward analysis, the difference between the robot's actual motion state and the target state can be accurately captured, providing a reliable basis for feedback adjustment; micro-dynamic event-based conversion transforms motion details into processable pulse signals, amplitude normalization ensures the consistency of pulse signals, and pulse sparsity eliminates redundant information and improves feedback efficiency; the real-time feedback mechanism builds a closed-loop control system, enabling optimized control commands to dynamically correct motion deviations, significantly improving the adaptability, robustness, and accuracy of robot control, and ensuring that the robot can complete tasks stably and efficiently even in complex scenarios.

[0080] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0081] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A robot control method based on a spiking neural network, characterized by, The method comprises: S1, spatiotemporal feature coding of the robot's environment perception information and ontology state information to obtain the robot's pulse feature sequence; S2, pulse-dependent fusion of the pulse feature sequence and the robot's feedback pulse signal to obtain the robot's bionic information processing process, and membrane potential integration of the bionic information processing process to obtain the robot's integrated output quantity; S3, pulse firing processing of the integrated output quantity to obtain the robot's driving pulse signal; S4, path fitting of the driving pulse signal to obtain the robot's continuous motion trajectory; S5, joint space mapping of the continuous motion trajectory based on the continuous motion trajectory and the bionic information processing process to obtain the robot's execution control instruction; S6, application of the execution control instruction, synchronous acquisition of the robot's pulse response signal, and feedback of the optimized pulse response signal to S2 to obtain the robot's optimized control instruction.

2. The pulse-based neural network-based robot control method of claim 1, wherein, The spatiotemporal feature coding of the robot's environment perception information and ontology state information to obtain the robot's pulse feature sequence comprises: Asynchronous event-driven sampling of the robot to obtain the robot's environment perception information; Event-based sparse representation of the environment perception information to obtain the robot's spatiotemporal event stream; Pose acquisition of the robot to obtain the robot's ontology state information; Joint coordination analysis of the ontology state information to obtain the robot's ontology motion stream; Pulse coding of the robot based on the spatiotemporal event stream and the ontology motion stream to obtain the robot's pulse feature sequence.

3. The pulse neural network based robot control method of claim 1, wherein, The pulse-dependent fusion of the pulse feature sequence and the robot's feedback pulse signal to obtain the robot's bionic information processing process, and the membrane potential integration of the bionic information processing process to obtain the robot's integrated output quantity comprises: Bionic time sequence alignment of the pulse feature sequence and the robot's feedback pulse signal to obtain the robot's synchronous pulse group; Based on the synchronous pulse group, the dependence relationship evolution of the pulse feature sequence is obtained to obtain the connection strength of the pulse feature sequence; Based on the connection strength, the firing behavior of the synchronous pulse group is shaped to obtain the robot's bionic information processing process; Spatiotemporal accumulation of local potential changes in the bionic information processing process to obtain the cumulative membrane potential field of the bionic information processing process; Field domain screening of the cumulative membrane potential field to obtain the gating potential field of the cumulative membrane potential field; Transmembrane phase convergence of the gating potential field to obtain the robot's integrated output quantity.

4. The pulse neural network based robot control method of claim 1, wherein, The pulse firing processing of the integrated output quantity to obtain the robot's driving pulse signal comprises: Membrane potential field construction of the integrated output quantity to obtain the spatial membrane potential field of the integrated output quantity; Bionic evolution of the spatial membrane potential field to obtain the updated membrane potential field of the spatial membrane potential field; Based on the updated membrane potential field, extreme value region detection is performed on the robot to obtain a candidate firing point of the robot; Competitive arbitration is performed on the candidate firing point to obtain a priority firing point of the candidate firing point; Based on the priority firing point, driving output is performed on the pulse firing of the robot to obtain a driving pulse signal of the robot.

5. The pulse neural network based robot control method of claim 4, wherein, The competitive arbitration on the candidate firing point to obtain the priority firing point of the candidate firing point comprises: Based on the updated membrane potential field, the potential amplitude of the candidate firing point is extracted to obtain the instantaneous membrane potential amplitude of the candidate firing point; Based on the bionic information processing process of the robot, the candidate firing point is searched for a historical firing time to obtain a historical firing time of the candidate firing point; The instantaneous membrane potential amplitude and the historical firing time are weighted and evaluated to obtain a firing priority score of the candidate firing point; Based on the firing priority score, global competitive sorting is performed on the candidate firing point to obtain a competitive dominance relationship of the candidate firing point; Based on the competitive dominance relationship, conflict resolution is performed on the candidate firing point to obtain the priority firing point of the candidate firing point.

6. The pulse neural network based robot control method of claim 5, wherein, The calculation formula of the firing priority score is as follows: ； wherein, is the release priority score of the candidate release point position, is the release priority score of the candidate release point position, is the release priority score of the candidate release point position, is the instantaneous membrane potential amplitude, is the historical release time, is the current time, is a preset weight coefficient, is a preset weight coefficient, is a preset time decay coefficient.

7. The pulse neural network based robot control method of claim 1, wherein, The path fitting of the driving pulse signal to obtain the continuous motion trajectory of the robot comprises: The firing timing of the driving pulse signal is mapped to a space-time pose to obtain a discrete space path of an end effector of the robot; The discrete space path of the end effector is subjected to curvature continuous processing to obtain a smooth path segment of the robot; Based on the smooth path segment, the robot is subjected to manifold embedding to obtain a smooth motion manifold of the robot; The smooth motion manifold is subjected to point-by-point fitting of the trajectory to obtain the continuous motion trajectory of the robot.

8. The pulse neural network based robot control method of claim 1, wherein, The joint space mapping of the continuous motion trajectory based on the continuous motion trajectory and the bionic information processing process to obtain the execution control instruction of the robot comprises: The task-related feature decoding of the bionic information processing process is performed to obtain bionic context information of the bionic information processing process; Based on the bionic context information, task semantic segmentation is performed on the continuous motion trajectory to obtain a trajectory segment set of the continuous motion trajectory; The trajectory segment set is marked for importance to obtain a hierarchical trajectory segment sequence of the continuous motion trajectory; Based on the hierarchical trajectory segment sequence, joint space projection is performed on the trajectory segment sequence to obtain a joint space path of the robot; Based on the bionic context information, bionic adjustment is performed on the joint space path to obtain the execution control instruction of the robot.

9. The pulse neural network-based robot control method of claim 8, wherein, The bionic adjustment of the joint space path based on the bionic context information to obtain the execution control instruction of the robot comprises: The bionic context information is subjected to pulse firing space-time analysis to obtain a motion rhythm synchronization signal of the robot; The bionic context information is subjected to joint coordination compilation to obtain an inter-joint coordination pulse code of the robot; Based on the motion rhythm synchronization signal, the joint space path is time axis dynamically scaled to obtain a pulse emission synchronization path of the robot; Based on the inter-joint collaborative pulse coding, a relative amplitude of joint motion in the pulse emission synchronization path is collaboratively gain adjusted to obtain a bionic coupling path of the robot; The bionic coupling path is mapped as an execution control instruction of the robot.

10. The pulse neural network based robot control method of claim 1, wherein, The application of the execution control instruction synchronously collects a pulse response signal of the robot, and feeds back the pulse response signal to S2 to obtain an optimized control instruction of the robot, including: Servo tracking of the execution control instruction obtains an actual joint motion of the robot; Task space feedforward analysis of the execution control instruction obtains an end pose change of the robot; Based on the actual joint motion and the end pose change, a motion state of the robot is micro-dynamic eventized converted to obtain an original pulse response stream of the robot; Amplitude normalization of the original pulse response stream obtains a standardized pulse response signal of the robot; Pulse sparsification of the standardized pulse response signal obtains an optimized feedback pulse signal of the robot; Real-time feedback of the optimized feedback pulse signal obtains the optimized control instruction of the robot.

Images