Parallel Intercoupled Photon Storage Pool System and Method for Robotic Arm Trajectory Prediction

By using a parallel mutually coupled photonic reservoir system and high-dimensional state coding and a bidirectional mutually coupled photonic reservoir module, ultra-low latency and high-precision prediction of robotic arm trajectories were achieved, solving the latency and accuracy problems of traditional methods in high-speed scenarios.

CN123060653APending Publication Date: 2026-11-10WUYI UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611206174.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-10
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing robotic arm trajectory prediction methods are insufficient to meet the real-time requirements of ultra-low latency and high precision in high-speed scenarios. Traditional digital computing methods have high computational overhead and long inference time, while single-node photon reservoir schemes lack the ability to characterize nonlinearities in high-dimensional coupled dynamics prediction.

Method used

A parallel, mutually coupled photonic reservoir system is adopted. The optical injection signal is generated by a high-dimensional state coding module, and a bidirectional, mutually coupled distributed feedback semiconductor laser is used for nonlinear mapping. Combined with the ridge regression algorithm, the prediction of the 8-dimensional dynamic state is achieved.

Benefits of technology

This method achieves ultra-low latency prediction of robotic arm trajectories, reduces system power consumption, improves prediction accuracy and stability, and solves the problem of effective transmission and characterization of high-dimensional coupled dynamic features.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123060653A_ABST
    Figure CN123060653A_ABST
Patent Text Reader

Abstract

This invention provides a parallel-coupled photonic reservoir system and method for predicting the trajectory of a robotic arm. The system consists of three cascaded modules: a high-dimensional state encoding module, a parallel-coupled photonic reservoir module, and an output readout and trajectory prediction module. Leveraging the inherent high-speed parallelism of photonic devices, the inference latency can be reduced to the picosecond to nanosecond level, while significantly reducing power consumption. This meets the ultra-low latency real-time prediction requirements of high-speed robotic arms, effectively overcoming the speed bottleneck of electronic computing power. The bidirectional coupled architecture effectively enhances the system's high-dimensional nonlinear characterization capability, significantly improving prediction accuracy and result stability. Furthermore, the pioneering high-dimensional state encoding method solves the problem of transmission and feature preservation of multivariable coupled dynamic signals in a single optical path, enabling effective prediction of the 8-dimensional dynamic state of the robotic arm by the photonic reservoir at the hardware level.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of photon reservoir computing technology, and in particular to a parallel, mutually coupled photon reservoir system and method for predicting robotic arm trajectories. Background Technology

[0002] With the rapid development of intelligent manufacturing and robotics, multi-degree-of-freedom robotic arms have been widely used in industrial scenarios such as precision assembly, human-robot collaboration, high-speed sorting, and complex dynamic operations. In these applications, high-precision real-time prediction of the robotic arm's end-effector trajectory is the core foundation for achieving high-performance motion planning, dynamic obstacle avoidance, and stable system operation. Currently, mainstream robotic arm control strategies are mainly based on analytical dynamic models, including model predictive control, adaptive control, and sliding mode control. These methods heavily rely on the accuracy of the system's state equations. However, actual robotic arm systems typically exhibit strong nonlinearity, strong coupling of multiple variables, time-varying parameters, and uncertainties caused by external disturbances. This results in high modeling complexity and difficulty in parameter identification. Furthermore, complex online matrix inversion and iterative optimization calculations often incur enormous computational overhead. Traditional digital microprocessors struggle to solve the dynamic equations within microsecond or even sub-microsecond timescales, failing to meet the ultra-low latency control requirements of modern high-speed robotic arms. To overcome the inherent limitations of analytical modeling, data-driven dynamic prediction methods have gradually become a research hotspot. Although methods such as recurrent neural networks, long short-term memory networks, Gaussian mixture models, and digital twin-driven methods have made some progress in trajectory prediction tasks, these methods usually rely on large-scale parameter training and complex backpropagation processes, which have inherent defects such as high computational overhead, large inference latency, and high power consumption, and still cannot meet the real-time requirements of high-speed robotic arms.

[0003] Photon reservoir computing, relying on the extremely high bandwidth and picosecond-level physical response of optical devices, provides a novel hardware implementation path for overcoming the speed and power consumption bottlenecks of digital computing. Currently, existing approximate solutions mainly fall into two categories. One category is robotic arm dynamics prediction schemes based on digital platforms, represented by deep neural networks, Gaussian mixture models, and digital twin technology. These schemes fit the dynamic characteristics of the robotic arm through software algorithms, achieving trajectory prediction on general-purpose CPU or GPU platforms. A typical example is a k-band... The Gaussian mixture model trajectory learning method with value selection and the robot error prediction and compensation method driven by digital twins have acceptable accuracy, but due to the limitations of the clock frequency of electronic devices, the inference latency is generally above the millisecond level, which cannot meet the ultra-low latency requirements of high-speed scenarios. Another type is the photon reservoir scheme based on a single semiconductor laser or one without bidirectional mutual coupling characteristics. These schemes mostly focus on abstract time-series benchmark tasks such as Mackey-Glass and Lorenz, and are mostly in the theoretical simulation stage. They lack real hardware verification for the task of predicting the dynamics of robotic arms, which has a clear engineering background, and there is no dedicated encoding method adapted to the state of high-dimensional coupled robotic arms. They cannot directly and efficiently map the dynamic state of multivariable coupled robotic arms into a single-channel light injection signal. Moreover, the single-node or uncoupled architecture has insufficient nonlinear characterization ability when dealing with high-dimensional strongly coupled dynamic tasks, and there are obvious shortcomings in prediction accuracy and stability. Summary of the Invention

[0004] This invention provides a parallel, mutually coupled photonic reservoir system and method for predicting the trajectory of a robotic arm. Leveraging the inherent high-speed parallel characteristics of photonic devices, inference latency can be reduced to the picosecond to nanosecond level, while system power consumption is significantly reduced. This meets the ultra-low latency real-time prediction requirements of high-speed robotic arms, effectively overcoming the speed bottleneck of electronic computing power. The bidirectional mutually coupled architecture effectively enhances the system's high-dimensional nonlinear characterization capability, significantly improving prediction accuracy and result stability. Furthermore, the pioneering high-dimensional state encoding method solves the problem of transmission and feature preservation of multivariable coupled dynamic signals in a single optical path, enabling effective prediction of the 8-dimensional dynamic state of the robotic arm by the photonic reservoir at the hardware level.

[0005] In a first aspect, embodiments of the present invention provide a parallel, mutually coupled photon reservoir system for predicting robotic arm trajectories, comprising: A high-dimensional state coding module is used to obtain the 8-dimensional state vector of the dual-joint robotic arm, perform dimension-by-dimensional normalization and fixed-order serial tiling on the 8-dimensional state vector to generate a one-dimensional discrete target sequence, and perform time-multiplexing modulation on the one-dimensional discrete target sequence and the chaotic mask signal to obtain an optical injection signal carrying dynamic information. The parallel-coupled photonic reservoir module includes a first distributed feedback semiconductor laser and a second distributed feedback semiconductor laser. Both the first and second distributed feedback semiconductor lasers are configured with independent self-feedback loops, and the first and second distributed feedback semiconductor lasers are connected by a bidirectional mutual coupling link. The optical injection signal is simultaneously injected into the first and second distributed feedback semiconductor lasers to excite nonlinear transient responses and complete the high-dimensional nonlinear mapping of the input features. The output readout and trajectory prediction module is used to collect the output optical signals of the first distributed feedback semiconductor laser and the second distributed feedback semiconductor laser, convert the output optical signals into electrical signals to construct a reservoir state matrix, solve the output weight matrix based on the training set data, and predict the test set data according to the output weight matrix to obtain the 8-dimensional dynamic state and end-effector trajectory of the robotic arm.

[0006] In some embodiments, the high-dimensional state coding module includes: The data generation unit is used to generate reference trajectories based on the rigid body dynamics model of the dual-joint robotic arm. It obtains the joint angle sequence through inverse kinematics and calculates the joint angular velocity and angular acceleration using the five-point center difference method. Substituting these into the dynamic equations, it obtains the joint torque and constructs an 8-dimensional state vector containing 2D joint angles, 2D joint angular velocities, 2D joint angular accelerations, and 2D joint torques. The normalization and tiling unit is used to perform min-max normalization processing on each component in the 8-dimensional state vector, map each component to the [0,1] interval, and then tile the 8-dimensional state components in a fixed order to splice them into the one-dimensional discrete target sequence. The time-multiplexed modulation unit is used to convert the one-dimensional discrete target sequence into an analog voltage signal, perform a dot product operation between the analog voltage signal and the chaotic mask signal through a high-speed multiplier to obtain a time-multiplexed injection signal, and use the time-multiplexed injection signal to drive a Mach-Zehnder phase modulator to modulate the robotic arm state signal onto the light field output by the continuous wave laser to form the light injection signal.

[0007] In some embodiments, in the normalization and tiling unit, the fixed order is to arrange the components of each dimension in the order of joint angle, joint angular velocity, joint angular acceleration, and joint torque; in the time multiplexing modulation unit, the chaotic mask signal is generated by a semiconductor laser, the mask period and the virtual node interval jointly determine the total number of virtual nodes, and the mask period is consistent with the self-feedback delay of the parallel mutually coupled photon reservoir module.

[0008] In some embodiments, the self-feedback loops of the first distributed feedback semiconductor laser and the second distributed feedback semiconductor laser are each composed of an optical fiber delay line, an electrically controlled optical amplifier, an electrically controlled optical attenuator, an optical circulator, and an optical isolator. The bidirectional mutual coupling link is equipped with an optical amplifier and an optical attenuator to adjust the mutual coupling strength; By adjusting the gain of the electro-optical amplifier and the attenuation of the electro-optical attenuator, the self-feedback coefficient and the mutual coupling coefficient are continuously controlled to change the short-term memory effect and nonlinear characterization capability of the reservoir.

[0009] In some embodiments, the output readout and trajectory prediction module includes: The photoelectric conversion and acquisition unit is used to convert the output optical signals of the first distributed feedback semiconductor laser and the second distributed feedback laser into current signals through a photodetector. After being amplified by an electric amplifier, the signals are sampled by a high-speed oscilloscope at virtual node intervals to obtain the virtual node state sequence corresponding to the two lasers. The training unit is used to construct the reservoir state matrix. Each column of the reservoir state matrix is ​​composed of a bias term, an input target value, and the virtual node states of the two lasers. The corresponding target matrix is ​​the target state sequence for the next time step. The ridge regression algorithm with L2 regularization is used to solve the output weight matrix. The prediction unit is used to substitute the state of the reserve pool corresponding to the test set into the output weight matrix obtained by solving to obtain the predicted normalized state sequence. After inverse normalization and dimension reconstruction, it is restored to the 8-dimensional dynamic state of the robotic arm and converted into the Cartesian space trajectory of the end effector through the forward kinematic equation.

[0010] In some embodiments, the regularization coefficient of the ridge regression algorithm in the training unit is 10 to the power of negative 6; in the process of solving the output weight matrix, only the output layer weights are updated, and the internal optical parameters and dynamic characteristics of the first distributed feedback semiconductor laser and the second distributed feedback semiconductor laser remain fixed throughout the process.

[0011] Secondly, embodiments of the present invention also provide a method for predicting the dynamic trajectory of a robotic arm, applied to a parallel, mutually coupled photon reservoir system for predicting robotic arm trajectories as described in the first aspect, the method comprising: Step S1: Obtain the 8-dimensional state vector of the dual-joint robotic arm, and perform dimension-by-dimensional normalization and fixed-order serial tiling on the 8-dimensional state vector to generate a one-dimensional discrete target sequence. Step S2: Time-multiplexed modulation of the one-dimensional discrete target sequence and the chaotic mask signal is performed to obtain an optical injection signal carrying dynamic information; Step S3: The optical injection signal is simultaneously injected into two distributed feedback semiconductor lasers with independent self-feedback loops. The two distributed feedback semiconductor lasers are connected by a bidirectional mutual coupling link. The high-dimensional nonlinear mapping of the input features is completed by utilizing the nonlinear transient response of the distributed feedback semiconductor lasers. Step S4: Collect the output optical signals of the two distributed feedback semiconductor lasers, construct the reservoir state matrix, and solve the output weight matrix based on the training set data using the ridge regression algorithm; Step S5: Substitute the state of the reservoir corresponding to the test set into the output weight matrix obtained by solving to obtain the predicted 8-dimensional dynamic state and end-effector trajectory of the robotic arm.

[0012] In some embodiments, step S1 includes: A reference trajectory is generated based on the rigid body dynamics model of a dual-joint robotic arm, and the reference trajectory includes a chaotic attractor trajectory and a periodic geometric trajectory. The joint angle sequence is obtained by solving the inverse kinematics, and the joint angular velocity and angular acceleration are calculated by the five-point center difference method. The joint torque is obtained by substituting them into the dynamic equation. An 8-dimensional state vector containing 2D joint angle, 2D joint angular velocity, 2D joint angular acceleration and 2D joint torque is constructed and divided into training set and test set in 8:2 ratio according to time sequence. Min-max normalization is performed on each component of the 8-dimensional state vector to map each component to the [0,1] interval. Then, the 8-dimensional state components are serially laid out in a fixed order of joint angle, joint angular velocity, joint angular acceleration, and joint torque to form the one-dimensional discrete target sequence.

[0013] In some embodiments, step S4 includes: The output optical signals of two distributed feedback semiconductor lasers are converted into current signals by a photodetector, amplified by an amplifier, and then sampled by a high-speed oscilloscope at virtual node intervals to obtain the virtual node state sequence. Construct a state matrix for the reservoir. Each column of the state matrix is ​​composed of a bias term, an input target value, and the virtual node states of the two lasers. The corresponding target matrix is ​​the target state sequence for the next time step. The output weight matrix is ​​solved using a ridge regression algorithm with L2 regularization. The regularization coefficient is set to 10 to the power of -6. During the training process, the optical parameters and dynamic characteristics inside the reservoir are kept fixed throughout, and only the output layer weights are updated.

[0014] Thirdly, embodiments of the present invention also provide a computer-readable storage medium storing computer-executable instructions for performing the robotic arm dynamic trajectory prediction method as described in the second aspect.

[0015] According to embodiments of the present invention, a parallel-coupled photonic reservoir system and method for predicting robotic arm trajectories are provided. The parallel-coupled photonic reservoir system for predicting robotic arm trajectories is composed of three cascaded modules: a high-dimensional state encoding module, a parallel-coupled photonic reservoir module, and an output readout and trajectory prediction module. The high-dimensional state encoding module acquires the 8-dimensional state vector of the dual-joint robotic arm, performs dimension-by-dimensional normalization and fixed-order serial tiling on the 8-dimensional state vector to generate a one-dimensional discrete target sequence, and performs time-multiplexed modulation of the one-dimensional discrete target sequence with a chaotic mask signal to obtain a light injection signal carrying dynamic information. The parallel-coupled photonic reservoir module includes a first distributed feedback semiconductor laser and a second distributed feedback semiconductor laser. Both the first and second distributed feedback semiconductor lasers are equipped with independent self-feedback loops, and are connected via a bidirectional mutual-coupled link. An optical injection signal is simultaneously injected into both lasers to excite nonlinear transient responses, completing a high-dimensional nonlinear mapping of the input features. The output readout and trajectory prediction module acquires the output optical signals from both lasers, converts them into electrical signals, constructs a reservoir state matrix, solves the output weight matrix based on training data, and predicts the test data using the output weight matrix to obtain the 8-dimensional dynamic state and end-effector trajectory of the robotic arm. This system is a parallel photon reservoir system based on mutually coupled self-feedback distributed feedback semiconductor lasers. It utilizes the high bandwidth and low latency inherent in optical devices to achieve ultra-low latency prediction of high-speed robotic arm dynamics, breaking through the speed bottleneck of traditional digital computing. This invention addresses the challenge of effectively transmitting and preserving the 8-dimensional multivariable coupled dynamic characteristics of a robotic arm in a single-link optical system by employing a method of dimension-by-dimensional normalization of the 8-dimensional state vector, fixed-order serial tiling, and time-multiplexed mask encoding. A bidirectional, mutually coupled optical reservoir architecture is adopted, introducing nonlinear dynamic interaction across reservoirs to enhance the system's ability to characterize high-dimensional coupled dynamics, effectively reducing prediction errors and improving result stability. Based on this, the embodiments of this invention can achieve effective prediction of the 8-dimensional dynamic state of a robotic arm using a photon reservoir. Attached Figure Description

[0016] Figure 1 This is a block diagram of a parallel, mutually coupled photon reservoir system for predicting robotic arm trajectories, provided in one embodiment of the present invention. Figure 2 This is a diagram of a hardware experimental device for a parallel mutually coupled photon reservoir provided in one embodiment of the present invention; Figure 3 This is a diagram showing the predicted motion trajectory of a dual-joint robotic arm according to an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the following drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0019] In this embodiment of the invention, the terms "furthermore," "exemplarily," or "optionally" are used as examples, illustrations, or descriptions and should not be construed as being more preferred or advantageous than other embodiments or designs. The use of the terms "furthermore," "exemplarily," or "optionally" is intended to present the relevant concepts in a specific manner.

[0020] First, let's analyze some of the terms used in this invention: Reservoir Computing (RC): A lightweight recurrent neural network architecture that reduces training complexity while maintaining strong nonlinear expressive power by fixing the high-dimensional dynamic system and training only the linear output layer.

[0021] Photonic Reservoir Computing (PRC): A reservoir system built on the GHz-level high bandwidth, picosecond-level low latency physical response and natural parallel computing characteristics of optical devices, used to overcome the speed and power consumption bottlenecks of digital computing.

[0022] Mutually Coupled DFB-SL: Two short-cavity distributed feedback semiconductor lasers with independent self-feedback branches interact dynamically across the reservoir via a bidirectional mutual coupling link to enhance the system's high-dimensional nonlinear characterization capability.

[0023] Mask signal: A periodic signal used to achieve time multiplexing. In this system, a chaotic sequence generated by a semiconductor laser is used as a mask to act as a high-dimensional nonlinear scatterer, mapping the one-dimensional input signal to a high-dimensional virtual node space.

[0024] Time-multiplexing: a technique that converts a one-dimensional discrete signal into a continuous-time optical signal through mask modulation. The mask period T and the virtual node spacing θ determine the total number of virtual nodes N=T / θ.

[0025] Normalized mean square error (NMSE): The core indicator for evaluating the accuracy of prediction results. NMSE=0 indicates that the prediction is completely consistent with the target, and NMSE<0.1 indicates that the system can achieve high-precision and effective prediction.

[0026] 8-dimensional robotic arm state vector: The complete dynamic state of a dual-joint robotic arm, which includes eight mutually coupled state components: 2-dimensional joint angle, 2-dimensional joint angular velocity, 2-dimensional joint angular acceleration, and 2-dimensional joint torque.

[0027] To facilitate a more convenient description of the working principle of the embodiments of the present invention, the following introduction of relevant technical scenarios is given first.

[0028] With the rapid development of intelligent manufacturing and robotics, multi-degree-of-freedom robotic arms have been widely used in industrial scenarios such as precision assembly, human-robot collaboration, high-speed sorting, and complex dynamic operations. In these applications, high-precision real-time prediction of the robotic arm's end-effector trajectory is the core foundation for achieving high-performance motion planning, dynamic obstacle avoidance, and stable system operation. Currently, mainstream robotic arm control strategies are mainly based on analytical dynamic models, including model predictive control, adaptive control, and sliding mode control. These methods heavily rely on the accuracy of the system's state equations. However, actual robotic arm systems typically exhibit strong nonlinearity, strong coupling of multiple variables, time-varying parameters, and uncertainties caused by external disturbances. This results in high modeling complexity and difficulty in parameter identification. Furthermore, complex online matrix inversion and iterative optimization calculations often incur enormous computational overhead. Traditional digital microprocessors struggle to solve the dynamic equations within microsecond or even sub-microsecond timescales, failing to meet the ultra-low latency control requirements of modern high-speed robotic arms. To overcome the inherent limitations of analytical modeling, data-driven dynamic prediction methods have gradually become a research hotspot. Although methods such as recurrent neural networks, long short-term memory networks, Gaussian mixture models, and digital twin-driven methods have made some progress in trajectory prediction tasks, these methods usually rely on large-scale parameter training and complex backpropagation processes, which have inherent defects such as high computational overhead, large inference latency, and high power consumption, and still cannot meet the real-time requirements of high-speed robotic arms.

[0029] Photon reservoir computing, relying on the extremely high bandwidth and picosecond-level physical response of optical devices, provides a novel hardware implementation path for overcoming the speed and power consumption bottlenecks of digital computing. Currently, existing approximate solutions mainly fall into two categories. One category is robotic arm dynamics prediction schemes based on digital platforms, represented by deep neural networks, Gaussian mixture models, and digital twin technology. These schemes fit the dynamic characteristics of the robotic arm through software algorithms, achieving trajectory prediction on general-purpose CPU or GPU platforms. A typical example is a k-band... The Gaussian mixture model trajectory learning method with value selection and the robot error prediction and compensation method driven by digital twins have acceptable accuracy, but due to the limitations of the clock frequency of electronic devices, the inference latency is generally above the millisecond level, which cannot meet the ultra-low latency requirements of high-speed scenarios. Another type is the photon reservoir scheme based on a single semiconductor laser or one without bidirectional mutual coupling characteristics. These schemes mostly focus on abstract time-series benchmark tasks such as Mackey-Glass and Lorenz, and are mostly in the theoretical simulation stage. They lack real hardware verification for the task of predicting the dynamics of robotic arms, which has a clear engineering background, and there is no dedicated encoding method adapted to the state of high-dimensional coupled robotic arms. They cannot directly and efficiently map the dynamic state of multivariable coupled robotic arms into a single-channel light injection signal. Moreover, the single-node or uncoupled architecture has insufficient nonlinear characterization ability when dealing with high-dimensional strongly coupled dynamic tasks, and there are obvious shortcomings in prediction accuracy and stability.

[0030] Based on this, the present invention provides a parallel-coupled photonic reservoir system and method for predicting the trajectory of a robotic arm. The parallel-coupled photonic reservoir system for predicting the trajectory of a robotic arm consists of three cascaded modules: a high-dimensional state encoding module, a parallel-coupled photonic reservoir module, and an output readout and trajectory prediction module. The high-dimensional state encoding module acquires the 8-dimensional state vector of the dual-joint robotic arm, performs dimension-by-dimensional normalization and fixed-order serial tiling on the 8-dimensional state vector to generate a one-dimensional discrete target sequence, and then performs time-multiplexing modulation of the one-dimensional discrete target sequence with a chaotic mask signal to obtain a light injection signal carrying dynamic information. The parallel-coupled photonic reservoir module includes a first distributed feedback semiconductor laser and a second distributed feedback semiconductor laser. Each device is equipped with an independent self-feedback loop, and the first and second distributed feedback semiconductor lasers are connected via a bidirectional mutual-coupled link. An optical injection signal is simultaneously injected into both the first and second distributed feedback semiconductor lasers to excite nonlinear transient responses, completing a high-dimensional nonlinear mapping of the input features. The output readout and trajectory prediction module collects the output optical signals from the first and second distributed feedback semiconductor lasers, converts the output optical signals into electrical signals, constructs a reservoir state matrix, solves the output weight matrix based on the training set data, and predicts the test set data based on the output weight matrix to obtain the 8-dimensional dynamic state and end-effector trajectory of the robotic arm. This system is a parallel photonic reservoir system based on mutually coupled self-feedback distributed feedback semiconductor lasers. It utilizes the high bandwidth and low latency inherent characteristics of optical devices to achieve ultra-low latency prediction of high-speed robotic arm dynamics, breaking through the speed bottleneck of traditional digital computing. By performing dimension-by-dimensional normalization of the 8-dimensional state vector, fixed-order serial tiling, and time-multiplexed mask encoding, it solves the problem of effective transmission and feature preservation of the 8-dimensional multivariable coupled dynamic features of the robotic arm in a single-link optical system. By employing a bidirectional, mutually coupled optical reservoir architecture and introducing nonlinear dynamic interactions across the reservoirs, the system's ability to characterize high-dimensional coupled dynamics is enhanced, effectively reducing prediction errors and improving result stability. Based on this, embodiments of the present invention can achieve effective prediction of the 8-dimensional dynamic state of a robotic arm using a photon reservoir.

[0031] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0032] like Figure 1 As shown, Figure 1This is a block diagram of a parallel-coupled photonic reservoir system for robotic arm trajectory prediction according to an embodiment of the present invention. EA is an electrical amplifier, DM is a discrete processing module, SC is a proportional circuit, PM is a phase modulator, CW Laser is a continuous laser, and DFB-SL is a short-cavity DFB laser. The parallel-coupled photonic reservoir system for robotic arm trajectory prediction of the present invention is composed of three cascaded modules: a high-dimensional state encoding module, a parallel-coupled photonic reservoir module, and an output readout and trajectory prediction module. The core working principle of the system is to first convert the 8-dimensional coupled dynamic state of the dual-joint robotic arm into a single-channel modulated signal that can drive the optical system through encoding, then inject it into two bidirectional mutually coupled distributed feedback semiconductor lasers with self-feedback. The nonlinear transient response of the lasers is used to complete the high-dimensional feature mapping. Finally, the laser output state is collected and linear readout weights are trained to predict the robotic arm state and end-effector trajectory at the next moment.

[0033] Understandably, this system is a parallel, mutually coupled photonic reservoir computing system for predicting the 8-dimensional coupled dynamic states of a dual-joint robotic arm. It establishes a reproducible experimental coupling framework between the robotic arm engineering scenario and the photonic reservoir hardware. By performing dimension-by-dimensional normalization, fixed-order serial tiling, and time-multiplexed mask encoding on the 8-dimensional state vector, it achieves efficient mapping from the 8-dimensional coupled robotic arm state to the single-channel optical injection signal, completely preserving the dynamic coupling characteristics. Based on a reservoir architecture design with two self-feedback distributed feedback lasers bidirectionally mutually coupled, the system enhances the high-dimensional dynamic characterization capability through cross-reservoir nonlinear dynamic interaction, reducing the average prediction error by 4.7% compared to a single-reservoir structure.

[0034] Understandably, the technical implementation of the high-dimensional state coding module is as follows: First, a reference trajectory is generated based on the rigid body dynamics model of the dual-joint robotic arm. The trajectory types cover two main categories: chaotic attractor trajectories and periodic geometric trajectories. The joint angle sequence is obtained through inverse kinematics. Then, the joint angular velocity and angular acceleration are calculated using the five-point central difference method. Finally, the joint torque is obtained by substituting these into the dynamic equations. This constructs an 8-dimensional state vector containing two joint angles, two joint angular velocities, two joint angular accelerations, and two joint torques. All samples are strictly divided into training and test sets in an 8:2 ratio according to time sequence to preserve temporal causality and avoid data leakage. Subsequently, min-max normalization is performed on each of the eight state components, mapping each component uniformly to the [0,1] interval. Then, the eight-dimensional state components are serially laid out and spliced ​​according to a fixed order. The target sequence is a one-dimensional discrete sequence, preserving the temporal characteristics and coupling relationships of each state component. Next, the one-dimensional discrete sequence is converted into an analog voltage signal by an arbitrary waveform generator. This signal is then multiplied with a periodic chaotic mask signal using a high-speed multiplier to obtain a time-multiplexed injection signal. The input scaling factor is fixed at 1.0 to match the signal amplitude to the linear operating range of the phase modulator. The total number of virtual nodes is determined by the mask period and the virtual node interval. The mask period is consistent with the reservoir feedback delay. In a preferred embodiment, the mask period is 30 nanoseconds, the virtual node interval is 0.1 nanoseconds, and the corresponding number of virtual nodes is 300. Finally, the masked electrical signal drives a Mach-Zehnder phase modulator to modulate the robotic arm's state signal onto the optical field output by a continuous-wave laser, forming an injection optical signal carrying complete dynamic information.

[0035] In one embodiment, the high-dimensional state encoding module may include a data generation unit, a normalization and tiling unit, and a time multiplexing modulation unit. The data generation unit generates a reference trajectory based on the rigid body dynamics model of the dual-joint robotic arm, obtains the joint angle sequence through inverse kinematics, calculates the joint angular velocity and angular acceleration using the five-point center difference method, substitutes these values ​​into the dynamic equations to obtain the joint torque, and constructs an 8-dimensional state vector containing 2D joint angles, 2D joint angular velocities, 2D joint angular accelerations, and 2D joint torques. The normalization and tiling unit performs min-max normalization on each component of the 8-dimensional state vector, mapping each component to the [0,1] interval, and then sequentially tiles the 8-dimensional state components in a fixed order, splicing them into a one-dimensional discrete target sequence. The time-multiplexing modulation unit converts a one-dimensional discrete target sequence into an analog voltage signal. A high-speed multiplier performs a dot product operation between the analog voltage signal and the chaotic mask signal to obtain a time-multiplexed injection signal. This signal drives a Mach-Zehnder phase modulator to modulate the robotic arm's state signal onto the optical field output by a continuous-wave laser, forming an optical injection signal. In the normalization and tiling unit, the components are arranged in a fixed order: joint angle, joint angular velocity, joint angular acceleration, and joint torque. In the time-multiplexing modulation unit, the chaotic mask signal is generated by a semiconductor laser. The mask period and the virtual node spacing jointly determine the total number of virtual nodes, and the mask period is consistent with the self-feedback delay of the parallel-coupled photon reservoir module.

[0036] Understandably, the parallel-coupled photonic reservoir module is the core computing unit of the system. It consists of two structurally symmetrical self-feedback distributed feedback semiconductor lasers connected by a bidirectional mutual coupling link. Each laser is equipped with an independent self-feedback loop, which consists of an optical fiber delay line, an electrically controlled optical amplifier, an electrically controlled optical attenuator, an optical circulator, and an optical isolator. By adjusting the gain of the optical amplifier and the attenuation of the optical attenuator, the self-feedback optical power can be continuously controlled, thereby changing the self-feedback coefficient. The self-feedback branch introduces a short-term memory effect into the reservoir, which is the physical basis for the reservoir's time-series processing capability. The two lasers are mutually coupled through a bidirectional optical fiber link. Each coupling branch is also equipped with an optical amplifier and an optical attenuator for continuously controlling the mutual coupling strength and the corresponding mutual coupling coefficient. The mutual coupling mechanism enables the nonlinear dynamics of the two lasers to influence and interact with each other, which can significantly improve the reservoir's high-dimensional feature representation capability and noise suppression capability. During operation, modulated light carrying the input signal is simultaneously injected into two distributed feedback lasers, exciting the lasers to generate nonlinear transient responses in the carrier concentration and optical field. Under the combined effect of self-feedback delay and mutual coupling, the low-dimensional input signal is mapped to a high-dimensional nonlinear state space, completing the nonlinear extension of the characteristics. The output optical signals of the two lasers together constitute the state output of the reservoir. In a preferred embodiment, the system's baseline operating parameters are set to a self-feedback coefficient of 0.1, an optical injection coefficient of 0.1, and a mutual coupling coefficient of 0.05 for both lasers. The operating temperature of the two lasers is stably controlled at approximately 28 degrees Celsius, with a center wavelength of approximately 1550 nanometers.

[0037] In one embodiment, the parallel-coupled photonic reservoir module includes a first distributed feedback semiconductor laser and a second distributed feedback semiconductor laser. The self-feedback loops of both the first and second distributed feedback semiconductor lasers are composed of an optical fiber delay line, an electrically controlled optical amplifier, an electrically controlled optical attenuator, an optical circulator, and an optical isolator. An optical amplifier and an optical attenuator are configured in the bidirectional mutual coupling link to adjust the mutual coupling strength. By adjusting the gain of the electrically controlled optical amplifier and the attenuation of the electrically controlled optical attenuator, the self-feedback coefficient and the mutual coupling coefficient are continuously controlled to change the short-term memory effect and nonlinear characterization capability of the reservoir.

[0038] Understandably, the technical implementation of the output readout and trajectory prediction module is as follows: the output optical signals of the two distributed feedback lasers are converted into current signals by photodetectors, then amplified to a suitable amplitude by an amplifier, and finally the data is acquired by a high-bandwidth, high-speed oscilloscope. The acquired voltage signals are sampled at virtual node intervals to obtain the virtual node state sequences corresponding to the two lasers. Subsequently, a reservoir state matrix is ​​constructed. Each column of the matrix consists of a bias term, the input target value, and the virtual node states of the two lasers. The corresponding target matrix is ​​the target state sequence for the next time step. The ridge regression algorithm with L2 regularization is used to solve the output weight matrix. The preferred value of the regularization coefficient is 10 to the power of -6. During the training process, only the output layer weights are updated. The optical parameters and dynamic characteristics inside the reservoir are fixed throughout the process, which conforms to the core design principle of reservoir calculation. After training, the reservoir states of the test set are substituted into the solved weight matrix to obtain the predicted normalized state sequence. Then, after inverse normalization and dimensional reconstruction, it is restored to the 8-dimensional dynamic state of the robotic arm. Finally, it is converted into the Cartesian space trajectory of the end effector through the forward kinematic equation.

[0039] It is understood that the present invention adopts a ridge regression linear readout training method based on time-series partitioning, strictly follows dynamic causality to avoid data leakage, and at the same time ensures the numerical stability of weight solution.

[0040] In one embodiment, the output readout and trajectory prediction module may include a photoelectric conversion and acquisition unit, a training unit, and a prediction unit. The photoelectric conversion and acquisition unit converts the output optical signals of the first distributed feedback semiconductor laser and the second distributed feedback laser into current signals using a photodetector. After amplification by an amplifier, the signals are sampled at virtual node intervals using a high-speed oscilloscope to obtain the virtual node state sequences corresponding to the two lasers. The training unit constructs a reservoir state matrix. Each column of the reservoir state matrix consists of a bias term, an input target value, and the virtual node states of the two lasers. The corresponding target matrix is ​​the target state sequence for the next time step. The output weight matrix is ​​solved using a ridge regression algorithm with L2 regularization. The prediction unit substitutes the reservoir states corresponding to the test set into the solved output weight matrix to obtain the predicted normalized state sequence. After inverse normalization and dimensional reconstruction, the sequence is restored to an 8-dimensional robotic arm dynamic state and converted into the Cartesian space trajectory of the end effector using forward kinematic equations.

[0041] Understandable, Figure 2This diagram shows the hardware experimental setup for a parallel-coupled photonic reservoir. All components are built using commercially available standard photonics and electronics, ensuring high reproducibility. In the diagram, P represents a computer, AWG an arbitrary waveform generator, DL an optical fiber delay line, OA an optical amplifier, VOA an optical attenuator, CIR a circulator, FBS / FC an optical fiber beam splitter, MC a modular controller, represents the injection current, TE represents temperature, IS a isolator, PM a power meter, PD a photodetector, DFB a distributed feedback semiconductor laser, EA an electrical amplifier, Mask a mask signal, MUL a multiplier, MZM a phase modulator, CW a continuous wave laser, and OSC an oscilloscope. The system is divided into two parts: optical path and circuit. The optical path is centered around two distributed feedback lasers, equipped with laser controllers for precise control of the laser injection current and operating temperature. A continuous wave laser serves as the carrier source for optical injection, and a Mach-Zehnder modulator performs electro-optic modulation. Each branch is equipped with fiber delay lines, electrically controlled optical amplifiers, and electrically controlled optical attenuators to adjust delay and optical power. An optical circulator guides the optical path, an optical isolator prevents light reflection from interfering with the stable operation of the laser, and a fiber beam splitter splits and combines the optical signals. The circuit includes a computer for data generation and subsequent processing, two arbitrary waveform generators to generate input state signals and mask signals respectively, a high-speed multiplier based on a field-programmable gate array (FPGA) to perform dot product operations on the input and mask signals, a photodetector for photoelectric conversion, an amplifier to amplify the detected weak electrical signals, and a high-speed oscilloscope for final signal acquisition and recording. The specific operating parameters of the two distributed feedback lasers are shown in Table 1.

[0042] Table 1: Operating parameters of two mutually coupled distributed feedback lasers

[0043] Furthermore, an embodiment of the present invention discloses a method for predicting the dynamic trajectory of a robotic arm, applied to the aforementioned parallel-coupled photon reservoir system for predicting robotic arm trajectories. This method for predicting the dynamic trajectory of a robotic arm includes, but is not limited to, the following steps: Step S1: Obtain the 8-dimensional state vector of the dual-joint robotic arm, perform dimension-by-dimensional normalization and fixed-order serial tiling on the 8-dimensional state vector to generate a one-dimensional discrete target sequence. Step S2: Time-multiplexed modulation of the one-dimensional discrete target sequence and the chaotic mask signal is performed to obtain an optical injection signal carrying dynamic information; Step S3: Simultaneously inject the optical injection signal into two distributed feedback semiconductor lasers with independent self-feedback loops. The two distributed feedback semiconductor lasers are connected through a bidirectional mutual coupling link. The high-dimensional nonlinear mapping of the input features is completed by utilizing the nonlinear transient response of the distributed feedback semiconductor lasers. Step S4: Collect the output optical signals of two distributed feedback semiconductor lasers, construct the reservoir state matrix, and solve the output weight matrix based on the training set data using the ridge regression algorithm; Step S5: Substitute the state of the reservoir corresponding to the test set into the output weight matrix obtained by solving to obtain the predicted 8-dimensional dynamic state and end-effector trajectory of the robotic arm.

[0044] Understandably, the robotic arm dynamic trajectory prediction method of this invention is executed according to the following process: First, a dual-joint robotic arm dynamic dataset is constructed, generating an 8-dimensional state sequence containing joint angles, angular velocities, angular accelerations, and torques. The training set and test set are divided according to time sequence. Second, the 8-dimensional states are normalized dimension by dimension and then serially laid out in a fixed order to obtain a one-dimensional target sequence. This one-dimensional target sequence is then modulated with a chaotic mask signal using time multiplexing and loaded onto the light field to complete the optical signal encoding of the high-dimensional state. Next, the encoded injected optical signal is input into a parallel mutually coupled photon reservoir to excite the nonlinear transient response of two lasers, completing the high-dimensional nonlinear mapping of the input features. Then, the output optical signals of the two lasers are collected, converted into electrical signals, and the reservoir state matrix is ​​constructed by sampling according to virtual nodes. The output weight matrix is ​​solved using the ridge regression algorithm based on the training set data. Finally, the reservoir state of the test set is substituted into the trained weight matrix to obtain the predicted state sequence. After inverse normalization and dimensional reconstruction, the predicted state and end-effector trajectory of the robotic arm are restored, completing the complete prediction task.

[0045] Understandably, in step S1, a reference trajectory is generated based on the rigid body dynamics model of the dual-joint robotic arm. The reference trajectory includes a chaotic attractor trajectory and a periodic geometric trajectory. The joint angle sequence is obtained by solving inverse kinematics, and the joint angular velocity and angular acceleration are calculated using the five-point central difference method. The joint torque is obtained by substituting them into the dynamic equation. An 8-dimensional state vector containing 2D joint angle, 2D joint angular velocity, 2D joint angular acceleration and 2D joint torque is constructed and divided into a training set and a test set in an 8:2 ratio according to the time sequence. Min-max normalization is performed on each component in the 8-dimensional state vector to map each component to the [0,1] interval. Then, the 8-dimensional state components are serially laid out in a fixed order of joint angle, joint angular velocity, joint angular acceleration and joint torque, and spliced ​​into a one-dimensional discrete target sequence.

[0046] Understandably, in step S4, the output optical signals of the two distributed feedback semiconductor lasers are converted into current signals by a photodetector, amplified by an amplifier, and then sampled by a high-speed oscilloscope at virtual node intervals to obtain a virtual node state sequence. A reservoir state matrix is ​​constructed, and each column of the reservoir state matrix is ​​composed of a bias term, an input target value, and the virtual node states of the two lasers. The corresponding target matrix is ​​the target state sequence for the next time step. The ridge regression algorithm with L2 regularization is used to solve the output weight matrix, with the regularization coefficient set to 10 to the power of -6. During the training process, the internal optical parameters and dynamic characteristics of the reservoir are fixed throughout, and only the output layer weights are updated.

[0047] Understandable, Figure 3 The figures show the predicted motion trajectory results of the dual-joint robotic arm, used to visually verify the prediction effect of the system. Subfigure (a) from left to right compares the original target trajectory in Cartesian space at the end of the robotic arm with the system's predicted trajectory under three typical working conditions: Lorenz chaotic trajectory, Chen chaotic trajectory, and Lissajous regular geometric trajectory. It can be observed that the predicted results of the three types of trajectories highly match the original target trajectories in terms of overall shape, key turning points, and long-term evolution trends, without significant trajectory divergence or phase shift, verifying the system's universal prediction capability for trajectories with different dynamic characteristics. Subfigure (b) shows the time series comparison of the 8-dimensional normalized joint state variables corresponding to the Lorenz chaotic trajectory, covering 2D joint angle, 2D angular velocity, 2D angular acceleration, and 2D joint torque, a total of 8 coupled state components. The predicted curves of each component almost completely overlap with the actual curves, further proving that the system can not only predict the end-effector spatial trajectory but also accurately reconstruct the complete high-dimensional coupled dynamic state of the robotic arm, with prediction accuracy meeting engineering application requirements.

[0048] It is understood that the parallel intercoupled photon reservoir system and method for predicting robotic arm trajectories of the present invention can be applied to industrial robotic arm trajectory prediction, high-speed robot motion control, intelligent manufacturing dynamic obstacle avoidance, and high-speed time-series signal processing.

[0049] Based on this, compared with existing robotic arm dynamics prediction methods based on digital platforms, this invention relies on the inherent high-speed parallel characteristics of photonic devices, reducing inference latency to the picosecond to nanosecond level, which is 3 to 4 orders of magnitude higher than digital algorithms. At the same time, the system power consumption is significantly reduced, which can meet the ultra-low latency real-time prediction requirements of high-speed robotic arms, effectively breaking through the speed bottleneck of electronic computing power, and providing a brand-new hardware implementation path for dynamics prediction in high-speed industrial scenarios.

[0050] Compared to existing single-cavity, uncoupled photon reservoirs, the bidirectional mutually coupled architecture proposed in this invention effectively enhances the system's high-dimensional nonlinear characterization capability, significantly improving both prediction accuracy and result stability. Simultaneously, the pioneering high-dimensional state encoding method solves the problem of transmission and feature preservation of multivariable coupled dynamic signals in a single optical path, achieving for the first time at the hardware level effective prediction of the 8-dimensional dynamic state of a robotic arm using a photon reservoir. Furthermore, the system maintains stable performance under a wide range of optical parameter perturbations and significant shifts in robotic arm structural parameters, demonstrating excellent parameter robustness and scenario generalization, making it more adaptable to complex and ever-changing real-world industrial scenarios.

[0051] In addition, one embodiment of the present invention discloses a computer-readable storage medium storing computer-executable instructions for performing the robotic arm dynamic trajectory prediction method as described in any of the preceding embodiments.

[0052] The above provides a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A parallel, mutually coupled photon reservoir system for predicting robotic arm trajectories, characterized in that, include: A high-dimensional state coding module is used to obtain the 8-dimensional state vector of the dual-joint robotic arm, perform dimension-by-dimensional normalization and fixed-order serial tiling on the 8-dimensional state vector to generate a one-dimensional discrete target sequence, and perform time-multiplexing modulation on the one-dimensional discrete target sequence and the chaotic mask signal to obtain an optical injection signal carrying dynamic information. The parallel-coupled photonic reservoir module includes a first distributed feedback semiconductor laser and a second distributed feedback semiconductor laser. Both the first and second distributed feedback semiconductor lasers are configured with independent self-feedback loops, and the first and second distributed feedback semiconductor lasers are connected by a bidirectional mutual coupling link. The optical injection signal is simultaneously injected into the first and second distributed feedback semiconductor lasers to excite nonlinear transient responses and complete the high-dimensional nonlinear mapping of the input features. The output readout and trajectory prediction module is used to collect the output optical signals of the first distributed feedback semiconductor laser and the second distributed feedback semiconductor laser, convert the output optical signals into electrical signals to construct a reservoir state matrix, solve the output weight matrix based on the training set data, and predict the test set data according to the output weight matrix to obtain the 8-dimensional dynamic state and end-effector trajectory of the robotic arm.

2. The parallel mutually coupled photon reservoir system for robotic arm trajectory prediction according to claim 1, characterized in that, The high-dimensional state coding module includes: The data generation unit is used to generate reference trajectories based on the rigid body dynamics model of the dual-joint robotic arm. It obtains the joint angle sequence through inverse kinematics and calculates the joint angular velocity and angular acceleration using the five-point center difference method. Substituting these into the dynamic equations, it obtains the joint torque and constructs an 8-dimensional state vector containing 2D joint angles, 2D joint angular velocities, 2D joint angular accelerations, and 2D joint torques. The normalization and tiling unit is used to perform min-max normalization processing on each component in the 8-dimensional state vector, map each component to the [0,1] interval, and then tile the 8-dimensional state components in a fixed order to splice them into the one-dimensional discrete target sequence. The time-multiplexed modulation unit is used to convert the one-dimensional discrete target sequence into an analog voltage signal, perform a dot product operation between the analog voltage signal and the chaotic mask signal through a high-speed multiplier to obtain a time-multiplexed injection signal, and use the time-multiplexed injection signal to drive a Mach-Zehnder phase modulator to modulate the robotic arm state signal onto the light field output by the continuous wave laser to form the light injection signal.

3. The parallel mutually coupled photon reservoir system for robotic arm trajectory prediction according to claim 2, characterized in that, In the normalization and tiling unit, the fixed order is to arrange the components of each dimension in the order of joint angle, joint angular velocity, joint angular acceleration, and joint torque; in the time multiplexing modulation unit, the chaotic mask signal is generated by a semiconductor laser, the mask period and the virtual node interval jointly determine the total number of virtual nodes, and the mask period is consistent with the self-feedback delay of the parallel mutually coupled photon reservoir module.

4. The parallel mutually coupled photon reservoir system for robotic arm trajectory prediction according to claim 1, characterized in that, The self-feedback loops of both the first and second distributed feedback semiconductor lasers consist of an optical fiber delay line, an electrically controlled optical amplifier, an electrically controlled optical attenuator, an optical circulator, and an optical isolator. The bidirectional mutual coupling link is equipped with an optical amplifier and an optical attenuator to adjust the mutual coupling strength; By adjusting the gain of the electro-optical amplifier and the attenuation of the electro-optical attenuator, the self-feedback coefficient and the mutual coupling coefficient are continuously controlled to change the short-term memory effect and nonlinear characterization capability of the reservoir.

5. The parallel mutually coupled photon reservoir system for robotic arm trajectory prediction according to claim 1, characterized in that, The output readout and trajectory prediction module includes: The photoelectric conversion and acquisition unit is used to convert the output optical signals of the first distributed feedback semiconductor laser and the second distributed feedback laser into current signals through a photodetector. After being amplified by an electric amplifier, the signals are sampled by a high-speed oscilloscope at virtual node intervals to obtain the virtual node state sequence corresponding to the two lasers. The training unit is used to construct the reservoir state matrix. Each column of the reservoir state matrix is ​​composed of a bias term, an input target value, and the virtual node states of the two lasers. The corresponding target matrix is ​​the target state sequence for the next time step. The ridge regression algorithm with L2 regularization is used to solve the output weight matrix. The prediction unit is used to substitute the state of the reserve pool corresponding to the test set into the output weight matrix obtained by solving to obtain the predicted normalized state sequence. After inverse normalization and dimension reconstruction, it is restored to the 8-dimensional dynamic state of the robotic arm and converted into the Cartesian space trajectory of the end effector through the forward kinematic equation.

6. The parallel mutually coupled photon reservoir system for robotic arm trajectory prediction according to claim 5, characterized in that, In the training unit, the regularization coefficient of the ridge regression algorithm is 10 to the power of negative 6; in the process of solving the output weight matrix, only the output layer weights are updated, and the internal optical parameters and dynamic characteristics of the first distributed feedback semiconductor laser and the second distributed feedback semiconductor laser remain fixed throughout the process.

7. A method for predicting the dynamic trajectory of a robotic arm, characterized in that, The method, applied to the parallel inter-coupled photon reservoir system for robotic arm trajectory prediction as described in any one of claims 1 to 6, comprises: Step S1: Obtain the 8-dimensional state vector of the dual-joint robotic arm, and perform dimension-by-dimensional normalization and fixed-order serial tiling on the 8-dimensional state vector to generate a one-dimensional discrete target sequence. Step S2: Time-multiplexed modulation of the one-dimensional discrete target sequence and the chaotic mask signal is performed to obtain an optical injection signal carrying dynamic information; Step S3: The optical injection signal is simultaneously injected into two distributed feedback semiconductor lasers with independent self-feedback loops. The two distributed feedback semiconductor lasers are connected by a bidirectional mutual coupling link. The high-dimensional nonlinear mapping of the input features is completed by utilizing the nonlinear transient response of the distributed feedback semiconductor lasers. Step S4: Collect the output optical signals of the two distributed feedback semiconductor lasers, construct the reservoir state matrix, and solve the output weight matrix based on the training set data using the ridge regression algorithm; Step S5: Substitute the state of the reservoir corresponding to the test set into the output weight matrix obtained by solving to obtain the predicted 8-dimensional dynamic state and end-effector trajectory of the robotic arm.

8. The method for predicting the dynamic trajectory of a robotic arm according to claim 7, characterized in that, Step S1 includes: A reference trajectory is generated based on the rigid body dynamics model of a dual-joint robotic arm, and the reference trajectory includes a chaotic attractor trajectory and a periodic geometric trajectory. The joint angle sequence is obtained by solving the inverse kinematics, and the joint angular velocity and angular acceleration are calculated by the five-point center difference method. The joint torque is obtained by substituting them into the dynamic equation. An 8-dimensional state vector containing 2D joint angle, 2D joint angular velocity, 2D joint angular acceleration and 2D joint torque is constructed and divided into training set and test set in 8:2 ratio according to time sequence. Min-max normalization is performed on each component of the 8-dimensional state vector to map each component to the [0,1] interval. Then, the 8-dimensional state components are serially laid out in a fixed order of joint angle, joint angular velocity, joint angular acceleration, and joint torque to form the one-dimensional discrete target sequence.

9. The method for predicting the dynamic trajectory of a robotic arm according to claim 7, characterized in that, Step S4 includes: The output optical signals of the two distributed feedback semiconductor lasers are converted into current signals by a photodetector, amplified by an electric amplifier, and then sampled by a high-speed oscilloscope at virtual node intervals to obtain the virtual node state sequence. Construct a state matrix for the reservoir. Each column of the state matrix is ​​composed of a bias term, an input target value, and the virtual node states of the two lasers. The corresponding target matrix is ​​the target state sequence for the next time step. The output weight matrix is ​​solved using a ridge regression algorithm with L2 regularization. The regularization coefficient is set to 10 to the power of -6. During the training process, the optical parameters and dynamic characteristics inside the reservoir are kept fixed throughout, and only the output layer weights are updated.

10. A computer-readable storage medium storing computer-executable instructions for performing the robotic arm dynamic trajectory prediction method as described in any one of claims 7 to 9.