Controllers and control systems for flexible actuators, bionic robots

Abstract

This application discloses a controller and control system for a flexible actuator, and a biomimetic robot, belonging to the field of flexible actuator control technology. The controller includes: a pulse encoding module for converting target motion commands into pulse sequences; a spiking neural network processing unit for receiving the pulse sequences and collaboratively generating multi-channel pulse control signals based on its internal state evolution; and a multi-channel drive circuit connected to the output of the spiking neural network processing unit, where each channel receives one pulse control signal and converts it into a drive signal suitable for driving the corresponding flexible drive unit. By integrating the spiking neural network processing unit and the multi-channel drive circuit, the controller achieves end-to-end pulse domain communication from command input to drive execution, and can adaptively adjust the multi-channel control signals based on real-time feedback, thereby providing precise and compliant collaborative control for the flexible actuator.

Description

Technical Field

[0001] This application relates to the field of flexible actuator control technology, specifically to a controller and control system for a flexible actuator, and a bionic robot. Background Technology

[0002] Flexible actuators are novel devices made of flexible materials that can undergo continuous deformation under external stimuli. They can be applied in fields such as soft robots, bionic muscles, medical assistive devices, and flexible grasping devices. Unlike traditional rigid joints, flexible actuators typically contain multiple independently actuated flexible drive units. These units achieve various complex movements such as bending, twisting, and extension through coordinated deformation. To precisely control the motion of the flexible actuator, multiple drive signals need to be generated based on the target motion command, and the deformation state needs to be sensed in real time for closed-loop adjustment.

[0003] However, existing control methods for flexible actuators often employ general-purpose microcontrollers (MCUs) or digital signal processors (DSPs) as the control core. Their architecture, based on continuous numerical processing, struggles to accurately handle the timing relationships between multi-channel signals. This results in a lack of unified coordination mechanisms among the deformation units during motion, making it difficult to precisely match the timing and amplitude of actions, thus affecting the overall motion accuracy and smoothness. To generate the required drive signals, existing solutions often use discrete components to build multi-stage amplifier circuits. As the number of drive units increases, the circuit size significantly increases, and consistency between channels becomes difficult to guarantee, limiting the system's integration and reliability. Existing feedback mechanisms typically only collect a single physical quantity (such as displacement at a point), while the deformation state of a flexible actuator manifests as distributed spatial morphological changes. A single physical quantity cannot fully reflect the actual deformation, preventing the control system from effectively correcting the drive signal based on the accurate deformation state. Existing controllers are usually separated from the flexible actuators, resulting in long signal transmission paths that are susceptible to interference. Furthermore, controllers are often rigid structures, making it difficult to maintain good mechanical compatibility during deformation. Summary of the Invention

[0004] This application provides a controller and control system for a flexible actuator, and a bionic robot, aiming to at least solve the prior art problems described in the background art.

[0005] In a first aspect, this application provides a controller for a flexible actuator, the flexible actuator including multiple independently drivable flexible drive units, the controller including a pulse coding module for converting target motion commands into pulse sequences; a spiking neural network processing unit for receiving the pulse sequences and collaboratively generating multi-channel pulse control signals based on its internal state evolution; and a multi-channel drive circuit connected to the output terminal of the spiking neural network processing unit, each channel being used to receive one pulse control signal and convert it into a drive signal suitable for driving the corresponding flexible drive unit.

[0006] In one possible implementation, the controller is configured to be at least partially located inside the flexible actuator.

[0007] In one possible implementation, the controller is integrated inside the flexible actuator, forming an integral structure with the flexible actuator.

[0008] In one possible implementation, the controller is a flexible controller, with at least a portion of its circuitry or substrate made of a flexible material to accommodate the deformation of the flexible actuator.

[0009] In one possible implementation, the controller is integrated on a flexible substrate as a system-on-a-chip and configured to share a substrate or co-package with the body of the flexible actuator.

[0010] In one possible implementation, the spiking neural network processing unit includes a neural network topology that simulates a biological spinal cord central pattern generator, and is configured to autonomously generate a multi-channel pulse control signal with a preset timing relationship based on the input pulse sequence.

[0011] In one possible implementation, the spiking neural network processing unit includes a collaborative control layer comprising: multiple input neurons for receiving the pulse sequence; multiple interneurons forming excitatory and / or inhibitory connections with the input neurons; and multiple output neurons connected to the interneurons, wherein the pulse activity of the output neurons corresponds to the multi-channel pulse control signal; wherein the interneurons are configured to collaboratively drive the output neurons to generate pulses based on the pulse activity of the input neurons, thereby establishing a preset collaborative relationship between the multi-channel pulse control signals.

[0012] In one possible implementation, the spiking neural network processing unit is configured to: uniformly determine the generation strategy of pulse control signals for each channel according to the target motion command, so that the pulse control signals of each channel are correlated in terms of timing, frequency or pulse density, so as to jointly realize the motion output indicated by the target motion command.

[0013] In one possible implementation, the multi-channel drive circuit includes multiple boost circuits, each boost circuit being used to amplify the pulse control signal output by the spiking neural network processing unit to a voltage or field strength sufficient to drive the corresponding flexible drive unit to produce deformation, and the multiple boost circuits are physically configured to be distributed inside or on the surface of the flexible actuator body, with each boost circuit correspondingly connected to a flexible drive unit.

[0014] In one possible implementation, the multi-channel driving circuit includes at least one of the following energy form conversion circuits: an electro-optic conversion circuit for converting a pulse control signal into an optical driving signal; an electro-thermal conversion circuit for converting a pulse control signal into a thermal driving signal; and an electromagnetic conversion circuit for converting a pulse control signal into a magnetic driving signal.

[0015] In one possible implementation, the multi-channel driving circuit includes: a programmable gain amplifier for adjusting the amplitude of the driving signal of each output channel; a pulse width modulator for adjusting the pulse width of the driving signal of each output channel; and a timer for adjusting the duration of the driving signal of each output channel; wherein the programmable gain amplifier, the pulse width modulator, and the timer are configured to dynamically adjust the output parameters of the corresponding channels according to the target motion command or a feedback pulse sequence from the feedback interface.

[0016] In one possible implementation, the controller further includes a feedback interface for receiving sensing signals from the flexible actuator and converting them into a feedback pulse sequence input to the spiking neural network processing unit.

[0017] In one possible implementation, the spiking neural network processing unit further includes an error detection circuit, which is configured to: compare the feedback pulse sequence with the target pulse sequence corresponding to the target motion command, and adjust the synaptic weights inside the spiking neural network processing unit or send an adjustment signal to the multi-channel driving circuit according to the comparison result, so as to change the gain, bias or timing relationship of each channel driving signal.

[0018] In one possible implementation, the spiking neural network processing unit is configured to dynamically adjust the coordination relationship between the multi-channel pulse control signals according to the feedback pulse sequence to adapt to the deformation state changes of the flexible actuator.

[0019] In one possible implementation, the controller is configured to: cause the first drive unit to undergo axial extension and contraction when the drive signal is applied to the first drive unit of the flexible actuator; and cause the second drive unit to undergo bending deformation when the drive signal is applied to the second drive unit of the flexible actuator.

[0020] In one possible implementation, the spiking neural network processing unit is configured to dynamically adjust the relative timing relationship between the pulse control signals of each channel according to the target motion command, so as to form dynamic antagonism among multiple flexible drive units.

[0021] Secondly, embodiments of this application also provide a flexible actuator control system, including: multiple subsystems, each subsystem corresponding to a bionic part or a motion area; each subsystem including: multiple controllers provided in the first aspect; multiple flexible actuators, each flexible actuator being connected to a corresponding controller and independently controlled by the corresponding controller; wherein, the multiple controllers are distributed within the subsystem and are respectively located inside or near the flexible actuators they control.

[0022] In one possible implementation, the multiple subsystems are connected via a synchronization bus or wireless communication interface to achieve motion coordination between the subsystems.

[0023] In one possible implementation, at least a portion of the plurality of controllers is configured as a master controller, and the remainder as slave controllers; the master controller is used to receive the total target motion command, decompose it into sub-target motion commands for each slave controller, and distribute them to the corresponding slave controller.

[0024] In one possible implementation, the plurality of flexible actuators include different types of drive units, and the controller outputs an adapted drive signal according to the type of each flexible actuator.

[0025] In one possible implementation, an upper-level coordination controller is also included, which is connected to all subsystems and is used to send motion commands to each subsystem according to the global task plan and collect feedback status information from each subsystem.

[0026] Thirdly, embodiments of this application also provide a biomimetic robot, including at least one flexible actuator control system provided in the second aspect.

[0027] The above technical solution first converts target motion commands (such as bending angles and speeds) into pulse sequences using a pulse coding module. Then, a spiking neural network processing unit receives these pulse sequences and, based on the dynamic evolution of its internal neuronal membrane potentials and synaptic weights, collaboratively generates multi-channel pulse control signals. This means that the pulses in each channel are correlated in terms of timing, frequency, or density. Finally, a multi-channel drive circuit converts each pulse control signal into a drive signal (such as a high-voltage electric field, light pulse, or thermocurrent) suitable for driving the corresponding flexible drive unit. This controller addresses the problem of existing control systems being unable to coordinate the coordinated actions of multiple drive units and having difficulty in signal timing coordination. It utilizes the state evolution mechanism of a spiking neural network to achieve unified planning and coordinated output of multi-channel signals, avoiding timing conflicts and action misalignment caused by independent control of each unit. Simultaneously, the direct pulse domain connection between the pulse coder and the drive circuit reduces the number of digital-to-analog conversion steps, lowering latency and power consumption. The multi-channel parallel drive architecture enables the controller to simultaneously and accurately control multiple flexible drive units, significantly improving the synchronization, compliance, and response speed of the flexible actuator in multi-unit coordinated motion. Attached Figure Description

[0028] Figure 1 A schematic diagram of the structure of a controller for a flexible actuator provided in one embodiment of this application;

[0029] Figure 2 This is a schematic diagram of the structure of a flexible actuator control system provided in one embodiment of this application. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the following description is intended to demonstrate how to implement multiple specific solutions of this application to fully support the scope of protection claimed in the claims, and does not constitute a limitation on the scope of protection.

[0031] This application provides a controller for a flexible actuator. The controller integrates a pulse coding module, a spiking neural network processing unit, and a multi-channel drive circuit. It can directly convert target motion commands into pulse sequences. The spiking neural network processing unit collaboratively generates multi-channel pulse control signals, which are then converted into drive signals suitable for driving the flexible actuator by the multi-channel drive circuit. The controller also includes a feedback interface to receive deformation state information from the flexible actuator and convert it into a feedback pulse sequence, which is then input to the spiking neural network processing unit to form a closed-loop control. By integrating the spiking neural network processing unit with the multi-channel drive circuit, the controller achieves end-to-end pulse domain communication from command input to drive execution and can adaptively adjust the multi-channel control signals based on real-time feedback, thereby providing precise and compliant coordinated control for the flexible actuator.

[0032] The controller of the flexible actuator provided in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0033] Figure 1 This is a schematic diagram of the structure of a controller for a flexible actuator provided in one embodiment of this application.

[0034] Reference Figure 1 As shown, the controller 100 may include: a pulse code module 101, a pulse neural network processing unit 102, and a multi-channel drive circuit 103.

[0035] The pulse coding module 101 is used to convert target motion commands into pulse sequences.

[0036] In some embodiments, when a target motion command (such as "bend 30 degrees, medium speed") is input, the command is decomposed into one or more motion parameters, which may include motion direction, motion amplitude, motion speed, and motion trajectory. These motion parameters are then converted into a pulse sequence with specific pulse quantity, frequency, and timing characteristics according to a preset encoding mapping relationship. For example, for the command "bend 30 degrees, medium speed," the pulse encoding module 101 can output 300 pulses within 100 milliseconds at a pulse frequency of 3 kHz, where the number of pulses is proportional to the target amplitude (30 degrees), and the pulse frequency is positively correlated with the motion speed. The pulse encoding module 101 outputs one pulse sequence, which serves as the input to the spiking neural network processing unit 102, which collaboratively generates a multi-channel pulse control signal based on this sequence.

[0037] The spiking neural network processing unit 102 is used to receive pulse sequences and generate multi-channel pulse control signals based on the evolution of its internal state.

[0038] In some embodiments, the spiking neural network processing unit 102 is configured to run a spiking neural network model, which consists of multiple neuron nodes and synaptic connections arranged in a preset topological relationship. When the pulse sequence output by the pulse encoding module 101 is input into the spiking neural network processing unit 102, each pulse signal in the pulse sequence propagates along the synaptic connections between neuron nodes. Each neuron node integrates its membrane potential based on the received pulse, and outputs a pulse when the membrane potential exceeds the firing threshold, which is then transmitted to the downstream neuron node through the synaptic connection.

[0039] The "internal state" of the spiking neural network processing unit 102 includes at least one of the following: the membrane potential value of each neuron node, the synaptic connection weight, and the pulse firing history. These internal states evolve dynamically over time, and ultimately, a pre-defined set of output neuron nodes (e.g., extensor motor neurons and flexor motor neurons) collaboratively output pulse signals to form a multi-channel pulse control signal. That is, the pulse signals of each output channel are correlated in terms of timing, frequency, or pulse density, and together achieve the motion output indicated by the target motion command. The collaborative relationship between the pulse control signals of each channel (such as relative delay, pulse frequency ratio, etc.) is determined collaboratively by the excitatory and inhibitory connections within the neural network topology, rather than being directly specified by external commands.

[0040] For example, when the input pulse sequence represents the instruction to "bend 30 degrees", the position-sensing neurons inside the spiking neural network processing unit 102 determine the difference between the target value and the current value based on the input pulse frequency, the velocity-sensing neurons determine the direction of motion velocity change based on the membrane potential change rate, and the extensor and flexor motor neurons collaboratively generate phase-shifted pulse outputs through antagonistic connections. The extensor channel pulses drive the corresponding flexible actuator to produce elongation deformation, while the flexor channel pulses drive another set of actuators to produce contraction deformation. The timing coordination between the two (e.g., the extensor pulse precedes the flexor pulse by a few milliseconds) enables the flexible actuator to produce a smooth bending motion. In this way, the spiking neural network processing unit 102 collaboratively "evolves" the input single-channel pulse sequence into a multi-channel pulse control signal with a specific collaborative relationship. The multi-channel pulse control signal output by the spiking neural network processing unit 102 is then transmitted to the multi-channel drive circuit 103, which converts it into a drive signal suitable for driving the flexible actuator.

[0041] The multi-channel driving circuit 103 is connected to the output of the pulse neural network processing unit 102. Each channel is used to receive a pulse control signal and convert it into a driving signal suitable for driving the corresponding flexible driving unit.

[0042] In some embodiments, the multi-channel drive circuit 103 includes drive channels corresponding one-to-one with the pulse control signal channels. Each drive channel independently receives a low-voltage pulse control signal and amplifies and conditions it to generate a drive signal with sufficient voltage, current, or field strength. During the conversion process, the drive circuit transmits the timing, pulse width, frequency, and other characteristics of the input pulse to the output drive signal through charge pumps, switched capacitor networks, or transformer coupling. Simultaneously, based on the material properties of the flexible drive unit (such as dielectric elastomers, piezoelectric materials, shape memory alloys, etc.), the low-voltage pulse is converted into corresponding forms of drive energy. For example, for an electro-actuated unit, the drive circuit amplifies the low-voltage pulse into a high-voltage pulse of hundreds to thousands of volts, generating a strong electric field on the flexible drive unit, causing electro-induced deformation. For a thermally actuated unit, the drive circuit converts the pulse signal into a controllable heating current, causing thermal deformation due to temperature changes in the unit. The relative timing relationship of the input pulse control signals is maintained between the drive channels to ensure that multiple flexible drive units deform collaboratively according to a predetermined time coordination relationship. The converted drive signal is directly applied to the corresponding flexible drive unit through the output terminal, driving it to produce the expected deformation action.

[0043] based on Figure 1 The controller for the flexible actuator provided in the illustrated embodiment, for a flexible actuator comprising multiple independently driveable flexible drive units, converts the target motion command into a pulse sequence through a pulse coding module 101. A spiking neural network processing unit 102 receives this pulse sequence and, based on the evolution of its internal state, collaboratively generates multi-channel pulse control signals. Then, a multi-channel drive circuit 103 converts the pulse control signals of each channel into drive signals suitable for driving the corresponding flexible drive unit. By utilizing the internal state evolution mechanism and collaborative generation capability of the spiking neural network, parallel, independent, and collaborative control of multiple flexible drive units is achieved. This enables each drive unit to produce mutually coordinated deformations according to a unified motion intention, thereby solving the technical problem that traditional single-channel controllers cannot simultaneously drive multiple units and ensure timing coordination. This significantly improves the control accuracy and compliance of the flexible actuator in multi-unit collaborative motion.

[0044] based on Figure 1 The controller for the flexible actuator provided in the illustrated embodiment will be described in detail below with reference to the accompanying drawings, including its specific structure and the functions of each component module. First, the internal implementation of the pulse coding module 101, the pulse neural network processing unit 102, and the multi-channel drive circuit 103, and their roles in coordinated control, will be explained sequentially. Finally, the signal transmission relationships and coordinated working mechanisms between the modules will be described.

[0045] The internal implementation of the pulse code module 101 and its role in coordinated control are as follows:

[0046] In some embodiments, the pulse coding module 101 converts the target motion command into a pulse sequence, including: generating a corresponding pulse sequence by using at least one of rate coding, time coding, or group coding according to at least one of the motion direction, motion amplitude, motion speed, or motion trajectory of the target motion command.

[0047] In a specific implementation, when the pulse coding module 101 converts the target motion command into a pulse sequence, it uses different coding strategies to generate the corresponding pulse sequence according to the motion parameters such as motion direction, motion amplitude, motion speed or motion trajectory contained in the command. Specifically: For motion direction, a population coding method is used to map the direction angle to the firing ratio of a group of direction-selecting neurons. For example, 0° (directly forward) corresponds to 100% firing rate of the first group of neurons and 0% firing rate of the second group, and 90° (directly to the right) corresponds to 50% firing rate of the first group and 50% firing rate of the second group, so that the direction information is represented by the parallel pulse activity of multiple neurons. For motion amplitude (such as bending angle, extension length), a rate coding method is used to map the amplitude value to the pulse frequency linearly or nonlinearly. For example, 0° corresponds to 0Hz, 30° corresponds to 3kHz, and 60° corresponds to 6kHz. The larger the amplitude, the higher the pulse frequency. For motion speed, a time coding method is used to map the speed value to the total duration of the pulse train or the pulse interval. For example, at low speed, the pulse train lasts for 200ms and the pulse interval is sparse, while at high speed, the pulse train lasts for 50ms and the pulse interval is close. For motion trajectory (such as S-curve, uniform straight line), time coding and population coding are used in combination to discretize the trajectory into a series of instantaneous directions and amplitudes at a series of time points, which are expressed by the time pattern of the pulse sequence. This method of encoding multidimensional parameters separately enables the pulse sequence to carry complete motion commands with high information density and low redundancy, while maintaining natural compatibility with the pulse domain processing logic inside the spiking neural network processing unit, avoiding the delay and accuracy loss caused by digital-to-analog conversion.

[0048] For example, when the target motion instruction is "bend 30 degrees to the right, at a moderate speed, and move along an S-shaped trajectory", the pulse coding module first outputs the direction "right" through population coding, with the first group of neurons firing at 100% and the second group firing at 0%; it sets the pulse frequency to 3kHz for the amplitude "30 degrees" through rate coding; it sets the pulse duration to 100ms for the speed "moderate" through time coding; and it discretizes the "S-shaped trajectory" into 50 time segments through time coding, with each segment using population coding to represent the instantaneous direction (gradually changing from right to left and then back to right), ultimately generating a time-varying pulse sequence that is input into the spiking neural network processing unit, enabling it to accurately understand and execute complex continuous trajectory motion.

[0049] The internal implementation of the spiking neural network processing unit 102 and its role in cooperative control are as follows:

[0050] In some embodiments, the spiking neural network processing unit includes a neural network topology that simulates a biological spinal cord central pattern generator, and is configured to autonomously generate a multi-channel pulse control signal with a preset timing relationship based on the input pulse sequence.

[0051] In a specific implementation, the spiking neural network processing unit adopts a neural network topology that simulates a central pattern generator (CPG) in the spinal cord. This topology consists of multiple neuronal nodes connected according to specific excitatory and inhibitory relationships, forming a neural circuit capable of autonomously generating rhythmic oscillations. Specifically, this neural network topology includes bilaterally symmetrical neuronal groups or vertically coupled oscillating units. Each neuronal node is connected to a network structure with a stable phase relationship through preset synaptic connection weights and connection polarities (such as cross-inhibition and mutual excitation). When a pulse sequence input from the pulse coding module is injected into the network as a start signal, the network does not require external clock or instruction intervention. Instead, it autonomously generates multiple pulse outputs with fixed phase differences, frequency ratios, or timing misalignments based on its internal membrane potential integration and firing mechanism, forming a multi-channel pulse control signal. This CPG topology significantly reduces the computational burden on the control system. It eliminates the need for real-time calculation of the drive timing for each channel; simple start or speed control commands are all that's required, and the network can autonomously generate complex cooperative timing sequences. The hardwired structure of the network topology ensures the stability and anti-interference capabilities of the phase relationship between the pulse control signals of each channel, enabling the flexible actuator to maintain smooth and compliant dynamic characteristics during multi-unit cooperative motion. Simultaneously, the biomimicry characteristics of CPG make the controller's output closer to the antagonistic cooperative mechanism of a biological muscle system, making it particularly suitable for applications requiring high naturalness of movement, such as bionic robots and prosthetics.

[0052] For example, the spiking neural network processing unit internally constructs a CPG topology that simulates the control of the fish tail's swing in the spinal cord. This topology contains two groups of neurons, left and right. Each group maintains synchronous oscillation through excitatory connections, and the two groups form a 180-degree phase difference through cross-inhibition connections. When a pulse sequence representing "start" is input, the CPG topology autonomously generates two pulse control signals with opposite phases and the same frequency, which drive the drive units on both sides of the flexible actuator to alternately contract and extend, so that the bionic fish tail produces continuous and stable swinging motion without the need for an external controller to calculate the switching timing of each drive unit in real time.

[0053] In some embodiments, the spiking neural network processing unit includes a collaborative control layer, which comprises: multiple input neurons for receiving pulse sequences; multiple interneurons forming excitatory and / or inhibitory connections with the multiple input neurons; and multiple output neurons forming connections with the multiple interneurons. The pulse activities of the multiple output neurons correspond to multi-channel pulse control signals. The interneurons are configured to collaboratively drive the multiple output neurons to generate pulses based on the pulse activities of the multiple input neurons, thereby forming a preset collaborative relationship between the multi-channel pulse control signals.

[0054] In a specific implementation, to achieve the aforementioned collaborative generation function, the spiking neural network is internally configured with a collaborative control layer. This layer consists of multiple input neurons, intermediate neurons, and output neurons connected according to a specific topological structure. Specifically, the input neurons employ an integrator-fire circuit. Each input neuron receives a pulse sequence output by the pulse coding module and transmits it in parallel to multiple intermediate neurons. The intermediate neurons are divided into several functional groups. Within each group, excitatory synaptic connections form positive feedback loops to maintain oscillations (e.g., enhancing input through output feedback to ensure continuous firing). Intergroups are mutually constrained through inhibitory synaptic connections to generate phase differences (e.g., when neurons in group A fire, the membrane potential of neurons in group B is reduced through inhibitory synapses). Intermediate neurons and output neurons are connected through configurable synaptic weights. Each output neuron can receive input from multiple intermediate neurons and determines its firing mode based on a weighted sum. The intermediate neurons are configured to collaboratively drive multiple output neurons to generate pulses based on the pulse activity of the input neurons, thus establishing a preset collaborative relationship between the multi-channel pulse control signals. This structural design embeds the complexity of collaborative control into the hardware connection, enabling parallelization of control logic, low-latency processing, and flexible expansion of multi-channel output.

[0055] For example, in a bionic muscle application controlling an eight-channel drive, the collaborative control layer is configured as follows: two input neurons receive a "start" pulse and a "velocity regulation" pulse, respectively; four interneurons form a first group of oscillators, which generate synchronous oscillations of 50Hz through mutual excitation; another four interneurons form a second group of oscillators, which form a 90° phase difference with the first group through cross-inhibition; each interneuron then drives an output neuron through a one-to-one excitatory connection, while each output neuron also receives weak inhibitory connections from two adjacent interneurons. Specifically, the interneurons A1, A2, A3, and A4 in the first group maintain synchronous oscillations through excitatory connections (weight 0.8); the interneurons B1, B2, B3, and B4 in the second group form a phase lag with the first group through cross-inhibition connections (weight -0.5); output neuron O1 receives excitation from A1 (weight 1.0) and inhibition from B2 (weight -0.2), O2 receives excitation from A2 and inhibition from B3, and so on. When the input neurons receive the "start" pulse, the first group of interneurons fire simultaneously at 50Hz oscillations, driving the output neurons O1-O4 to fire synchronously through excitatory connections. After a phase lag, the second group of interneurons fires 10ms later, driving the output neurons O5-O8 to fire synchronously. Simultaneously, weak inhibition of adjacent channels causes the pulses of each output neuron to attenuate slightly at the edges. Ultimately, the eight output pulses form a complex coordinated pattern: the first group of four channels is synchronous, the second group of four channels is synchronous, the two groups are 10ms apart, and the pulse intensity of the edge channels within each group is slightly weaker. This drives the bionic muscle to produce a peristaltic wave-like movement that propagates from one end to the other. All coordinated relationships are automatically achieved by hardware connections, without the need for software intervention or real-time calculations.

[0056] In some embodiments, the collaborative control layer consists of a three-level circuitry: an input neuron array, an intermediate neuron grid, and output neurons. All neurons are implemented using integrator-discharge hardware circuitry. Each neuron in the input neuron array receives an external pulse sequence, converts it into an integral membrane potential signal, and broadcasts it to the intermediate neuron grid via a parallel bus. The intermediate neuron grid configures synaptic weights according to a preset excitation / inhibition connection routing table: excitatory connections transmit membrane potential increments through positive weights, while inhibitory connections reduce the membrane potential of the target neuron through negative weights. Neurons within the grid form mutually stimulating or mutually inhibiting topologies, such as symmetrical cross-inhibition loops or cascaded delay chains. The output neurons receive weighted pulse inputs from specific nodes in the intermediate neuron grid and generate the final multi-channel pulse control signal. The entire routing configuration can be fixed in hardware (e.g., read-only memory) or programmed online via registers, thereby enabling flexible switching between different collaborative modes (e.g., fixed phase difference, frequency ratio, or timing misalignment).

[0057] In some embodiments, the spiking neural network processing unit is configured to: uniformly determine the generation strategy of the pulse control signals of each channel according to the target motion command, so that the pulse control signals of each channel are correlated in terms of timing, frequency or pulse density, so as to jointly realize the motion output indicated by the target motion command.

[0058] In a specific implementation, the spiking neural network processing unit includes an instruction parsing and strategy mapping module. This module first parses parameters such as motion type (e.g., bending, twisting), amplitude, and velocity from the target motion instruction. Then, based on a preset cooperative rule base, it independently calculates the start time, frequency, pulse width, and phase offset from other channels for each output channel's pulse signal, and writes these parameters into the pulse generation register of each channel, thereby achieving unified planning and correlated output of pulse signals from each channel. This implementation avoids timing conflicts or incoordination that may result from independent decision-making by each channel, ensuring that multiple flexible drive units coordinate their actions according to a preset cooperation relationship from the initial motion stage. This significantly improves the synchronization and compliance of the overall motion of the flexible actuator while reducing the real-time computational burden.

[0059] For example, when the target motion command is "bend 30 degrees to the right", the spiking neural network processing unit uniformly determines that: the extensor channel outputs pulses at a frequency of 5 kHz starting at 0 ms, and the flexor channel outputs pulses at a frequency of 1 kHz starting at a delay of 5 ms. The pulse density ratio of the two channels is 5:1, which makes the extensor contract quickly and the flexor then provide damping, together achieving a smooth 30-degree bending motion.

[0060] In some embodiments, the spiking neural network processing unit further includes an error detection circuit, which is configured to: compare the feedback pulse sequence with the target pulse sequence corresponding to the target motion command, and adjust the synaptic weights inside the spiking neural network processing unit or send an adjustment signal to the multi-channel driving circuit according to the comparison result, so as to change the gain, bias or timing relationship of the driving signals of each channel.

[0061] In a specific implementation, the error detection circuit integrated within the spiking neural network processing unit adopts a dual-path comparison architecture. One path inputs the feedback pulse sequence to a frequency counter, comparing it in real time with the expected frequency of the target pulse sequence to calculate the frequency deviation Δf. The other path measures the time offset Δt between the feedback pulse and the target pulse (e.g., the absolute time difference between the feedback pulse leading or lagging the target pulse) through a time-to-digital converter. Based on the magnitude and direction of Δf and Δt, the error detection circuit determines the adjustment path according to preset control logic: for the frequency deviation Δf, when the deviation exceeds a threshold (e.g., Δf > 10%), the weights of the corresponding synapses within the spiking neural network processing unit are adjusted through the pulse timing-dependent plasticity (STDP) mechanism to optimize the network output over the long term; when the deviation is small, adjustment commands are directly sent to the multi-channel drive circuit to change the gain of the corresponding channel in real time (e.g., adjusting the gain of the programmable gain amplifier via the I2C bus). For a time offset Δt, if the feedback pulse leads the target pulse (Δt is positive), the output delay of the drive signal for that channel is increased via a programmable delay line, causing the actual action to lag. If the feedback pulse lags the target pulse (Δt is negative), the delay is reduced or the trigger is advanced (by adjusting the initial count value of the pulse generation register), thereby correcting the timing errors between the drive units. This hierarchical control logic separates long-term learning from short-term rapid correction, maintaining the system's adaptability to material aging and environmental changes while ensuring millisecond-level real-time correction response, thus significantly improving the motion accuracy and disturbance rejection capability of the flexible actuator.

[0062] For example, during the repeated grasping process of the bionic finger, if the bending angle is insufficient due to increased load, the error detection circuit detects that the feedback pulse frequency is 5% lower than the target frequency and immediately sends a command to the drive circuit to increase the extensor channel gain from 600V to 720V, thus restoring the angle. If the bending is caused by the difference in response speed between the two drive units, and the feedback pulse on one side is detected to be 2ms ahead of the other side, the output of the leading channel is delayed by 2ms through the delay line to synchronize the two sides and restore pure bending motion.

[0063] In some embodiments, the spiking neural network processing unit is configured to dynamically adjust the coordination relationship between multi-channel pulse control signals according to the feedback pulse sequence to adapt to changes in the deformation state of the flexible actuator.

[0064] In a specific implementation, the spiking neural network processing unit is configured to receive the feedback pulse sequence generated by the feedback interface of the controller in real time. Through an internal coordination relationship adjustment circuit, it analyzes the relative timing, frequency ratio, and pulse density distribution between the feedback pulses of each channel, and compares this with the currently output multi-channel pulse control signal. It dynamically adjusts the synaptic connection weights between neurons or the phase coupling strength between output channels, thereby changing the coordination relationship (such as phase difference, frequency ratio, or pulse density distribution pattern) between the pulse control signals of each channel to adapt to changes in the deformation state of the flexible actuator caused by load variations, material fatigue, or external disturbances. This implementation enables the spiking neural network to automatically optimize the coordination between multiple channels based on the real-time deformation feedback of the flexible actuator, avoiding motion misalignment or energy waste caused by fixed coordination relationships, and significantly improving the system's adaptability in dynamic environments.

[0065] For example, in the continuous rotational movement of the bionic wrist, when the load suddenly increases and causes the torsional speed to decrease, the pulse frequency of the corresponding torsional channel in the feedback pulse sequence decreases. The spiking neural network immediately increases the frequency ratio between the torsional channel and the bending channel (from 2:1 to 3:1) and fine-tunes the phase difference between the two (from 90° to 80°), so that the wrist can automatically compensate for the load while maintaining rotation and keep the movement stable.

[0066] In some embodiments, the spiking neural network processing unit is configured to dynamically adjust the relative timing relationship between the pulse control signals of each channel according to the target motion command, so as to form dynamic antagonism among multiple flexible drive units. The target motion command may include at least one of motion direction, velocity, and acceleration.

[0067] In a specific implementation, the spiking neural network processing unit is configured to dynamically calculate the relative timing relationship between each output channel based on the target motion command (such as parameters like motion direction, speed, and acceleration). Through a built-in programmable delay chain and phase coupling logic, it adjusts the start time difference, pulse interval, or phase offset of the pulse control signals of different channels in real time, thereby forming dynamic antagonism among multiple flexible actuators. That is, the agonist muscle unit and the antagonist muscle unit alternately contract or cooperate in braking according to motion requirements, so as to achieve precise control of motion speed, acceleration, and end-effector compliance. This implementation method differs from static fixed-timing antagonism. Dynamic antagonism can adjust the force sequence and intensity of the agonist and antagonist muscles in real time according to the real-time changes of the command (such as sudden stop, speed change, and direction change), so that the flexible actuator remains stable during high-speed motion, avoids overshoot during sudden stop, and achieves smooth transition during direction change, significantly improving the dynamic response performance and biomimetic anthropomorphism of the system.

[0068] For example, in a scenario where the bionic arm swings rapidly and then stops abruptly, the spiking neural network dynamically adjusts according to the "stop" command: it immediately cuts off the extensor channel pulses, while triggering the flexor channel pulses 10ms in advance and increasing the pulse width, so that the antagonist muscles exert force in advance to generate braking torque, thereby stopping the arm smoothly within 50ms and avoiding end-effector tremors.

[0069] The internal implementation of the multi-channel drive circuit 103 and its role in cooperative control are as follows:

[0070] In some embodiments, the multi-channel driving circuit includes multiple boost circuits, each boost circuit being used to amplify a first pulse control signal output by the spiking neural network processing unit to a second pulse control signal, wherein the voltage or field strength of the second pulse control signal is sufficient to drive the corresponding flexible driving unit to produce deformation.

[0071] In specific implementations, each boost circuit integrates a charge pump circuit, a switched capacitor network, or a transformer-coupled boost circuit. It progressively boosts and amplifies the first pulse control signal (typically a low-voltage, low-current logic-level pulse) output from the pulse neural network processing unit, converting it into a second pulse control signal. This ensures that the voltage amplitude or electric field strength of the second pulse control signal is sufficient to drive the corresponding flexible actuator to produce the desired deformation. Localized layout significantly shortens the transmission path of the high-voltage drive signal, reducing electromagnetic interference and signal attenuation introduced by long-distance transmission, and ensuring the timing accuracy and waveform fidelity of the drive signals in each channel. Each drive unit is independently packaged or integrated, providing good electrical isolation between channels and avoiding crosstalk between multiple channels. Simultaneously, the boost circuit can deform along with the flexible actuator, avoiding the mechanical constraints imposed on the flexible body by the centralized drive circuit during deformation, thus improving the overall flexibility and integration of the system.

[0072] For example, eight boost circuits are directly mounted on a flexible circuit board in bare die form. Each boost circuit is adjacent to the dielectric elastomer driving unit it drives. The 0~3.3V pulse signal output by the pulse neural network processing chip is boosted to 800V by the cascaded charge pump circuit in the boost circuit. It is then applied directly to the electrodes on both sides of the dielectric elastomer driving unit through short-pitch silver paste traces on the flexible circuit board. This enables each driving unit to generate contraction or extension deformation precisely and synchronously according to the timing of the pulse control signal, realizing the smooth bending movement of the biomimetic muscle complex.

[0073] It should be noted that "sufficient to drive the corresponding flexible actuator to produce deformation" means that, based on the material properties and structural design of the flexible actuator, it can generate the voltage amplitude or electric field strength required to produce the expected deformation. For example, for dielectric elastomer actuators, a voltage of several hundred to several thousand volts is typically required to produce significant deformation; for piezoelectric ceramic actuators, a voltage of tens to several hundred volts is typically required; and for shape memory alloy actuators, current heating may be required instead of simple voltage driving. Those skilled in the art can determine the specific driving voltage or field strength range through conventional experiments based on the type of flexible actuator used.

[0074] In some embodiments, each boost circuit is used to amplify the pulse control signal output by the spiking neural network processing unit to a voltage or field strength sufficient to drive the corresponding flexible drive unit to produce deformation, and the multiple boost circuits are physically configured to be distributed inside or on the surface of the flexible actuator body, with each boost circuit correspondingly connected to a flexible drive unit.

[0075] In another implementation, each boost circuit can employ a charge pump, transformer, or switched capacitor topology to progressively boost the low-voltage pulse control signal (e.g., 3.3V) output by the pulse neural network processing unit to hundreds or even thousands of volts, so that its voltage or field strength is sufficient to drive the corresponding flexible actuator to produce the desired deformation. At the same time, multiple boost circuits are physically configured to be distributed and embedded inside the flexible actuator body (e.g., silicone inner wall, muscle bundle gap) or attached to its surface. Each boost circuit forms a one-to-one direct connection with a flexible actuator through flexible wires or conductive ink lines. This implementation ensures that the driving signal has sufficient energy to trigger reliable deformation at the signal amplification level, avoiding motion failure due to insufficient driving capability. At the physical structure configuration level, the distributed layout significantly shortens the transmission path of the high-voltage signal, reduces electromagnetic interference and signal attenuation, and allows the boost circuit to deform along with the flexible actuator, avoiding the mechanical constraints of centralized circuits on the flexible body. At the one-to-one connection level, each boost circuit independently drives a flexible driving unit, eliminating crosstalk between channels and facilitating independent energy regulation and timing control of each unit, thereby significantly improving the accuracy of multi-unit coordinated motion and system integration. For example, in a bionic finger, eight miniature boost circuits (2mm×2mm in size) are embedded in the inner wall of the silicone knuckle in bare form. Each circuit is connected to a dielectric elastomer driving unit through short-pitch silver paste lines, amplifying the 3.3V pulse to 600V, enabling each driving unit to contract independently and precisely according to the timing output of the pulse neural network, achieving smooth bending of the finger.

[0076] Because the bionic muscles of different parts of a bionic robot (such as fingers, wrists, and upper arms) have significantly different requirements for driving voltage, output power, size, and compliance, a single type of boost circuit cannot meet all these needs. Therefore, this application provides three different sizes of boost circuits based on different application scenarios: micro-embedded, small flexible integrated, and compact modular. These are described below.

[0077] A miniature embedded boost circuit is configured to be fully embedded within the flexible inner wall of the distal end (such as fingers or toes) of a bionic robot or within a miniature bionic muscle body. Its overall contour matches the tiny volume of the implanted cavity, and its thickness does not exceed the elastic deformation limit of the flexible epidermis of that area. For example, its dimensions can be specifically configured as <5mm × 5mm × 1mm. In one embodiment, this miniature embedded boost circuit employs a fully integrated charge pump topology, utilizing an on-chip MOSFET switch array and flying capacitors to achieve multi-stage voltage multiplication without the need for external inductors. Its power amplification is achieved as follows: a low-voltage pulse signal (e.g., 3.3V / 10mA) output from a pulse neural network processing unit is shaped and buffered before driving a non-overlapping clock generator inside the charge pump. Through three to six cascaded cross-coupled charge pump units, each stage boosts the voltage by approximately 1.5 to 2 times, ultimately outputting a drive signal in the 200-400V / mA range (power approximately 0.2-2W).

[0078] In some applications, this miniature embedded boost circuit can be directly embedded in the silicone inner wall of a bionic finger or integrated into the surface of a miniature bionic muscle within the finger. Its extreme miniaturization allows it to be implanted in the most distal bionic part. The all-solid-state, inductive-free structure is resistant to bending and impact, and exhibits high reliability during repeated finger flexion and extension. Simultaneously, the adjustable switching frequency allows for precise control of the contraction force of the fingertip muscles. For example, in a bionic dexterous hand, a 3mm × 3mm charge pump chip is embedded in the fingertip of each finger, boosting the 3.7V battery voltage to 300V to drive a 2mm diameter dielectric elastomer muscle unit at the fingertip, enabling the bionic finger to perform delicate movements such as gently pinching objects like strawberries.

[0079] A small, flexible, integrated boost circuit is configured to fit within a flexible cavity or muscle bundle gap in a medium-sized part of a bionic robot (such as the palm or wrist). Its length and width are adapted to the usable area on the dorsal side of the joint, and it possesses overall flexibility, without substantially restricting the normal flexion and extension freedom of the joint. For example, its dimensions are specifically configured as 10mm × 10mm × 2mm to 20mm × 15mm × 3mm. In one embodiment, the small, flexible, integrated boost circuit employs a hybrid integration process, mounting the boost controller, a miniature planar transformer or coupled inductor, capacitor, and rectifier diodes onto a flexible printed circuit board (FPC). Its power amplification is achieved as follows: a low-voltage pulse signal drives the controller to generate a high-frequency PWM wave, which is boosted by a miniature transformer (turns ratio 1:20 to 1:50), and then smoothed into high-voltage DC or a high-voltage pulse sequence by a voltage doubler rectifier network, resulting in an output voltage of 400–800V and an output power of 1–5W.

[0080] In some applications, this small, flexible, integrated boost circuit is bendable and can be encased in a flexible cavity on the back of a bionic hand or wrist, or encapsulated in a silicone jacket over a bionic muscle bundle. Compared to a charge pump, the transformer topology offers a higher boost ratio and power density, with lower output voltage ripple, making it suitable for driving medium loads requiring a stable electric field. The flexible substrate allows the circuit to adapt to dynamic deformation during joint bending, while reducing the number of cables. For example, in the flexion and extension movements of a bionic wrist, a 15mm × 15mm flexible boost circuit is embedded under the skin on the back of the wrist, providing 600V / 2W drive to three sets of parallel bionic muscles, enabling smooth up-and-down wrist movements.

[0081] A compact modular boost circuit is configured to be embedded within the internal cavity of a rigid or semi-rigid shell of a large proximal part of a bionic robot (such as the upper arm or thigh). Its compact shape avoids space constraints related to the arrangement of the main drive muscle bundles. It connects to the distal load via a flexible cable, without affecting the overall mobility and appearance of the robot. For example, its dimensions can be specifically configured as 30mm × 25mm × 5mm to 50mm × 40mm × 10mm. In one embodiment, the compact modular boost circuit employs a multi-phase interleaved Boost or flyback topology, using a small planar transformer, solid-state capacitors, and shielded inductors, all encapsulated in a metal / plastic shielded shell. It has an output power of 10–50W and an output voltage exceeding 1000V. Its power amplification is achieved by a low-voltage pulse signal generating a multi-phase interleaved control sequence via a driver chip, which drives multiple power MOSFETs. Energy is collected and output through a planar transformer and a voltage doubler rectifier network. The interleaving technique reduces input current ripple and output capacitor stress, resulting in a conversion efficiency typically >90%.

[0082] In some applications, this compact modular boost circuit can be embedded in the internal cavity of a bionic upper arm or thigh, connecting to multiple distributed bionic muscle groups via flexible cables. A silicone buffer layer can be added to the bottom of the module to reduce vibration transmission. High power output meets the explosive power requirements of large muscle groups such as the upper arm and thigh, such as rapid arm raising or leg kicking movements. The modular design facilitates maintenance and replacement, and the shielded outer shell effectively suppresses high-voltage switching noise. For example, in a bionic upper limb robot, a 40mm × 30mm × 8mm modular boost circuit is embedded inside the upper arm shell, boosting the 12V battery voltage to 1200V, simultaneously driving six bionic muscle bundles, enabling the robot to lift a 5kg weight.

[0083] In some embodiments, the multi-channel driving circuit includes at least one of the following energy form conversion circuits: an electro-optic conversion circuit for converting a pulse control signal into an optical driving signal; an electro-thermal conversion circuit for converting a pulse control signal into a thermal driving signal; and an electromagnetic conversion circuit for converting a pulse control signal into a magnetic driving signal.

[0084] In a specific implementation, the pulse control signal of each channel is first amplified by a pre-drive stage to enhance its driving capability; then, different processing paths are selected according to the type of flexible drive unit. It should be noted that the types of flexible drive units include electro-actuated, thermal-actuated, optical-actuated, and magnetic-actuated types, and the corresponding drive signals must be adapted to their driving principles.

[0085] In some embodiments, converting the pulse control signal into a driving signal suitable for driving the corresponding flexible driving unit can specifically involve power amplification and / or energy form conversion of the multi-channel pulse control signal to generate a driving signal with sufficient voltage, field strength, or energy to drive the corresponding flexible driving unit to produce deformation. The driving signal is at least one of an electric field signal, an optical signal, a thermal signal, or a magnetic field signal.

[0086] In a specific implementation, the pulse control signal of each channel is first amplified by a pre-drive stage to enhance its driving capability; then, different processing paths are selected according to the type of flexible drive unit. It should be noted that the types of flexible drive units include electro-actuated, thermal-actuated, optical-actuated, and magnetic-actuated types, and the corresponding drive signals must be adapted to their driving principles.

[0087] For electrically actuated units (such as dielectric elastomers and piezoelectric ceramics), a charge pump, transformer, or switched capacitor boost circuit is used to amplify the voltage by tens to hundreds of times to generate a high-voltage electric field signal.

[0088] For photo-actuated units (such as photoresponsive hydrogels), electrical pulse signals are converted into light pulses of corresponding wavelengths through light-emitting diodes or laser diodes, and optical power is amplified by modulating the drive current during the conversion process.

[0089] For thermally actuated units (such as shape memory alloys and thermally expanding polymers), amplified pulse currents drive micro heating elements (such as metal thin film resistors and Joule heating wires) to convert electrical energy into heat energy, raising the temperature to the phase transition threshold.

[0090] For magneto-actuated units (such as magnetostrictive materials), the pulsed current is amplified and converted into a pulsed magnetic field through coils or electromagnets.

[0091] The above implementation ensures that the drive signal has sufficient energy to drive the load through power amplification, avoiding deformation failure due to insufficient drive capability; through energy form conversion, the same control method of the controller can be compatible with flexible drive units based on various physical principles, greatly expanding the versatility of the controller; at the same time, the configurability of voltage, field strength or energy allows the controller to flexibly adjust the output parameters according to the threshold characteristics of different drive units, achieving precise matching between drive energy and load requirements.

[0092] For example, when the flexible driving unit is a dielectric elastomer biomimetic muscle, the 3.3V pulse control signal is amplified (current increased to 100mA), and then the voltage is boosted to 800V through a cascaded charge pump to generate a high-voltage electric field signal applied to the dielectric elastomer electrode, causing the dielectric elastomer to contract rapidly. When the driving unit is a shape memory alloy wire, the same pulse control signal is amplified (pulse current increased from 10mA to 2A), and electrical energy is converted into Joule heat through a micro nickel-chromium alloy heating wire, causing the alloy wire temperature to rise from room temperature to 70℃ phase transition temperature within 50ms, resulting in contraction deformation. When the driving unit is a photoresponsive hydrogel, the pulse control signal is amplified to drive a high-power blue LED, outputting a light pulse with a wavelength of 450nm and a light intensity of 50mW / cm², causing local volume contraction of the hydrogel and achieving bending motion.

[0093] In some embodiments, the multi-channel driving circuit includes: a programmable gain amplifier for adjusting the amplitude of the driving signal of each output channel; a pulse width modulator for adjusting the pulse width of the driving signal of each output channel; and a timer for adjusting the duration of the driving signal of each output channel; wherein the programmable gain amplifier, the pulse width modulator, and the timer are configured to dynamically adjust the output parameters of the corresponding channel according to the target motion command or the feedback pulse sequence from the feedback interface.

[0094] In a specific implementation, the spiking neural network processing unit calculates the required deformation and deformation rate of each flexible drive unit in the current motion cycle based on the target motion command and real-time feedback signal. These requirements are then mapped to configuration values ​​for the output parameters: the deformation determines the total energy of the drive signal, achieved by adjusting the amplitude or pulse train duration; the deformation rate determines the instantaneous power, achieved by adjusting the pulse width or amplitude. The implementation of dynamic adjustment is explained in detail below with specific circuit structures and examples.

[0095] In some embodiments, during the process of converting multi-channel pulse control signals into drive signals, the output parameters of each drive signal corresponding to a channel are dynamically adjusted; wherein the output parameters include at least one of the following: amplitude, pulse width, or duration; the output parameters of the channel are associated with the deformation or deformation rate required by the corresponding flexible drive unit.

[0096] In a specific implementation, each drive channel integrates a programmable drive circuit, which includes: a digital control interface (such as an SPI or I2C slave), a configuration register set, a programmable gain amplifier (PGA), a pulse width modulator (PWM), and a high-precision timer / counter. The spiking neural network processing unit calculates the required deformation and deformation rate of each flexible drive unit in the current motion cycle based on the target motion command and real-time feedback signal through a cooperative control algorithm. These requirements are then mapped to configuration values ​​for the output parameters: the deformation determines the total energy of the drive signal, achieved by adjusting the amplitude or pulse train duration; the deformation rate determines the instantaneous power, achieved by adjusting the pulse width or amplitude. The mapping coefficients can be pre-calibrated and stored in the configuration register. Subsequently, the pulse neural network processing unit writes parameter values ​​to the configuration registers of each channel via the control bus: for example, writing an 8-bit binary number to the PGA gain control word to make the output voltage amplitude adjustable in approximately 4V steps within the range of 0~1000V; writing a count value to the PWM comparison register to make the output pulse width adjustable in 1μs resolution within the range of 10μs~10ms; and writing a count value to the timer reload value register to make the duration of the drive signal output programmable within the range of 100μs~1s. Each channel performs adjustment independently without interference. Through closed-loop dynamic adjustment, the drive energy received by each flexible drive unit is precisely matched to its instantaneous deformation requirements, avoiding deformation saturation, overshoot, or energy waste in traditional constant voltage drives, while allowing the same controller to drive hybrid flexible drive unit arrays of different sizes and materials.

[0097] For example, in a scenario where a bionic hand grasps an irregular object, the spiking neural network analyzes the contact force distribution in real time: the thumb requires large deformation (60° bend) but a low rate (30° / s), so the amplitude of the thumb channel is set to 800V, pulse width to 5ms, and duration to 200ms; the index finger requires moderate deformation (30° bend) and a high rate (60° / s), so the amplitude of the index finger channel is set to 500V, pulse width to 2ms, and duration to 50ms; the middle finger requires almost no deformation but needs a fast response, so the pulse width is set to 0.5ms and the amplitude to 200V. After each channel is adjusted independently, the bionic hand wraps around the object with a stable and coordinated grasping posture, and the grasping success rate and energy efficiency are better than the control group with uniform output parameters.

[0098] Additionally, it's important to note that for both optical and thermal driving methods, the "pulse width" and "duration" in the dynamically adjusted output parameters have specific physical meanings. In optical driving, the "pulse width" corresponds to the duration of a single light pulse (i.e., the conduction time of an LED or laser diode). A wider pulse width results in higher energy per irradiation, potentially causing greater local deformation in photoresponsive materials (such as photosensitive hydrogels). The "duration" corresponds to the total length of the light pulse train (i.e., the total irradiation time of multiple consecutive light pulses). A longer duration results in higher accumulated light energy, allowing the optical driving unit to continuously bend or contract. In thermal driving, the "pulse width" corresponds to the width of a single heating current pulse (i.e., the conduction time of a power MOSFET). A wider pulse width results in more energy injected by a single Joule heating pulse, leading to a higher instantaneous temperature rise in shape memory alloys or thermally expanding polymers. The "duration" corresponds to the total duration of the heating current pulse train (i.e., the total heating time of multiple pulses). A longer duration results in more accumulated heat, causing the driving unit's temperature to gradually rise to the phase transition point and maintain deformation. By adjusting the pulse width and duration separately, instantaneous power and total energy can be independently controlled, achieving precise decoupling of the deformation speed and amplitude of the optical / thermal actuation unit. For example, in an optically driven micro-gripper, when rapid bending with a small amplitude is required, a narrow pulse width (1ms) but high-frequency optical pulse train with a short duration (10ms) is used; when large-amplitude, slow bending is required, a wide pulse width (5ms) but low-frequency pulse train with a long duration (50ms) is used. In a thermally driven bionic finger, when rapid grasping is required, a large pulse width (20ms) high-current pulse is used to rapidly heat and contract the alloy wire with a short duration; when maintaining the grasping force is required, a narrow pulse width (5ms) low-current pulse train with a long duration (500ms) is used to maintain the temperature. This parameter decoupling allows the controller's control method to flexibly adapt to the different dynamic characteristics of the optical / thermal actuation unit.

[0099] In some embodiments, the multi-channel drive circuitry may be physically distributed on or near the flexible actuator body.

[0100] In specific implementations, each drive circuit, in the form of a bare chip or packaged chip, is directly fixed to the surface, inner wall, or embedded in the elastic material of the flexible actuator body through surface mounting or embedded processes. Each drive circuit is responsible for driving only one or a few flexible drive units closest to it. The drive circuits are connected to a common pulse neural network processing unit through flexible wires or conductive ink lines. This distributed layout significantly shortens the transmission path of the high-voltage drive signal, reduces electromagnetic interference, signal attenuation, and timing distortion caused by long-line transmission, and ensures that the cooperative relationship in the multi-channel pulse control signal can be transmitted to each drive unit with high fidelity. At the same time, the distributed circuit and the flexible actuator body form a mechanically compatible integrated structure, avoiding the traction constraint of the centralized drive circuit on the movement of the flexible body, and improving the integration and reliability of the system.

[0101] It should also be noted that this application adopts a one-to-one localized layout for the distributed driving circuits of photoactuated and thermally actuated units. For photoactuated units (e.g., photoresponsive hydrogels, photoflexible polymers), the corresponding driving circuit is a micro LED driver chip. This chip is integrated with the light-emitting diode (or laser diode) in the same flexible package, and is attached to the surface of the photoactuated unit or embedded in its substrate via a flexible circuit board or conductive ink lines. The pulse control signal output by the pulse neural network directly drives the chip, and the light intensity and exposure time of the LED are controlled by adjusting the pulse width or amplitude, so that the light pulse irradiates the driving unit with minimal transmission loss. For thermally actuated units (e.g., shape memory alloy wires, thermally expanding polymers), the corresponding driving circuit is a micro heating driver chip (including a power MOSFET and a current-limiting resistor), which can be integrated on a flexible circuit board and attached near the shape memory alloy wire, or directly bonded to the alloy wire with thermally conductive adhesive. The pulse control signal is amplified into a pulse current by the chip, which flows through the alloy wire to generate Joule heat. Because the driving circuit is in close contact with the alloy wire, the heat loss is minimal and the response is rapid. This distributed layout avoids light attenuation and line loss caused by long-distance transmission of optical signals or high currents, improving energy conversion efficiency and response speed. Simultaneously, each drive unit is independently configured with a drive circuit, facilitating local closed-loop temperature or light intensity control and enhancing the accuracy and consistency of multi-unit coordinated motion. For example, in a biomimetic soft crawling robot, multiple photo-bending units are arranged along both sides of the body, with a micro-module integrating an LED and a drive chip attached next to each unit. A pulsed neural network independently controls the light emission sequence of each module, causing the units on both sides to alternately bend and generate peristaltic waves. In a shape memory alloy-driven biomimetic hand, a micro-drive chip is welded near each alloy wire. The chip outputs pulsed current according to control signals, causing the alloy wire to rapidly heat and contract, achieving independent bending of the finger.

[0102] In some embodiments, the flexible actuator includes a plurality of independently driveable flexible drive units, and the output of the multi-channel drive circuit is configured to be connected to the plurality of flexible drive units.

[0103] In specific implementations, each output channel of the multi-channel drive circuit is electrically connected to a corresponding flexible drive unit through an independent conductive path (such as a flexible wire, conductive ink line, or embedded metal trace), forming a "one-to-one" independent drive link; alternatively, multiple drive units are grouped according to function (such as agonist muscle groups and antagonist muscle groups), and the multiple output channels of the multi-channel drive circuit are respectively connected to the drive units within each group to achieve grouped collaborative drive. Under this corresponding connection method, each flexible drive unit can be independently controlled, enabling precise allocation of drive energy according to the target motion command, and achieving coordinated coordination of each unit in terms of timing and amplitude; the independent drive link avoids crosstalk between channels, improves the fidelity of multi-channel signal transmission, and ensures that the timing relationship in the pulse control signal can be accurately transmitted to each drive unit; at the same time, the corresponding connection layout simplifies system wiring, reduces parasitic parameters, and is beneficial to improving the system's integration and reliability.

[0104] For example, the flexible actuator is a biomimetic muscle complex containing eight strip-shaped dielectric elastomer drive units arranged along the axial direction. The eight output channels of the multi-channel drive circuit are respectively connected to these eight drive units one by one through flexible conductive silver paste lines embedded in the silicone body. When a bending action is required, the controller outputs drive signals to five channels to cause the corresponding unit to contract and outputs drive signals to three channels to cause the corresponding unit to extend, according to a preset coordination mode. The timing and amplitude of the signals of each channel are precisely matched, so that the biomimetic muscle produces a smooth and controllable bending motion.

[0105] In some embodiments, the controller is configured to be at least partially disposed within the flexible actuator.

[0106] In a specific implementation, "at least part of the controller is located inside the flexible actuator" means that one or more components of the controller (e.g., but not limited to, at least one of a spiking neural network processing unit, a multi-channel drive circuit, a feedback interface, or a pulse coding module) are located within the internal space of the flexible actuator body, embedded in the material of the flexible actuator body, or attached to the inner surface of the flexible actuator body. Other components of the controller (such as external connectors, power management modules, or communication interfaces) may be located outside the flexible actuator or partially exposed.

[0107] In one embodiment, the spiking neural network processing unit and the multi-channel drive circuit are embedded in the silicone body of the flexible actuator in the form of a chip, while the pulse coding module and the feedback interface are connected as external modules via flexible cables.

[0108] In another implementation, the entire controller is attached to the inner wall of the flexible actuator in the form of a flexible circuit board and is covered by an outer material.

[0109] In some embodiments, the controller is integrated inside the flexible actuator, forming an integral structure with the flexible actuator.

[0110] In specific implementations, the core components of the controller (such as the pulse neural network processing unit, multi-channel drive circuit, and feedback interface) can be embedded in the body material of the flexible actuator as a packaged chip or module, or the controller can be entirely encapsulated in the internal cavity of the flexible actuator, making the controller and the flexible actuator physically inseparable. This integrated structure can significantly shorten the transmission path of pulse control signals and drive signals, reduce signal transmission delay and electromagnetic interference, and improve the accuracy of multi-channel signal timing; it eliminates external connectors and cables between the controller and the flexible actuator, enhancing the mechanical reliability and vibration resistance of the system; at the same time, the controller moves with the flexible actuator, improving the overall compliance and service life of the system.

[0111] For example, the controller can be entirely encapsulated inside the silicone body of the bionic muscle composite, with only the power supply and communication interfaces remaining on the outside of the controller. When the bionic muscle bends, the silicone body causes the internal controller to deform along with it, and the internal connection lines between the controller and the drive unit bend synchronously with the body. This ensures the long-term reliability of the entire actuator under repeated large deformations while achieving precise multi-channel collaborative control.

[0112] In some embodiments, the controller is a flexible controller, with at least a portion of its circuitry or substrate made of a flexible material to accommodate the deformation of the flexible actuator.

[0113] In specific implementations, the controller's circuitry is fabricated on a flexible circuit board (such as a polyimide substrate), giving the entire board bendable and foldable characteristics. Alternatively, the controller's core components (such as a pulse neural network processing unit or a multi-channel drive circuit) are directly integrated onto a stretchable elastic substrate (such as a silicone film or polyurethane film). The components are interconnected via stretchable conductive ink or serpentine wires, allowing the controller to stretch, bend, or twist along with the flexible actuator. This flexible design allows the controller to achieve physical mechanical compatibility with the flexible actuator, avoiding stress concentration or interface delamination caused by rigid structures during deformation. The flexible circuitry or substrate maintains reliable electrical connections even under repeated deformation, extending system lifespan. Furthermore, the flexible controller can be attached to the surface of the flexible actuator or embedded within it, eliminating the need for additional rigid support structures and further enhancing system integration and flexibility.

[0114] For example, a pulse neural network processing chip and a multi-channel driver chip are surface-mounted onto a polyimide flexible circuit board. The chips are interconnected by copper foil traces on the flexible circuit board. The entire flexible circuit board is attached to the outer surface of the bionic muscle composite. When the bionic muscle bends, the flexible circuit board bends accordingly. The mechanical stress borne by the chip and solder joint is much lower than that of the rigid circuit board. Thus, while ensuring accurate transmission of multi-channel signals, the controller and flexible actuator can achieve long-term stable operation under repeated large deformations.

[0115] In some embodiments, the controller is integrated on a flexible substrate as a system-on-a-chip and is configured to share a substrate or co-package with the body of the flexible actuator.

[0116] In specific implementations, core components such as the spiking neural network processing unit, multi-channel drive circuit, and feedback interface can be integrated into a single chip (System-on-a-Chip). This chip is directly fabricated or bonded to a flexible substrate (such as polyimide film, flexible glass, or elastomer film) to form a flexible chip assembly. Subsequently, the flexible chip assembly and the body of the flexible actuator are designed using a common substrate (i.e., the flexible substrate and the actuator body share the same substrate layer) or a co-package design (i.e., the flexible chip assembly and the actuator body are packaged in the same shell or covering layer), making the controller and actuator structurally integrated. With this integration method, the System-on-a-Chip significantly reduces the number of components and interconnections, significantly reduces the controller size, reduces parasitic parameters, and improves signal transmission integrity and anti-interference capability. The flexible substrate enables the controller to have mechanical compliance that matches the actuator body, and the common substrate or co-package design eliminates the physical interface between the controller and the actuator, avoiding stress concentration or interface delamination caused by differences in material stiffness during deformation, further improving the long-term reliability of the system under repeated large deformations.

[0117] For example, a pulse neural network processing unit and a multi-channel high-voltage drive circuit are integrated into a single system-on-a-chip (SoC). This chip is bonded to a polyimide flexible circuit board via flip-chip bonding to form a flexible chip assembly. Subsequently, the flexible chip assembly is placed inside the silicone body of the bionic muscle composite. Through a secondary injection molding process, the silicone body and the flexible chip assembly are co-encapsulated as one unit. When the formed bionic muscle performs bending, twisting, or other movements, the flexible chip assembly deforms synchronously with the silicone body. The interconnection between the chip and the flexible substrate remains stable under repeated deformation, thereby achieving precise multi-channel coordinated control while ensuring the reliability of the actuator during long-term dynamic use.

[0118] In some embodiments, the controller is configured to cause the first drive unit to undergo axial extension and contraction when a drive signal is applied to the first drive unit of the flexible actuator, and to cause the second drive unit to undergo bending deformation when a drive signal is applied to the second drive unit of the flexible actuator.

[0119] In a specific implementation, when a driving signal is applied to the first driving unit of the flexible actuator, the first driving unit undergoes axial extension and contraction deformation; when a driving signal is applied to the second driving unit of the flexible actuator, the second driving unit undergoes bending deformation. In the biomimetic muscle scenario, the flexible actuator simulates the collaborative working mechanism of the agonist and antagonist muscles of biological muscles: the first driving unit adopts a high axial stiffness dielectric elastomer columnar structure, embedded on the outer side of the biomimetic joint, and is responsible for rapid axial contraction (shortening) under the action of the driving signal, generating active tension, simulating the force exertion of the agonist muscle; the second driving unit adopts a low axial stiffness dielectric elastomer thin film structure, attached to the inner side of the joint, and is responsible for generating bending deformation under the action of the driving signal, simulating the damping and guiding effect of the antagonist muscle. The controller uses a pulse neural network to collaboratively generate two pulse control signals with a specific timing relationship: the first signal is amplified into a high-voltage electric field signal by a boost circuit, causing the first drive unit to contract axially and bend the joint; the second signal, after appropriate delay and amplitude adjustment, causes the second drive unit to bend, assisting in the bending direction and providing buffering at the end of the bend to prevent overshoot. This division of labor utilizes the different mechanical properties of the two drive units to achieve active-antagonistic synergy similar to biological muscles, giving the joint movement both sufficient driving force and compliant braking capability, avoiding the impact and instability caused by purely rigid drive. For example, in controlling the bending of a bionic elbow joint from an extended position to 90°, the spiking neural network, based on the "bend 90°" command, outputs a pulse sequence with a frequency of 8kHz, an amplitude of 900V, and a pulse width of 10ms to the first drive unit (outer high-stiffness column), causing it to axially contract by 20mm within 0-80ms, thus rapidly bending the elbow joint. Simultaneously, it outputs a pulse sequence with a frequency of 4kHz, an amplitude of 400V, a pulse width of 8ms, and a lag of 15ms to the second drive unit (inner low-stiffness film), causing it to bend within 80-120ms, providing gradually increasing counter-damping to accurately stop the joint at the 90° position, avoiding overshoot and rebound. Throughout the control process, the feedback interface monitors the bending angle in real time through a curvature sensor. If the detected angle exceeds 90°, it immediately increases the pulse width of the second drive unit (to 12ms) and decreases the amplitude of the first drive unit (to 700V), achieving closed-loop correction and ensuring positioning accuracy.

[0120] In some embodiments, the controller of the flexible actuator may further include: a feedback interface for receiving sensing signals from the flexible actuator and converting them into a feedback pulse sequence input to the spiking neural network processing unit.

[0121] In a specific implementation, the feedback interface, as a key component of the controller, receives sensing signals from the flexible actuator and converts them into a feedback pulse sequence, which is then input to the spiking neural network processing unit. Specifically, the feedback interface includes signal conditioning circuitry (such as amplification, filtering, and level conversion) and pulse coding circuitry. The signal conditioning circuitry first denoises, amplifies, and normalizes the analog or digital sensing signals output from various sensors distributed on the flexible actuator (such as flexible strain sensors, curvature sensors, and capacitive deformation sensors), obtaining continuous physical quantities reflecting the current deformation state (such as local curvature values, elongation, and contact force). Subsequently, the pulse coding circuitry converts these continuous physical quantities into a feedback pulse sequence with a specific pulse frequency, pulse density, or pulse interval according to a preset coding strategy (such as rate coding, time coding, or group coding). This feedback pulse sequence is input to the spiking neural network processing unit in real time as an adjustment signal for the network's internal state evolution, enabling the processing unit to dynamically adjust the generation strategy of the multi-channel pulse control signal based on the difference between the actual deformation and the target deformation. Based on this feedback mechanism, a complete closed-loop control system is formed, which significantly improves the motion accuracy and anti-interference capability of the flexible actuator. Through the unified information representation in the pulse domain, the feedback signal is naturally compatible with the internal processing logic of the pulse neural network processing unit, avoiding the information loss and delay caused by the repeated conversion between continuous signals and pulse signals in the traditional scheme. At the same time, the feedback interface supports multiple sensor types and encoding methods, which can flexibly adapt to the deformation sensing requirements of different flexible actuators.

[0122] For example, multiple flexible strain sensors are attached to the surface of the flexible actuator to detect the elongation in different regions. The feedback interface reads the resistance change value of each sensor at a sampling rate of 1kHz, and after conditioning, obtains a voltage signal of 0-3.3V. Then, through rate encoding, the voltage amplitude is linearly mapped to the pulse frequency (0V corresponds to 0Hz, 3.3V corresponds to 10kHz), generating a multi-channel feedback pulse sequence that is input to the spiking neural network processing unit. The error detection neuron in the processing unit compares the feedback pulse frequency with the target pulse frequency, and adjusts the output pulse timing of the extensor and flexor motor neurons according to the comparison result, so that the flexible actuator can quickly recover to the target deformation state when subjected to external disturbances, realizing high-precision closed-loop collaborative control.

[0123] In some embodiments, the feedback interface and the multi-channel drive circuit form a closed-loop control architecture in the flexible drive system. The multi-channel drive circuit, located in the forward channel, converts the control commands (low-voltage pulses) generated by the spiking neural network processing unit into powerful driving energy (high voltage / strong electric field), driving the flexible drive unit to deform. The feedback interface, located in the feedback channel, conditions and encodes the sensing signals generated by the actual deformation of the flexible actuator into a feedback pulse sequence, which is then sent back to the spiking neural network processing unit for correcting and optimizing subsequent control commands. Together, they form a complete closed loop of "command, drive, deformation, sensing, and correction" through the spiking neural network processing unit.

[0124] For example, taking a dielectric elastomer bionic muscle as an example, the driving circuit takes the 0-3.3V pulse control signal output from the pulse neural network, and gradually boosts it to an 800V high-voltage pulse through a charge pump circuit. This pulse is then applied to the flexible electrodes on both sides of the dielectric elastomer film, generating a strong electric field that compresses the film thickness and expands its area, thereby driving the bionic muscle to contract. A flexible strain sensor (resistance-changing type) attached to the surface of the dielectric elastomer changes its resistance value when the film deforms (e.g., from 10kΩ to 15kΩ). The bridge and differential amplifier in the signal conditioning circuit convert this resistance change into a 0-3.3V voltage signal (e.g., 1V→2V corresponds to a deformation of 0%→30%), which is then converted by the pulse encoding circuit into a feedback pulse sequence with a pulse frequency of 10kHz→20kHz, and input into the pulse neural network. The pulse neural network processing unit adjusts the density or timing of the output pulses at the next moment according to the deviation between the feedback pulse frequency and the target pulse frequency, so that the deformation of the dielectric elastomer bionic muscle gradually approaches the target value.

[0125] In some embodiments, the controller of the flexible actuator may further include: a configuration and parameter storage unit for directly supporting the flexible configuration and adaptive capabilities of the spiking neural network, enabling the controller to load different network topologies, synaptic weights and encoding mapping tables for different flexible actuators or motion tasks.

[0126] In a specific implementation, the configuration and parameter storage unit is integrated within the controller and connected to the spiking neural network processing unit, the multi-channel drive circuit, and the feedback interface via an internal bus. It stores topological parameters of the spiking neural network (such as the number of neurons, synaptic connection matrix, excitation / inhibition type), initial synaptic weight values, the encoding mapping table of the feedback interface, and the gain configuration of the drive circuit. Specifically, this configuration and parameter storage unit uses non-volatile memory (such as flash memory or EEPROM) as its physical carrier and, in conjunction with access control logic, allows the spiking neural network processing unit to automatically load pre-stored network parameters during power-on initialization. It also supports online reconfiguration via external communication interfaces (such as JTAG or SPI). This design enables the controller to be "reconfigurable" and "learnable": on the one hand, it can adapt to various types (dielectric elastomers, piezoelectrics, shape memory alloys) or different sizes of flexible actuators by loading different parameter files without replacing the hardware, and perform different motion tasks such as bending, twisting, and stretching; on the other hand, during system operation, the spiking neural network processing unit can update the synaptic weights or encoding mapping table in the storage unit online according to the difference between the feedback pulse sequence and the target instruction, so as to achieve adaptive optimization.

[0127] For example, the same controller chip was used to control a small flexible gripper and a large bionic muscle. For the gripper, the pre-installed network topology in the storage unit is a sparse connection structure with fast response, and the encoding mapping table maps the "grab" command to a high-frequency pulse sequence. For the bionic muscle, the CPG topology with a phase-coupled structure and the corresponding encoding table are reloaded through an external SPI interface. The original data in the storage unit is overwritten, and the controller switches its operating mode without any hardware modifications. Simultaneously, during use, the controller senses the slowed deformation response of the flexible actuator due to material fatigue through the feedback interface. The spiking neural network processing unit gradually increases the weight value of the corresponding synapse in the storage unit according to the error magnitude, increasing the output pulse frequency and compensating for the performance degradation caused by material aging, thus achieving online adaptive adjustment of the controller.

[0128] In some embodiments, the controller of the flexible actuator may further include: an SNN hardware accelerator for reducing the computational latency and power consumption of the spiking neural network processing unit, enabling higher frequency pulse inputs, more complex network topologies, and online learning rules (such as STDP).

[0129] In a specific implementation, the SNN hardware accelerator, as a functional module within the spiking neural network processing unit, is tightly coupled with the neuron node array, synaptic weight storage unit, and output pulse routing logic through a dedicated data path. This is used to accelerate computationally intensive or time-sensitive operations in the network. Specifically, the accelerator integrates multiple parallel multiply-accumulate units, a group of membrane potential accumulators, a fast comparator array, and event sequencing logic. When the pulse sequence input from the pulse encoding module arrives, the accelerator performs parallel weighted accumulation of synaptic currents, integral updates of membrane potentials, and threshold transconductance detection. This transforms tasks that would otherwise be processed cyclically by a general-purpose processor into a pipelined or combinational logic process within a single clock cycle. Simultaneously, the accelerator embeds a dedicated pulse timing-dependent plasticity (STDP) circuit, which can monitor the time difference between input and output pulses in real time and incrementally update synaptic weights according to preset rules. SNN hardware accelerators can significantly reduce the computational latency of spiking neural network processing units (reaching microsecond-level response), support higher pulse input frequencies (such as above 100kHz) and complex network topologies containing thousands of neurons; significantly reduce the power consumption of processing units (by 1 to 2 orders of magnitude compared to running SNN algorithms on general-purpose processors); enable online learning rules to take effect in real time without interrupting control, and enhance the controller's adaptability to conditions such as aging of flexible actuators and load changes.

[0130] For example, the flexible actuator needs to control 32 driving units simultaneously to generate biomimetic peristaltic motion. The SNN network contains 512 neurons and nearly 10,000 synaptic connections. The hardware accelerator parallelizes the membrane potential integration and threshold comparison operations of all neurons, completes a full network state update every 10 microseconds, and responds to changes in input pulses with a delay of less than 1 microsecond. At the same time, the STDP acceleration circuit continuously monitors the timing difference between the feedback pulses and the actual output pulses of the driving units, and fine-tunes the relevant synaptic weights every millisecond. This allows the biomimetic peristalsis to automatically optimize the waveform with the lowest energy consumption after dozens of cycles. The power consumption of the entire processing unit during operation is only 15mW, achieving high energy efficiency and high real-time collaborative control.

[0131] In some embodiments, the controller of the flexible actuator may further include: a clock unit for providing a high-precision, low-jitter time reference for the spiking neural network processing unit, ensuring the accuracy and repeatability of neuronal membrane potential integration, pulse firing timing, and phase relationship of multi-channel output signals.

[0132] In a specific implementation, the clock unit is integrated within the controller and connected to each neuron node, synaptic circuit, and output pulse routing logic in the spiking neural network processing unit via a dedicated clock tree distribution network. This provides a high-precision, low-jitter time reference for the dynamic evolution of the entire spiking neural network. Specifically, the clock unit can use a high-stability crystal oscillator (such as a temperature-compensated crystal oscillator, TCXO) or a phase-locked loop frequency multiplier circuit to generate the master clock signal. Different frequency clock domains are generated through frequency division / multiplication chains: for example, providing a microsecond-level resolution sampling clock for the membrane potential integrator, a nanosecond-level precision trigger pulse for the pulse output comparator, and a configurable synchronization clock for the multi-channel output to calibrate the phase relationship of the pulse signals in each channel. This clock unit also supports an external synchronization input interface, allowing multiple controllers to share the same reference clock, achieving global timing synchronization of a large-scale flexible drive unit array. By configuring this clock unit, it is possible to ensure that the integral of the membrane potential of each neuron and the comparison of the threshold are performed under a strictly uniform time step, so that the repeatability error of the pulse firing time is controlled at the nanosecond level; it ensures that the preset phase difference between the multi-channel output signals (such as the extensor channel leading the flexor channel by 5ms) remains stable during long-term operation or under different ambient temperatures, avoiding coordination deformation disorder caused by clock drift; at the same time, the high-precision clock reference provides an accurate time difference measurement basis for online learning based on pulse timing-dependent plasticity (STDP).

[0133] For example, the bionic muscle requires the two driving units to alternately contract with a 180-degree phase difference to generate oscillating motion. The clock unit uses a 24MHz temperature-compensated crystal oscillator as the master clock, and generates a 1MHz membrane potential update clock and a 100Hz output synchronization clock through frequency division. The left and right groups of neurons in the spiking neural network processing unit update the membrane potential based on this. In 10,000 consecutive oscillation cycles, the phase difference of the output pulses of the extensor and flexor channels is measured to be stable within the range of 180°±0.5°, without cumulative drift. When the ambient temperature changes from 0℃ to 50℃, the frequency drift of the temperature-compensated crystal oscillator is less than ±2ppm, and the phase difference change is less than 0.1°, ensuring the consistency of the bionic muscle's movement over a wide temperature range without the need for additional external synchronization signal intervention.

[0134] In some embodiments, the controller of the flexible actuator may further include a programmable gain and filtering circuit that dynamically adjusts the amplification factor of the feedback signal and filters out noise based on the sensor type and deformation amplitude.

[0135] In a specific implementation, the programmable gain and filter circuit is integrated within the feedback interface, located between the sensor signal input and the pulse code circuit. It is used to dynamically condition the raw signals output from various sensors on the flexible actuator. Specifically, this programmable gain and filter circuit includes a programmable gain amplifier (PGA) and a configurable active filter. The PGA adjusts its feedback resistor network via a digital control interface (such as SPI or parallel control lines) to achieve step or continuous gain adjustment (e.g., from 1x to 1000x). The configurable filter employs a switched capacitor or op-amp-resistor array structure and can be set to low-pass, band-pass, or high-pass modes as needed, with its cutoff frequency configured by software. The control logic automatically or via instructions from the pulse neural network processing unit switches the gain and filter parameters based on the currently selected sensor type (e.g., strain gauge, capacitive sensor, piezoelectric sensor) and the detected deformation range. Programmable gain and filtering circuits enable the same feedback interface to adapt to various flexible sensors with different output amplitudes and frequency characteristics without replacing hardware. By dynamically adjusting the gain, it ensures that the feedback signal is always within the optimal input range of the pulse code circuit (e.g., 0-3.3V) under different deformation amplitudes, avoiding quantization distortion at small signals or saturation at large signals. Configurable filters effectively remove mechanical vibration noise (typically tens to hundreds of hertz) and electrical high-frequency noise coupled from the drive circuit introduced during the movement of the flexible actuator, improving the signal-to-noise ratio of the feedback signal. This provides accurate and stable deformation state information to the pulse neural network processing unit, which is a key guarantee for high-precision closed-loop control.

[0136] For example, the same controller is used sequentially for two different flexible actuators: the first uses a flexible strain gauge (full-scale output ±20mV), and the second uses a capacitive sensor (full-scale output 0-5V). When the strain gauge is connected, the programmable gain and filter circuit are set to a gain of 100x and a low-pass filter cutoff frequency of 100Hz to suppress 50Hz power frequency interference introduced by human touch; when the capacitive sensor is connected, the gain is set to 1x and the filter is configured as a bandpass (10Hz-500Hz) to filter out high-frequency ripple introduced by the drive pulse. The pulse neural network processing unit automatically adjusts these parameters via the SPI bus based on the stability and amplitude range of the feedback pulse sequence, achieving plug-and-play and high-precision closed-loop control for different sensors.

[0137] The controller of the flexible actuator provided in this application will be described in detail below through several specific embodiments.

[0138] Example 1

[0139] This embodiment provides a flexible actuator controller for a bionic finger, which is configured to be embedded inside the silicone knuckle of the bionic finger and has a size of 5mm × 5mm × 1mm. The controller includes: a pulse coding module that converts the target motion command of "bending 30°" into a pulse sequence with a frequency of 5kHz and a duration of 100ms; a spiking neural network processing unit that adopts a spiking neural network topology with a cooperative control layer (containing 8 input neurons, 16 intermediate neurons, and 2 output neurons), which, after receiving the pulse sequence, collaboratively generates two pulse control signals based on internal state evolution—an extensor channel outputting a pulse sequence with a frequency of 5kHz and an amplitude reference of 600V, and a flexor channel outputting a pulse sequence with the same frequency but a lag of 8ms and an amplitude of 300V; and a multi-channel drive circuit including two miniature boost circuits (each measuring 1.5mm × 1.5mm), which respectively adopt a three-stage charge pump topology to boost the 3.3V logic level to 600V and 300V step by step, and are embedded in the inner wall of the silicone knuckle. Each boost circuit is connected to the corresponding dielectric elastomer drive unit through a short-pitch silver paste line. The controller also integrates a feedback interface, connecting to a flexible curvature sensor. This sensor converts the actual bending angle into a feedback pulse sequence (30° corresponds to 30kHz), which is then input to the pulse neural network processing unit. An error detection circuit compares the feedback frequency with the target frequency. When the deviation exceeds ±2kHz, the gain of the boost circuit is adjusted via the I2C bus (in 20V steps). The miniaturized design of this embodiment allows the controller to be completely embedded inside the finger. The distributed boost circuit shortens the high-voltage path, and the collaborative generation and closed-loop feedback achieve precise and smooth bending control of the finger. For example, when grasping a strawberry, the controller drives the extensor channel to contract rapidly first, while the flexor channel provides damping with a lag. This feedback adjustment stabilizes the bending angle at 30°±1°, with a contact force ≤0.5N, preventing damage to the fruit.

[0140] Example 2

[0141] This embodiment provides a flexible actuator controller for a bionic arm. The controller adopts a modular design, measuring 40mm × 30mm × 10mm, and is installed inside the upper arm shell. The controller includes: a pulse encoding module, which converts the "rapid lift 60°" command into a pulse sequence with a frequency of 8kHz and a pulse width of 10ms; a spiking neural network processing unit, configured with a central pattern generator (CPG) topology, containing 6 groups of cross-inhibitory neurons, which autonomously generate 12 channels of pulse control signals (corresponding to 3 groups of muscle bundles, each group having 4 extension units + 2 flexion units). In the initial stage, all extension channels output an amplitude of 900V, and the flexion channel outputs an amplitude of 400V with a lag of 15ms; a multi-channel drive circuit containing 12 boost circuits, using a multi-phase interleaved Boost topology, with a total output power of 50W. Each boost circuit is connected to the corresponding dielectric elastomer muscle bundle via a flexible cable, and the boost circuits are distributed on a flexible circuit board on the inner side of the upper arm. The controller integrates an inertial measurement unit feedback interface to detect the arm's angle and angular velocity in real time, dynamically adjusting the parameters of each channel: maintaining full-amplitude output during the rapid lifting phase (0-100ms), reducing the central group amplitude to 600V and the side groups to 500V when approaching the target (100-200ms), reducing the amplitude of the extension channel to 400V and the pulse width to 2ms during the holding phase, and stopping output in the bending channel. The modular design of this embodiment facilitates integration and maintenance, while CPG auto-evolution reduces computational burden, and dynamic energy allocation achieves rapid response and energy saving. For example, when lifting a 5kg dumbbell, the controller raises the arm from 0° to 45° within 0-100ms and precisely stops it at 60° within 100-200ms, resisting disturbances during the holding phase, with overall power consumption reduced by 22% compared to constant voltage drive.

[0142] Example 3

[0143] This embodiment provides a controller for a hybrid flexible actuator. The controller is integrated on a flexible polyimide substrate as a system-on-a-chip and co-packaged with the flexible actuator body. The overall size is 10mm × 10mm × 0.5mm, and it can be bent to a radius of curvature of 5mm. The controller includes: a pulse encoding module, which converts target motion commands (such as "bend 30° and grasp simultaneously") into a pulse sequence with a frequency of 4kHz; a spiking neural network processing unit, which uses a collaborative control layer (containing 12 input neurons, 24 intermediate neurons, and 4 output neurons) to collaboratively generate 4-channel pulse control signals based on internal state evolution; and a multi-channel driving circuit containing four sub-circuits: the first channel uses a charge pump boost circuit (outputting a 600V high-voltage electric field), the second channel uses an LED driving circuit (outputting 450nm blue light pulses with a light intensity of 50mW / cm²), the third channel uses an H-bridge current amplification circuit (outputting a 2A heating current), and the fourth channel uses a coil driving circuit (outputting a 0.5T pulsed magnetic field). The controller also integrates feedback interfaces, connecting to a flexible curvature sensor, a photoresistor, a thermocouple, and a Hall sensor to convert various deformation / state information into feedback pulse sequences. The spiking neural network processing unit dynamically adjusts the output parameters (amplitude, pulse width, and duration) of each channel based on the feedback, controlling the first channel to drive the dielectric elastomer to produce axial extension, the second channel to drive the hydrogel to produce bending, the third channel to drive the shape memory alloy to produce contraction, and the fourth channel to drive the magnetostrictive material to produce torsion. This embodiment's single controller is compatible with four driving principles: electric, optical, thermal, and magnetic, making it suitable for hybrid flexible actuators. The flexible substrate and co-encapsulation design allow the controller to deform with the actuator, achieving high integration. Multi-physics collaborative control expands the functional range of flexible robots. For example, in a biomimetic octopus tentacle, the controller simultaneously drives the dielectric elastomer (rapid bending), the shape memory alloy (locking posture), and the hydrogel (adhering to objects), realizing a complete action chain of grasping, locking, and releasing.

[0144] Figure 2 This is a schematic diagram of the structure of a flexible actuator control system provided in one embodiment of this application.

[0145] Reference Figure 2 As shown in the figure, this application embodiment also provides a flexible actuator control system 200, including:

[0146] Multiple subsystems 201, each subsystem 201 corresponding to a biomimetic part or a motion area of ​​the biomimetic robot;

[0147] Each subsystem includes:

[0148] Multiple controllers 100;

[0149] Multiple flexible actuators, each of which is connected to a corresponding controller 100 and is independently controlled by that corresponding controller 100;

[0150] The multiple controllers 100 are distributed within the subsystem 201, and are respectively located inside or near the flexible actuator they control.

[0151] In some embodiments, the flexible actuator control system 200 adopts a hierarchical distributed architecture: First, the bionic robot is divided into multiple subsystems 201, each subsystem corresponding to an independent bionic part (such as a bionic arm, bionic hand, or bionic leg) or a motion area (such as an elbow joint or wrist joint); each subsystem contains multiple controllers 100 and multiple flexible actuators, each controller is connected one-to-one with the flexible actuator it controls, and the multiple controllers are distributed within the subsystem, that is, each controller is physically located inside (e.g., embedded in a silicone body) or adjacent to (e.g., attached to a surface) the flexible actuator it controls, thus forming a tightly integrated "one-to-one" unit. The subsystem division enables modular design and independent debugging of complex bionic robots, reducing system coupling; the distributed layout makes the drive path between each controller and its corresponding flexible actuator extremely short, significantly reducing high-voltage signal transmission loss and electromagnetic interference, and improving the response speed and timing accuracy of multi-actuator collaborative control; at the same time, each subsystem can be independently powered and communicated, facilitating fault isolation and maintenance, and significantly improving the overall system's scalability and reliability. For example, in a bionic arm system, there are three subsystems: "upper arm", "forearm" and "hand". Each subsystem contains multiple micro-controllers, which are embedded in the corresponding bionic muscles. Each controller independently drives a muscle, and the whole system works together to complete complex movements such as swinging the arm and grasping.

[0152] In some embodiments, multiple subsystems are connected via a synchronization bus or wireless communication interface to achieve motion coordination between the subsystems.

[0153] In specific implementations, multiple subsystems are interconnected via a synchronization bus (such as CAN bus, EtherCAT, or a dedicated LVDS synchronization link) or a wireless communication interface (such as UWB or Bluetooth Low Energy synchronization protocol) to form a unified motion coordination network. Specifically, the main controller or dedicated synchronization interface module within each subsystem receives global motion commands and broadcasts time base signals and motion parameters with microsecond-level precision via the synchronization bus, ensuring that the spiking neural network processing units of each subsystem evolve collaboratively under the same time base. For wireless solutions, a high-precision time synchronization protocol (such as IEEE 802.1AS) is used to correct clock deviations between subsystems and to periodically exchange status information. This interconnected architecture enables multiple subsystems distributed across different bionic parts (such as arms and legs) to coordinate movements according to preset phase relationships and force ratios, avoiding motion misalignment or conflicts caused by independent control. This supports the bionic robot in completing complex and smooth overall movements (such as coordinated carrying with both arms and alternating walking with both legs). For example, in the walking control of a bipedal robot, the left and right leg subsystems exchange joint angle and plantar pressure feedback at a frequency of 1kHz via a CAN synchronization bus and share the same gait phase clock. When the left leg enters the swing phase, the synchronization bus sends a support phase trigger signal to the right leg subsystem, and the pulse neural networks of the hip and knee joints on both sides automatically adjust the relative phase difference of 180° to achieve stable alternating steps.

[0154] In some embodiments, at least a portion of the multiple controllers of the subsystem are configured as master controllers, and the remainder are slave controllers; the master controller is used to receive the total target motion command, decompose it into sub-target motion commands for each slave controller, and distribute them to the corresponding slave controller.

[0155] In a specific implementation, the master controller receives the overall target motion command from the upper layer (such as a central planner or user instructions) via a high-speed communication interface (such as SPI, CAN, or a dedicated synchronous bus). Then, based on the kinematic models and task allocation strategies of the flexible actuators connected to each slave controller, it decomposes the overall target motion command into multiple sub-target motion commands (such as angles, angular velocities, forces / torques of each joint), and distributes these sub-target commands to the corresponding slave controllers via a communication network. Each slave controller independently executes the received sub-target commands, driving its connected flexible actuators to produce local motion, while simultaneously feeding back its own state information (such as actual deformation, current, temperature, etc.) to the master controller for closed-loop adjustment and fault monitoring. This architecture centralizes complex global motion planning in the master controller, reducing the computational burden on the slave controllers and allowing each slave controller to focus on local drive control, thus improving the system's real-time performance and reliability. Furthermore, master-slave collaboration enables multiple flexible actuators to coordinate their actions according to a unified motion intention, avoiding timing conflicts or motion interference caused by independent decision-making among the slave controllers. For example, in the elbow joint subsystem of a bionic arm, the main controller receives the general command "bend 60°" and, based on the mechanical model of the lateral extensor muscles and medial flexor muscles of the elbow joint, decomposes the command into two sub-commands: "lateral drive unit contracts 30%" and "medial drive unit relaxes 20%", which are then sent to the corresponding slave controllers. The two slave controllers independently drive their respective flexible drive units to work together to complete the smooth bending motion of the elbow joint.

[0156] In some embodiments, the plurality of flexible actuators include different types of drive units, and the controller outputs an adapted drive signal according to the type of each flexible actuator.

[0157] In a specific implementation, the flexible actuator control system may include multiple flexible actuators that can be driven by different types of driving units, such as electro-actuated (dielectric elastomers, piezoelectric ceramics), photo-actuated (photoresponsive hydrogels), thermally actuated (shape memory alloys), or magnetically actuated (magnetostrictive materials). The controller, based on the type of each flexible actuator, outputs an adapted driving signal through its internally integrated corresponding conversion circuits (such as electro-optic, electro-thermal, and electro-magnetic conversion circuits, as well as boost circuits)—a high-voltage electric field signal for electro-actuated units, a light pulse of a specific wavelength for photo-actuated units, a heating current for thermally actuated units, and a pulsed magnetic field for magnetically actuated units. This adaptation mechanism is closely linked to the controller's multi-channel driving circuit: the pulse neural network processing unit pre-selects the corresponding driving circuit path based on the target motion command and the flexible actuator type identifier, and further finely adjusts the energy characteristics of the driving signal through programmable parameters (such as amplitude, pulse width, and duration) to ensure that each type of driving unit receives the most suitable excitation for its physical principles. A single controller can be compatible with flexible actuators based on various driving principles, eliminating the need for separate controller design for each type and greatly improving the system's versatility and scalability. For example, in a bionic hand, the thumb uses a dielectric elastomer (electro-actuated), while the index finger uses a shape memory alloy (thermal-actuated). The controller automatically configures a boost circuit to output a 600V high-voltage electric field for the thumb channel and a heating drive circuit to output a 2A current pulse for the index finger channel. The two work together to complete the grasping action.

[0158] In some embodiments, the control system further includes an upper-level coordinating controller, which is connected to all subsystems and is used to send motion commands to each subsystem according to the global task plan and collect feedback status information from each subsystem.

[0159] In a specific implementation, a global task coordinator is also provided in the control system. This coordinator connects to all subsystems (such as the bionic arm subsystem, bionic hand subsystem, and bionic leg subsystem) via a synchronous bus or wireless network, forming a control architecture that combines centralized planning and distributed execution. Specifically, the global task coordinator receives the global task plan (such as "moving an object to a designated location"), decomposes the global task into sub-tasks for each subsystem based on the kinematic model, current state, and task priority of each subsystem, and generates corresponding motion commands (such as target angle, speed, force, etc.). These commands are then sent to the main controller of each subsystem via a communication interface (the main controller further decomposes the commands into slave controllers). Simultaneously, the global task coordinator collects real-time status information from each subsystem (such as joint angles, load force, energy consumption, fault codes, etc.) for global status monitoring and task adjustment. This architecture separates high-level decision-making from low-level execution in complex biomimetic robot systems. The global task coordinator focuses on task planning and global coordination, the main controller within a subsystem focuses on subsystem-level instruction decomposition and coordination, and the slave controller focuses on the drive control of individual flexible actuators. This forms a three-tiered hierarchical control system of "global coordination—subsystem coordination—single-drive control," improving the system's modularity and scalability. Simultaneously, by collecting global feedback information, the global task coordinator can dynamically adjust task allocation and replan in case of failures or load changes, enhancing the system's robustness and fault tolerance. For example, in a bipedal robot's handling task, the global task coordinator plans the walking path based on visual feedback, sends stepping commands to the main controllers of the left and right leg subsystems respectively, and monitors plantar pressure feedback in real time. When slippage of the left leg is detected, the global task coordinator immediately sends a command to the main controller of the right leg subsystem to adjust the support phase time and recalculates gait parameters to ensure the robot maintains balance.

[0160] The flexible actuator control system provided in this application will be described in detail below through a specific embodiment.

[0161] Example 4

[0162] This embodiment provides a flexible actuator control system, which includes three subsystems corresponding to the shoulder, elbow, and wrist joints of a bionic arm. Each subsystem contains multiple controllers as described in Embodiment 1 or 2, and corresponding flexible actuators (dielectric elastomer muscle bundles). Specifically, the elbow joint subsystem includes six controllers, each embedded within a bionic muscle, independently driving one muscle. The six controllers are interconnected via a CAN synchronization bus at a 1kHz frequency, with one controller configured as the master controller and the others as slave controllers. The master controller receives the overall command "bend 60°" from the global task coordinator, decomposes it into six sub-target commands based on the muscle's mechanical model (e.g., 30% contraction of the three outer muscles and 20% relaxation of the three inner muscles), and distributes them to each slave controller via the CAN bus. Each slave controller independently drives its corresponding muscle and simultaneously feeds back the actual deformation, current, temperature, and other statuses to the master controller, which performs consistency checks and fault monitoring. The master-slave distributed architecture of this embodiment enables collaborative control of complex movements. Each slave controller focuses on local drive, improving real-time performance and reliability. CAN bus synchronization ensures the timing accuracy of each muscle movement (error <100μs). For example, during elbow flexion, the master controller monitors in real time insufficient contraction of the lateral muscles and immediately adjusts the drive amplitude of the corresponding slave controller (from 600V to 700V), while simultaneously fine-tuning the delay of the medial muscles to ensure smooth joint alignment and avoid flexion deviations caused by differences in muscle performance.

[0163] This application also provides a bionic robot, which may include the flexible actuator control system provided in any embodiment of this application.

[0164] In one implementation, each moving part of the bionic robot (such as arms, fingers, legs, face, etc.) integrates a corresponding subsystem. Multiple controllers and flexible actuators are distributed within each subsystem, and they are connected to a global task coordinator via a synchronous bus or wireless network, forming a complete control link from global planning to local coordination and then to single-point actuation. This robot utilizes the lightweight and compliant characteristics of the flexible actuators and the high-precision coordination capabilities of the controllers to achieve more natural, safe, and efficient movement, making it particularly suitable for scenarios with high requirements for compliance and adaptability, such as human-computer interaction, medical rehabilitation, and special detection.

[0165] The bionic robot provided in this application will be described in detail below through specific embodiments.

[0166] Example 5

[0167] This embodiment provides a biomimetic robot, specifically a bipedal biomimetic robot, including a flexible actuator control system as described in Embodiment 3. The robot comprises a left leg subsystem, a right leg subsystem, a torso subsystem, and a dual-arm subsystem. Each subsystem contains multiple controllers and flexible actuators (dielectric elastomer muscles, shape memory alloy wires, photoresponsive hydrogels, etc.) distributed throughout. A global task coordinator (central planning controller) connects to the main controllers of each subsystem via an EtherCAT synchronization bus, exchanging data at a 2kHz frequency. The global task coordinator performs gait planning based on visual and inertial sensor information, decomposing the "moving forward 1 meter" task into sub-tasks such as left leg stepping, right leg stepping, torso leaning forward, and arm swinging, which are then sent to the corresponding main controllers of the subsystems. Each main controller further decomposes the task into slave controllers, driving the coordinated movement of the biomimetic muscles of the hip, knee, and ankle joints. Simultaneously, the global task coordinator collects real-time feedback from each subsystem, including plantar pressure, joint angles, and motor current, dynamically adjusting gait parameters. The beneficial effects of this embodiment are as follows: the three-level hierarchical control system (global coordination—subsystem collaboration—single-point drive) achieves stable control of complex bipedal movements; the combination of flexible actuators and distributed controllers gives the robot advantages such as impact resistance, low energy consumption, and high biomimeticity. For example, during walking, the global task coordinator detects abnormal pressure on the left leg (slipping) and immediately shifts the center of gravity to the right leg in the next gait cycle, adjusting the support phase time of the right leg from 0.5s to 0.7s, while simultaneously sending a command to the left leg subsystem to reduce the stride amplitude. The robot stably regains its balance and avoids falling. This robot can be applied to scenarios such as rescue and detection, and home services.

[0168] The above embodiments are merely illustrative of the technical solutions of this application. Those skilled in the art will understand that, without departing from the spirit and scope defined by the claims, various combinations, modifications, and substitutions can be made to the specific materials, shapes, sizes, arrangements, and control parameters of the driving component and the constraint guide. For example, the driving component can also be other electroactive polymers; the constraint guide can also be a unidirectional glass fiber or Kevlar fiber reinforced film; the number of driving units can be a single unit for simple driving, or more units for more complex motion synthesis; the control module can also be integrated onto the supporting substrate of the bionic nose wing or communicate with an external main controller. All these variations fall within the protection scope of this application.

Claims

1. A controller for a flexible actuator, characterized in that, The flexible actuator includes multiple independently driveable flexible drive units, including: The pulse coding module is used to convert target motion commands into pulse sequences; A spiking neural network processing unit is used to receive the pulse sequence and collaboratively generate multi-channel pulse control signals based on its internal state evolution; A multi-channel driving circuit is connected to the output of the pulse neural network processing unit. Each channel is used to receive a pulse control signal and convert it into a driving signal suitable for driving the corresponding flexible driving unit.

2. The controller according to claim 1, characterized in that, The controller is configured to be at least partially located inside the flexible actuator.

3. The controller according to claim 2, characterized in that, The controller is integrated inside the flexible actuator, forming an integral structure with the flexible actuator.

4. The controller according to claim 2, characterized in that, The controller is a flexible controller, at least part of which is made of flexible material to accommodate the deformation of the flexible actuator.

5. The controller according to claim 2, characterized in that, The controller is integrated on a flexible substrate as a system-on-a-chip and is configured to share a substrate or co-package with the body of the flexible actuator.

6. The controller according to claim 1, characterized in that, The spiking neural network processing unit includes a neural network topology that simulates a biological spinal cord central pattern generator, and is configured to autonomously generate a multi-channel pulse control signal with a preset timing relationship based on the input pulse sequence.

7. The controller according to claim 1, characterized in that, The spiking neural network processing unit has a collaborative control layer, which includes: Multiple input neurons are used to receive the pulse sequence; Multiple interneurons form excitatory and / or inhibitory connections with the multiple input neurons; Multiple output neurons are connected to multiple intermediate neurons, and the pulse activity of the multiple output neurons corresponds to the multi-channel pulse control signal respectively; The intermediate neuron is configured to collaboratively drive the multiple output neurons to generate pulses based on the pulse activity of the multiple input neurons, thereby forming a preset collaborative relationship among the multi-channel pulse control signals.

8. The controller according to claim 1, characterized in that, The spiking neural network processing unit is configured to: uniformly determine the generation strategy of pulse control signals for each channel according to the target motion command, so that the pulse control signals of each channel are correlated in terms of timing, frequency or pulse density, so as to jointly realize the motion output indicated by the target motion command.

9. The controller according to claim 1, characterized in that, The multi-channel driving circuit includes multiple boost circuits. Each boost circuit amplifies the pulse control signal output by the pulse neural network processing unit to a voltage or field strength sufficient to drive the corresponding flexible driving unit to produce deformation. The multiple boost circuits are physically configured to be distributed inside or on the surface of the flexible actuator body, and each boost circuit is connected to a corresponding flexible driving unit.

10. The controller according to claim 1, characterized in that, The multi-channel drive circuit includes at least one of the following energy form conversion circuits: An electro-optical conversion circuit is used to convert pulse control signals into optical drive signals; An electrothermal conversion circuit is used to convert pulse control signals into thermal drive signals; An electromagnetic conversion circuit is used to convert pulse control signals into magnetic drive signals.

11. The controller according to claim 1, characterized in that, The multi-channel driving circuit includes: A programmable gain amplifier used to adjust the amplitude of the drive signal for each output channel; A pulse width modulator is used to adjust the pulse width of the drive signal for each output channel. A timer is used to adjust the duration of the drive signals for each output channel; The programmable gain amplifier, pulse width modulator, and timer are configured to dynamically adjust the output parameters of the corresponding channels according to the target motion command or the feedback pulse sequence from the feedback interface.

12. The controller according to claim 1, characterized in that, Also includes: A feedback interface is used to receive sensing signals from the flexible actuator and convert them into a feedback pulse sequence, which is then input to the spiking neural network processing unit.

13. The controller according to claim 12, characterized in that, The spiking neural network processing unit further includes an error detection circuit, which is configured to: The feedback pulse sequence is compared with the target pulse sequence corresponding to the target motion command, and the synaptic weights inside the spiking neural network processing unit are adjusted or an adjustment signal is sent to the multi-channel driving circuit according to the comparison result, so as to change the gain, bias or timing relationship of each channel driving signal.

14. The controller according to claim 12, characterized in that, The spiking neural network processing unit is configured to dynamically adjust the coordination relationship between the multi-channel pulse control signals according to the feedback pulse sequence in order to adapt to the deformation state changes of the flexible actuator.

15. The controller according to claim 1, characterized in that, The controller is configured to: When the driving signal is applied to the first driving unit of the flexible actuator, the first driving unit is subjected to axial extension and contraction deformation. When the driving signal is applied to the second driving unit of the flexible actuator, the second driving unit is caused to undergo bending deformation.

16. The controller according to claim 1, characterized in that, The spiking neural network processing unit is configured to dynamically adjust the relative timing relationship between the pulse control signals of each channel according to the target motion command, so as to form dynamic antagonism among multiple flexible drive units.

17. A flexible actuator control system, characterized in that, include: Multiple subsystems, each corresponding to a bionic part or a motion area; Each subsystem includes: Multiple controllers as described in any one of claims 1 to 16; Multiple flexible actuators, each connected to a corresponding controller and independently controlled by the corresponding controller; The multiple controllers are distributed within the subsystem and are respectively located inside or near the flexible actuator they control.

18. The system according to claim 17, characterized in that, The multiple subsystems are connected via a synchronous bus or wireless communication interface to achieve motion coordination between the subsystems.

19. The system according to claim 17, characterized in that, At least a portion of the plurality of controllers are configured as master controllers, and the remainder as slave controllers; the master controller is used to receive the total target motion command, decompose it into sub-target motion commands for each slave controller, and distribute them to the corresponding slave controllers.

20. The system according to claim 17, characterized in that, The plurality of flexible actuators include different types of drive units, and the controller outputs an adapted drive signal according to the type of each flexible actuator.

21. The system according to claim 17, characterized in that, It also includes an upper-level coordination controller, which is connected to all subsystems and is used to send motion commands to each subsystem according to the global task plan and collect feedback status information from each subsystem.

22. A biomimetic robot, characterized in that, It includes at least one flexible actuator control system as described in any one of claims 17-21.

Images