Adaptive haptic feedback robot force-position hybrid control system and method thereof

Abstract

The application discloses a self-adaptive haptic feedback robot force-position hybrid control system and method thereof, and relates to the technical field of industrial robot precision control. The self-adaptive haptic feedback robot force-position hybrid control system comprises a bionic haptic perception module, a pulse coding module connected to the output end of the bionic haptic perception module, a force-position collaborative controller connected to the output end of the pulse coding module, and a dynamic impedance regulator connected to the output end of the force-position collaborative controller. The dynamic decoupling control of the contact force and the position error is realized through the double closed-loop architecture, the accurate control of the force and the position in the complex and changeable assembly environment is simultaneously considered for the robot, and the problem of mutual interference of the force and the position control in the traditional control method is avoided.

Description

Technical Field

[0001] This invention belongs to the field of precision control technology for industrial robots, and specifically refers to an adaptive tactile feedback robot force-position hybrid control system and method. Background Technology

[0002] Traditional robot assembly control suffers from drawbacks such as force control lag, limited sensing dimensions, and poor environmental adaptability. Traditional PID control struggles to adapt to sudden changes in contact stiffness, leading to force overshoot (typically >15%); a single six-dimensional force sensor cannot capture the details of pressure distribution on the contact surface; and fixed impedance parameters are ill-suited to the dynamic characteristics of assemblies made of different materials. While existing technologies have proposed multi-sensor fusion control, they have not resolved the timing matching problem between pulse signals and continuous control, thus failing to meet the requirements of micro-assembly. Summary of the Invention

[0003] The main objective of this invention is to provide an adaptive tactile feedback robot force-position hybrid control system and method to solve the problems in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a force-position hybrid control method for a precision assembly robot based on adaptive bionic tactile feedback, comprising the following steps:

[0005] S1. Bionic tactile perception: A flexible tactile sensor array is deployed on the surface of the robot's end effector to collect the pressure distribution matrix P(x,y,t) of the contact area in real time. Its spatial resolution is no greater than 1mm2. The surface is covered with a bionic microstructure layer (protrusion height h = 50μm ± 5μm, spacing d = 200μm).

[0006] S2, Pulse Neural Coding: Pressure signals are converted into spatiotemporal pulse sequences using a center-peripheral inhibition model, with a weight matrix W. i,j Satisfies a Gaussian difference distribution:

[0007]

[0008] Where σ = 0.8, σ c =1.6, λ=0.3, inhibition ratio is 2:1;

[0009] S3. Force-Position Coordinated Control: Construct a dual closed-loop control architecture, with the outer loop employing an adaptive sliding mode controller to track the position error e. p The inner loop dynamically adjusts the impedance parameters based on a spiking neural network (SNN) to generate a compensation force. The control law is:

[0010]

[0011] S4. Dynamic impedance adjustment: The virtual stiffness parameter is adjusted in real time according to the rate of change of the contact force gradient.

[0012]

[0013] The adjustment coefficient β satisfies the following relationship with the assembly material hardness E:

[0014]

[0015] Preferably, in step S1, the flexible tactile sensor array is composed of multilayer graphene piezoresistive units, with a single-point sensitivity S = ΔR / R0 ≥ 0.5% / kPa, temperature drift ≤ 0.01% / ℃, pressure detection range 0.1-10N, and nonlinear error ≤ 1.5%.

[0016] Preferably, in step S1, the ratio of the protrusion height to the spacing d of the biomimetic microstructure layer is / d = 0.25, where the protrusion height = 50μm ± 5μm and the spacing d = 200μm, which is used to enhance the contact force anisotropy detection capability.

[0017] Preferably, in step S2, the pulse triggering condition is:

[0018]

[0019] Where the dynamic baseline threshold Γ(t) = median(P(t-10ms:t)), the time differential gain η = 0.1, and the sliding window width k = 3.

[0020] The preferred dynamic adjustment rule for the parameters of the adaptive sliding mode controller in step S3 is as follows:

[0021]

[0022] Where e max =10μm, K p0 =120N / m, K d0 = 8 N·s / m.

[0023] Preferably, in step S3, the synaptic weight update rule of the spiking neural network (SNN) is:

[0024]

[0025] The time constant τ = 20ms, the learning rate η = 0.02, the sparsity coefficient γ = 0.05, and the network load rate ≤ 30%.

[0026] Preferably, in step S4, when the contact force gradient satisfies When this occurs, the emergency load reduction mode is triggered:

[0027] The virtual stiffness decreases from K(t) to 0.3K(t);

[0028] The pulse compensation gain is increased to 2α;

[0029] The position tracking error threshold is tightened to e max =5μm;

[0030] The exit condition for emergency load reduction mode is that the following conditions are met within 10 consecutive control cycles. And perform the recovery operation:

[0031] K(t)←min(1.2K(t),K0),α←0.5α.

[0032] A preferred adaptive haptic feedback robot force-position hybrid control system, characterized in that it includes:

[0033] A biomimetic tactile sensing module, comprising a flexible tactile sensor array (100×100 dot matrix, graphene piezoresistive unit) and a signal conditioning circuit (24-bit ADC, sampling rate ≥1kHz);

[0034] The pulse coding module includes an FPGA chip (Xi li nx Zynq UltraScale+) to implement a center-periphery suppression model and a hardware clock synchronization unit (IEEE 1588PTP protocol, synchronization error ≤1μs);

[0035] Force-position co-controller, which includes an adaptive sliding mode controller (TIC2000 series DSP) and a spiking neural network processor (Intel Loi Hi neuromorphic chip, 128 cores);

[0036] A dynamic impedance regulator, comprising a real-time parameter calculation unit (ARM Cortex-A72); and an emergency mode trigger circuit (comparator response time <2μs);

[0037] The preferred device communicates with the industrial robot controller via an EtherCAT bus and supports hot switching between OPC UA protocols. It automatically switches protocols when the network latency is >50ms, and the data packet loss rate is <0.1%.

[0038] Preferably, the spiking neural network processor has an energy efficiency ratio of ≥5 TOPS / W, and the synaptic weight storage adopts a non-volatile memristor array, supporting in-situ updates.

[0039] The adaptive tactile feedback robot force-position hybrid control system and method proposed in this invention have significant technical advantages and practical effects in real-world applications. First, this method, through a highly sensitive biomimetic tactile sensor array, can perceive the pressure distribution on the contact surface in real time and with precision, endowing the robot with "tactile sensation." During assembly, it can "sensor" minute pressure changes, achieving precise control of the contact force. Second, combined with a pulse neural coding mechanism, the pressure signal is converted into a spatiotemporal pulse sequence, effectively compressing the data volume and simulating the signal processing mode of a biological nervous system. The robot can process and respond to tactile information in a more biological manner, greatly improving the system's response speed and adaptability.

[0040] Furthermore, the force-position cooperative controller based on spiking neural networks (SNNs) achieves dynamic decoupling control of contact force and position error through a dual closed-loop architecture. This allows the robot to simultaneously achieve precise force and position control in complex and variable assembly environments, avoiding the problem of mutual interference between force and position control in traditional control methods. The introduction of a dynamic impedance adjustment algorithm enables the robot to adaptively adjust virtual stiffness parameters according to different assembly materials and changes in contact force, thereby achieving efficient and precise assembly of various materials.

[0041] This invention demonstrates immense potential and value in the field of precision assembly. This method not only effectively solves the problems of force overshoot and response lag inherent in traditional force control methods in micron-level assembly, but also significantly improves assembly success rate and accuracy. For example, in the semiconductor packaging field, this method can improve the yield rate of chip mounting, greatly reducing production costs and rework rates. Simultaneously, in optical assembly scenarios, this method can reduce manual calibration time and significantly improve production efficiency. Attached Figure Description

[0042] Figure 1 This is a flowchart of the steps of the present invention;

[0043] Figure 2 This is a system architecture diagram of the present invention;

[0044] Figure 3 This is the biomimetic tactile sensor structure of the present invention;

[0045] Figure 4 This is the control flowchart of the present invention;

[0046] Figure 5 This is the flowchart of the dual closed-loop control of the present invention. Detailed Implementation

[0047] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0048] This invention provides a force-position hybrid control method for an adaptive haptic feedback robot, comprising the following steps:

[0049] S1, Bionic tactile perception

[0050] 1. Sensor Deployment: A flexible tactile sensor array is embedded on the surface of the robot's end effector, employing multilayer graphene piezoresistive units with a spatial arrangement density of 25 points / cm². 2 Coverage area 10cm ×

[0051] 10cm.

[0052] The sensor surface is covered with a biomimetic microstructure layer with a protrusion height of 50μm and a spacing d of 200μm, forming a fingerprint-like mechanical filtering structure.

[0053] 2. Pressure signal acquisition: Real-time output of the pressure distribution matrix P(x,y,t) in the contact area, where:

[0054] x,y∈[1,100]: Spatial coordinates (corresponding to a 100×100 dot matrix)

[0055] t: timestamp (sampling rate f) s =1kHz)

[0056] The single-point pressure detection range is 0.1N≤P≤10N, and the nonlinear error is ≤1.5%.

[0057] 3. Anisotropy detection: The mechanical filtering effect of biomimetic microstructures enhances the distinction between normal and tangential forces.

[0058]

[0059] Where P x ,P y The pressure components are in the x / y directions.

[0060] S2, pulse neural coding

[0061] 1. Center-periphery suppression model: Spatial filtering is applied to the original pressure signal, with weight matrix W... i,j Satisfies a Gaussian difference distribution:

[0062]

[0063] Parameter settings: σ = 0.8, σ c =1.6, λ=0.3, suppression ratio 2:1. Improve the signal-to-noise ratio in the edge contact region (SNR improvement ≥10dB);

[0064] 2. Pulse generation rules: The dynamic pulse triggering condition is as follows:

[0065]

[0066] Γ(t) = median(P(t-10ms:t)): the baseline of the sliding window mid-range filter;

[0067] The pressure time-domain differential term enhances transient response.

[0068] 3. Pulse Sequence Output: Generates a sparse spatiotemporal pulse stream S(x,y,t), compressing the data volume to 15%-20% of the original signal. Simulates the "phasic-tonic" response mode of tactile nerves, adapting to SNN processing.

[0069] S3, Force-Position Coordinated Control

[0070] 1. Dual closed-loop architecture design: Outer loop position control:

[0071] Adaptive sliding mode controller is used to track the target trajectory θ d (t), position error e p =θ d -θ actual .

[0072] Dynamically adjust control parameters:

[0073]

[0074] Inner ring force control:

[0075] Real-time generation of compensating force using spiking neural networks (SNNs):

[0076]

[0077] Where τ i ∈[1,5]ms represents the synaptic delay, and the weight w i Update via STDP rules.

[0078] 2. Control Law Fusion: The total control force output is:

[0079]

[0080] The outer loop position control and inner loop force control are adjusted independently, solving the coupling oscillation problem of traditional PID control; Kp (t) and K d (t) Adaptive adjustment with error, the range of stiffness variation is expanded by 3 times; SNN compensates for high-frequency disturbances (such as tool head vibration), and the compensation force resolution reaches 0.01N.

[0081] S4, Dynamic Impedance Adjustment

[0082] 1. Virtual stiffness adjustment: Real-time calculation of contact force gradient Dynamically adjust virtual stiffness:

[0083]

[0084] K0 = 500 N / m: Nominal stiffness;

[0085] E: Elastic modulus of assembly material (e.g., silicon wafer) ).

[0086] 2. Emergency Load Reduction Mode: When hour:

[0087] The virtual stiffness is reduced to 0.3K(t) to avoid overshoot;

[0088] The pulse compensation gain is increased to α = 0.3, enhancing the anti-interference capability;

[0089] The position error threshold is tightened to e max =5μm;

[0090] Stiffness matching of seven types of materials, including silicon, glass, and ceramics, is achieved through β(E), thus enabling material self-adaptation.

[0091] The hyperbolic tangent function (tanh) ensures a smooth dynamic transition of stiffness, and the rate of change of stiffness...

[0092]

[0093] In emergency mode, the force overshoot is reduced to ≤2% to prevent damage to precision components.

[0094] This invention provides an adaptive haptic feedback robot force-position hybrid control system, comprising:

[0095] A biomimetic tactile sensing module, comprising a flexible tactile sensor array (100×100 dot matrix, graphene piezoresistive unit) and a signal conditioning circuit (24-bit ADC, sampling rate ≥1kHz);

[0096] The pulse coding module includes an FPGA chip (Xi li nx Zynq UltraScale+) to implement a center-periphery suppression model and a hardware clock synchronization unit (IEEE 1588PTP protocol, synchronization error ≤1μs);

[0097] Force-position co-controller, which includes an adaptive sliding mode controller (TIC2000 series DSP) and a spiking neural network processor (Intel Loi Hi neuromorphic chip, 128 cores);

[0098] A dynamic impedance regulator, comprising a real-time parameter calculation unit (ARM Cortex-A72); and an emergency mode trigger circuit (comparator response time <2μs);

[0099] The preferred device communicates with the industrial robot controller via an EtherCAT bus and supports hot switching between OPC UA protocols. It automatically switches protocols when the network latency is >50ms, and the data packet loss rate is <0.1%.

[0100] Preferably, the spiking neural network processor has an energy efficiency ratio of ≥5 TOPS / W, and the synaptic weight storage adopts a non-volatile memristor array, supporting in-situ updates.

[0101] Example: Semiconductor chip flip-chip bonding scenario

[0102] In a semiconductor packaging production line, high-precision chip mounting is required. Traditional assembly methods suffer from insufficient force control precision and slow response speed, resulting in a chip mounting yield of only 95%, and chip damage due to force overshoot is frequent. To address this, the present invention introduces a force-position hybrid control method for precision assembly robots based on adaptive bionic tactile feedback to improve assembly accuracy and success rate. The assembly objects are silicon-based processors (12mm × 12mm, solder ball diameter 80μm, spacing 150μm) and ceramic substrates (thermal expansion coefficient 6.5ppm / ℃, pad size 75μm).

[0103] Implementation process

[0104] S1, Bionic tactile perception

[0105] A flexible tactile sensor with a 100×100 dot matrix was deployed on the surface of the robot's end effector, employing multilayer graphene piezoresistive units, with a spatial resolution ≤0.8mm. 2The sensor surface is covered with a biomimetic microstructure layer, with protrusions 50 μm high and spaced 200 μm apart, forming a fingerprint-like mechanical filtering structure with an anisotropic detection coefficient ≥0.35. This design enables the sensor to acquire the pressure distribution matrix of the contact area in real time, with a single-point pressure detection range of 0.1-10 N, a sensitivity of 0.6% / kPa, and a temperature drift ≤0.01% / ℃. Through this highly sensitive biomimetic tactile perception, the robot can accurately sense pressure changes on the contact surface, providing fundamental data for subsequent force and position control.

[0106] S2, pulse neural coding

[0107] The acquired pressure signal is converted into a spatiotemporal pulse sequence using a center-periphery suppression model. The weight matrix follows a Gaussian difference distribution with a suppression ratio of 2:1. The weight matrix is ​​as follows:

[0108]

[0109] (Parameters: σ = 0.8, σ c =1.6, inhibition ratio 2:1).

[0110] The dynamic pulse triggering condition is:

[0111]

[0112] The dynamic baseline threshold P(t) is calculated using a sliding window mid-range filter, with a time differential gain k = 0.05 and a sliding window width of 5 ms. The generated pulse sequence data is compressed to 18% of the original signal, with a processing delay of 1.8 ms. This pulse neural coding mechanism not only effectively reduces the burden of data transmission and processing but also simulates the signal processing methods of biological nervous systems, enabling robots to process and respond to tactile information in a more biological manner.

[0113] S3, Force-Position Coordinated Control

[0114] A dual closed-loop control architecture was constructed. The outer loop uses an adaptive sliding mode controller to track the position error, while the inner loop dynamically adjusts the impedance parameters based on a spiking neural network (SNN) to generate compensation force.

[0115] outer ring sliding mode control position tracking error e p =θ d -θ actual The control law is:

[0116]

[0117] Where the error constraint is: e p ≤2μm,

[0118] The formula for generating SNN compensation force is:

[0119]

[0120] Synaptic weights are updated via STDP rules:

[0121]

[0122] This dual-closed-loop architecture enables dynamic decoupling control of contact force and position error, solving the problem of force and position coupling oscillation in traditional PID control.

[0123] S4, Dynamic Impedance Adjustment

[0124] The virtual stiffness parameters are adjusted in real time based on the rate of change of the contact force gradient. The stiffness adjustment formula is:

[0125]

[0126] The material hardness parameter β is calculated as follows:

[0127]

[0128] The silicon chip has an elastic modulus E = 170 GPa and a material hardness parameter β = 0.14. When the contact force gradient satisfies... When this occurs, an emergency load reduction mode is triggered: the virtual stiffness decreases from K(t) to 0.5K(t), the pulse compensation gain is increased to 1.5, and the position tracking error threshold is tightened to 1μm. Through this dynamic adjustment, stiffness matching for different materials (such as silicon, glass, ceramics, etc.) is achieved.

[0129] Implementation effect

[0130] This embodiment demonstrates the core advantages of the invention in a semiconductor flip-chip bonding scenario. It achieves 1.8μm accuracy through a haptic-SNN-sliding mode control closed loop, exhibiting ultra-fast response, completing emergency load reduction within 8ms to avoid damage to brittle materials, supporting cross-material adaptation, and enabling stiffness matching from silicon chips (170GPa) to flexible PCBs (2GPa). Energy efficiency is significantly improved, demonstrating great potential in high-precision assembly.

Claims

1. An adaptive haptic feedback robot force-position hybrid control method, characterized by: Includes the following steps: S1. Bionic Tactile Sensing: Deploy a flexible tactile sensor array on the surface of the robot's end effector to collect the pressure distribution matrix of the contact area in real time. Its spatial resolution is no greater than 1 mm², and its surface is covered with a biomimetic microstructure layer; S2, Pulse Neural Coding: Converts pressure signals into spatiotemporal pulse sequences using a center-peripheral inhibition model, with a weight matrix... Satisfies a Gaussian difference distribution: wherein , , , the inhibition ratio is 2:1; the pulse trigger condition is that wherein the dynamic baseline threshold , time differential gain , sliding window width ; S3. Force-Position Coordinated Control: Construct a dual closed-loop control architecture, with the outer loop employing an adaptive sliding mode controller to track position errors. The inner loop employs a spiking neural network processor, which dynamically adjusts the impedance parameters based on the spiking neural network to generate compensation force. The control law is: S4. Dynamic Impedance Adjustment: The virtual stiffness parameter is adjusted in real time according to the rate of change of the contact force gradient. where the adjustment factor with assembly material hardness satisfies: 。 2.The adaptive haptic feedback robot force-position hybrid control method of claim 1, wherein: In step S1, the flexible tactile sensor array is composed of multilayer graphene piezoresistive units, with single-point sensitivity... Temperature drift ≤0.01% / ℃, pressure detection range 0.1-10N, nonlinear error ≤1.5%. 3.The adaptive haptic feedback robot force-position hybrid control method of claim 1, wherein: In step S1, the protrusion height of the biomimetic microstructure layer Spacing The proportion is The height of the protrusion ,spacing It is used to enhance the ability to detect contact force anisotropy.

4. The adaptive haptic feedback robot force-position hybrid control method of claim 1, wherein: The dynamic parameter adjustment rules for the adaptive sliding mode controller in step S3 are as follows: wherein , , .

5. The adaptive haptic feedback robot force-position hybrid control method according to claim 1, wherein: In step S3, the synaptic weight update rule of the spiking neural network is as follows: where the time constant , learning rate , sparsification coefficient , network load rate ≤ 30%.

6. The adaptive haptic feedback robot force-position hybrid control method according to claim 1, wherein: In step S4, when the contact force gradient satisfies the emergency load shedding mode is triggered: virtual stiffness from ; The pulse compensation gain is raised to ; Position tracking error threshold tightened to ; The exit condition of the emergency load shedding mode is that the following is satisfied for 10 consecutive control cycles and the recovery operation is performed: 7.The adaptive tactile feedback hybrid force / position control system employed by the adaptive tactile feedback hybrid force / position control method of claim 1, wherein: include: A biomimetic tactile sensing module, comprising a flexible tactile sensor array and a signal conditioning circuit; A pulse coding module, wherein the pulse coding module includes an FPGA chip implementing a center-periphery suppression model and a hardware clock synchronization unit; A force-position co-controller, comprising an adaptive sliding mode controller and a spiking neural network processor; A dynamic impedance regulator, comprising a real-time parameter calculation unit and an emergency mode trigger circuit.

8. The adaptive haptic feedback robot force-position hybrid control system according to claim 7, wherein: The system communicates with the industrial robot controller via an EtherCAT bus and supports hot switching between OPC UA protocols. When the network latency is greater than 50ms, the protocol is automatically switched, and the data packet loss rate is less than 0.1%.

9. The adaptive haptic feedback robot force-position hybrid control system according to claim 7, wherein: The spiking neural network processor has an energy efficiency ratio of ≥5 TOPS / W, and the synaptic weight storage uses a non-volatile memristor array, supporting in-situ updates.

Images