Pneumatic artificial muscle biped walking control method and system based on imitation learning

Abstract

The invention discloses a pneumatic artificial muscle biped walking control method and system based on imitation learning, and the method comprises the steps: firstly collecting the walking posture data of a human body, and generating an expert track based on a Mujoo simulation platform; then training a strategy model by using an imitation learning algorithm, deploying the strategy model to an embedded controller, sending an action instruction to a PLC (Programmable Logic Controller) based on an action strategy, analyzing the action instruction by the PLC, generating a PWM (Pulse-Width Modulation) signal, and controlling the inflating and deflating time sequence of the lower limb pneumatic artificial muscle group; finally, lower limb joint angle and air pressure data are fed back in real time through a sensor, the PWM duty ratio is adjusted in a closed-loop mode, and the walking posture is optimized. The simulation learning algorithm and pneumatic artificial muscle driving are combined, the human walking mode is mapped, more natural and more stable bionic walking is achieved, and meanwhile the integration level and expandability of the system are improved. A PWM dynamic air pressure distribution mode is adopted, redundant energy loss is reduced, and the response speed and the control precision of the system are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of bionic robot control, and in particular relates to a pneumatic artificial muscle bipedal walking control method and system based on imitation learning. Background Art

[0002] In the field of biomimetic robotics, motion control technology for bipedal robots continues to advance. Bipedal robots currently on the market are primarily driven by biomimetic joint motors. While this motor-driven approach provides precise control, it has certain limitations, such as insufficient control flexibility and adaptability when navigating complex terrain and dynamic environments. European and American robotics companies initially employed hydraulic drive technology. While this technology excelled in power output and dynamic performance, the hydraulic system was complex and costly to maintain. Later, they transitioned to electric drive, which reduced system complexity but still left room for improvement in energy efficiency and response speed.

[0003] The present invention proposes an innovative bipedal walking control method that uses an imitation learning algorithm combined with PWM control technology to drive pneumatic artificial muscles. This method not only enables more natural and stable bionic walking, but also overcomes the difficulties of traditional control models in dealing with nonlinear dynamic problems. Through the imitation learning algorithm, the system can directly map human walking patterns without relying on complex mathematical models, thereby improving the flexibility and adaptability of control. At the same time, PWM control technology can dynamically allocate different duty cycles according to the coordinated contraction requirements of pneumatic artificial muscles, optimize muscle contraction and extension, and further enhance the naturalness and energy efficiency of movement.

[0004] This method also demonstrates innovation in the degrees of freedom and mathematical theory of motion control. Traditional mechanical dynamic control theory, typically based on rigid-body dynamics models, struggles to accurately describe and control systems with flexible and nonlinear characteristics. The introduction of pneumatic artificial muscles, however, allows the robot's lower limbs to possess greater degrees of freedom and compliance, enabling them to better simulate the complex movements of human muscles. Combined with imitation learning algorithms, this control approach enables effective control of complex, nonlinear systems without relying on precise mathematical models, thus advancing the development of motion control theory.

[0005] In summary, the present invention has obvious innovations and advantages in technical background, provides a new solution for the motion control of bipedal robots, and has broad application prospects and research value. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of the existing technology and propose a pneumatic artificial muscle bipedal walking control method and system based on imitation learning, aiming to achieve high-fidelity bipedal bionic walking through advanced control technology and innovative driving methods.

[0007] The object of the present invention is achieved through the following technical solution: a pneumatic artificial muscle bipedal walking control method based on imitation learning, the method comprising the following steps:

[0008] (1) Collect human walking posture data and generate expert trajectories based on the Mujoco simulation platform;

[0009] (2) Use imitation learning algorithms to train the policy model and deploy it to the embedded controller;

[0010] (3) The embedded controller sends action instructions to the PLC based on the action strategy output by the model;

[0011] (4) The PLC analyzes the motion instructions and generates PWM signals to control the inflation and deflation timing of the lower limb pneumatic artificial muscle group;

[0012] (5) Real-time feedback of lower limb joint angle and air pressure data is obtained through sensors, and the PWM duty cycle is adjusted in a closed loop to optimize walking posture.

[0013] Furthermore, the MIPI camera is used to collect RGB images of the human body while walking to perform posture estimation, extract the joint angle sequence, and obtain the human walking posture data.

[0014] Furthermore, the Mujoco platform is used to simulate human walking posture, generate expert trajectory data, and construct a training dataset for the imitation learning algorithm.

[0015] Furthermore, the imitation learning algorithm is behavior cloning (BC) or generative adversarial imitation learning (GAIL).

[0016] Furthermore, the imitation learning algorithm integrates the reinforcement learning mechanism to dynamically optimize the strategy model through environmental interaction.

[0017] Furthermore, the embedded controller is a Jetson Orin Nano 4GB, which is linked to the PLC through serial port communication.

[0018] Furthermore, the PWM signal adopts pulse width modulation technology to allocate different duty cycles according to the coordinated contraction requirements of the pneumatic artificial muscle group.

[0019] Furthermore, the closed-loop control adopts a PID algorithm, combined with multimodal data of the angle sensor and the pressure sensor.

[0020] Furthermore, the pneumatic artificial muscle group includes 14 groups of pneumatic artificial muscles, 4 groups corresponding to the anterior muscle groups of the quadriceps femoris, 3 groups corresponding to the posterior muscle groups of the hamstrings, and the rest are used for coordinated stability.

[0021] On the other hand, the present invention also provides a pneumatic artificial muscle bipedal walking control system based on imitation learning, the system comprising:

[0022] The data acquisition module is used to collect human walking posture data and generate expert trajectories based on the Mujoco simulation platform;

[0023] Imitation learning module, used to train the policy model using the imitation learning algorithm and deploy it to the embedded controller;

[0024] The muscle control module is used to send motion instructions to the PLC through the embedded controller based on the motion strategy output by the model. The PLC analyzes the motion instructions and generates PWM signals to control the inflation and deflation timing of the lower limb pneumatic artificial muscle group;

[0025] The closed-loop feedback module is used to provide real-time feedback of lower limb joint angles and air pressure data through sensors, adjust the PWM duty cycle in a closed loop, and optimize walking posture.

[0026] Beneficial effects of the present invention:

[0027] 1. Improved realism: This invention combines an imitation learning algorithm with pneumatic artificial muscle drive, overcoming the limitations of traditional control models in dealing with nonlinear dynamic problems. The imitation learning algorithm directly maps the human walking pattern, outperforming traditional PID control and achieving a more natural and stable bionic walking experience. It also improves the system's integration and scalability.

[0028] 2. Energy consumption optimization: The present invention adopts PWM to dynamically distribute air pressure, reducing redundant energy loss and improving the response speed and control accuracy of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1 It is a flow chart of the specific implementation of the method of the present invention;

[0031] Figure 2 It is the PWM control flow chart;

[0032] Figure 3 This is a schematic diagram of the arrangement of pneumatic artificial muscles. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are only used to explain the present invention and are not intended to limit the present invention.

[0034] like Figure 1 As shown, the present invention provides a pneumatic artificial muscle bipedal walking control method based on imitation learning, and the specific steps are as follows:

[0035] 1. Data collection and processing:

[0036] MIPI cameras are used to collect human walking posture data, which will be used as expert demonstrations to train imitation learning algorithms.

[0037] The collected data is processed and analyzed through the Mujoco simulation platform to simulate human walking posture and generate accurate expert trajectories, providing a basis for subsequent strategy model training.

[0038] 2. Strategy model training:

[0039] Imitation learning algorithms such as behavior cloning (BC) or generative adversarial imitation learning (GAIL) are used to learn and train the generated expert trajectories to obtain a policy model that can simulate human walking posture.

[0040] The trained policy model is deployed to an embedded controller (such as Jetson Orin Nano 4GB), which is responsible for running the policy model in real time and generating control commands.

[0041] 3. Control instruction transmission:

[0042] The embedded controller sends motion instructions to the PLC via a serial communication protocol (such as RS-485). These instructions contain all the information needed to control the pneumatic artificial muscle, such as the timing and strength of muscle contraction.

[0043] 4.PWM signal generation and control:

[0044] like Figure 2 As shown, after receiving the control instruction, the PLC parses it and generates the corresponding PWM signal according to the preset control logic.

[0045] The PWM signal is sent to the solenoid valve group that controls the pneumatic artificial muscle. By adjusting the duty cycle of the PWM signal, the inflation and deflation timing of the pneumatic artificial muscle is precisely controlled, thereby simulating the coordinated contraction of the human lower limb muscles and achieving natural and stable bipedal walking.

[0046] 5. Sensor feedback and closed-loop control:

[0047] The system integrates a variety of sensors, such as angle sensors and pressure sensors, to monitor the angle changes of the lower limb joints and the air pressure in the pneumatic artificial muscles in real time.

[0048] These sensor data are fed back to the PLC, which dynamically adjusts the PWM signal according to closed-loop control logic (such as PID algorithm) to achieve dynamic adjustment of muscle contraction timing, optimize walking posture, and ensure system stability and responsiveness.

[0049] 6. Hardware architecture and integration:

[0050] The entire system adopts a collaborative architecture of embedded controller and PLC, and realizes efficient data transmission and execution of control instructions through serial communication.

[0051] like Figure 3 As shown in FIG, 14 sets of pneumatic artificial muscles are installed on the bionic lower limbs, corresponding to different muscle groups, such as the quadriceps femoris in the front and the hamstrings in the back, to achieve multi-degree-of-freedom motion control.

[0052] On the other hand, the present invention also provides a pneumatic artificial muscle bipedal walking control system based on imitation learning, which includes a data acquisition module, an imitation learning module, a muscle control module and a closed-loop feedback module; for the specific implementation process of each module, please refer to the specific implementation steps of the above-mentioned pneumatic artificial muscle bipedal walking control method based on imitation learning.

[0053] The data acquisition module is used to collect human walking posture data and generate expert trajectories based on the Mujoco simulation platform;

[0054] The imitation learning module is used to train the policy model using the imitation learning algorithm and deploy it to the embedded controller;

[0055] The muscle control module is used to send action instructions to the PLC through the embedded controller based on the action strategy output by the model. The PLC analyzes the action instructions and generates PWM signals to control the inflation and deflation timing of the lower limb pneumatic artificial muscle group;

[0056] The closed-loop feedback module is used to provide real-time feedback of lower limb joint angles and air pressure data through sensors, adjust the PWM duty cycle in a closed loop, and optimize walking posture.

[0057] The specific deployment and implementation process of the present invention is as follows:

[0058] Hardware deployment:

[0059] 1. Install 14 sets of pneumatic artificial muscles on the bionic lower limb, 4 sets corresponding to the anterior muscle group (quadriceps femoris), 3 sets corresponding to the posterior muscle group (hamstrings), and the remaining sets for collaborative stability.

[0060] 2. Jetson Orin Nano establishes RS-485 communication with a PLC (example model: Siemens S7-1200) using the PySerial library.

[0061] Software process:

[0062] 1. Data preprocessing: Perform pose estimation (MediaPipe) on the RGB images collected by the MIPI camera and extract the joint angle sequence.

[0063] 2. Algorithm training: Load the human body model in Mujoco, generate a policy model (.onnx format) using the GAIL algorithm, and deploy it to Jetson.

[0064] 3. Real-time control: Jetson sends an action command frame (format: start bit + action code + check bit), and the PLC parses it and outputs a PWM signal to the solenoid valve group.

[0065] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modifications and changes made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A pneumatic artificial muscle bipedal walking control method based on imitation learning, characterized by: The method comprises the following steps: (1) Collect human walking posture data and generate expert trajectories based on the Mujoco simulation platform; (2) Use imitation learning algorithms to train the policy model and deploy it to the embedded controller; (3) The embedded controller sends action instructions to the PLC based on the action strategy output by the model; (4) The PLC analyzes the motion instructions and generates PWM signals to control the inflation and deflation timing of the lower limb pneumatic artificial muscle group; (5) Real-time feedback of lower limb joint angle and air pressure data is obtained through sensors, and the PWM duty cycle is adjusted in a closed loop to optimize walking posture.

2. The pneumatic artificial muscle bipedal walking control method based on imitation learning according to claim 1, characterized in that: The MIPI camera is used to collect RGB images of the human body while walking to estimate the posture, extract the joint angle sequence, and obtain the human walking posture data.

3. The pneumatic artificial muscle bipedal walking control method based on imitation learning according to claim 1, characterized in that: The Mujoco platform is used to simulate human walking posture, generate expert trajectory data, and construct a training dataset for the imitation learning algorithm.

4. The pneumatic artificial muscle bipedal walking control method based on imitation learning according to claim 1, characterized in that: The imitation learning algorithm is behavior cloning (BC) or generative adversarial imitation learning (GAIL).

5. The pneumatic artificial muscle bipedal walking control method based on imitation learning according to claim 1, characterized in that: The imitation learning algorithm integrates the reinforcement learning mechanism and dynamically optimizes the strategy model through environmental interaction.

6. The pneumatic artificial muscle bipedal walking control method based on imitation learning according to claim 1, characterized in that: The embedded controller is a Jetson Orin Nano 4GB, which is linked to the PLC through serial port communication.

7. The pneumatic artificial muscle bipedal walking control method based on imitation learning according to claim 1, characterized in that: The PWM signal adopts pulse width modulation technology and is allocated different duty cycles according to the coordinated contraction requirements of the pneumatic artificial muscle groups.

8. The pneumatic artificial muscle bipedal walking control method based on imitation learning according to claim 1, characterized in that: The closed-loop control uses a PID algorithm combined with multimodal data from an angle sensor and a pressure sensor.

9. The pneumatic artificial muscle bipedal walking control method based on imitation learning according to claim 1, characterized in that: The pneumatic artificial muscle group includes 14 groups of pneumatic artificial muscles, 4 groups corresponding to the anterior quadriceps muscle group, 3 groups corresponding to the posterior hamstring muscle group, and the rest are used for coordinated stability.

10. A pneumatic artificial muscle bipedal walking control system based on imitation learning that implements the method according to any one of claims 1 to 9, characterized in that: The system includes: The data acquisition module is used to collect human walking posture data and generate expert trajectories based on the Mujoco simulation platform; Imitation learning module, used to train the policy model using the imitation learning algorithm and deploy it to the embedded controller; The muscle control module is used to send motion instructions to the PLC through the embedded controller based on the motion strategy output by the model. The PLC analyzes the motion instructions and generates PWM signals to control the inflation and deflation timing of the lower limb pneumatic artificial muscle group; The closed-loop feedback module is used to provide real-time feedback of lower limb joint angles and air pressure data through sensors, adjust the PWM duty cycle in a closed loop, and optimize walking posture.

Images