Biped robot real machine deployment method
By optimizing control frequency, pseudo-parallel data acquisition, and IMU data synchronization, combined with reinforcement learning and extended Kalman filtering, the problems of control frequency mismatch and sensor noise differences in the actual deployment of bipedal robots were solved, enabling stable operation and efficient deployment of the robot in complex environments.
Patent Information
- Application Number
- CN202510957766.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-11
- Publication Date
- 2025-10-21
AI Technical Summary
In existing technologies, bipedal robots face problems such as control frequency mismatch, sensor noise differences, and mechanical structure errors during the process of moving from simulation environment to actual deployment, resulting in their inability to walk stably in reality and a lack of effective deployment solutions.
By optimizing control frequency, pseudo-parallel data acquisition, IMU data synchronization, and real-time calibration, combined with reinforcement learning and extended Kalman filtering, the consistency between the simulation environment and the actual environment is optimized. By adopting online learning and robust control strategies, the robot can achieve stable operation in complex environments.
It improves the performance of bipedal robots in real-world environments, achieving stable walking and turning capabilities, enhancing the robot's robustness and terrain adaptability, and shortening the conversion cycle from simulation to actual operation.
Smart Images
Figure CN120816539A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of biped robot deployment, in particular to a method for deploying a biped robot. Background Art
[0002] In today's rapidly developing world, humanoid robots, the product of the deep integration of artificial intelligence, high-end manufacturing, and new materials technologies, have become a strategic high ground in global technological competition and a core driving force for industrial transformation. Humanoid robots, integrating advanced technologies such as artificial intelligence, high-end manufacturing, and new materials, are expected to be disruptive, profoundly transforming human production and lifestyles and reshaping the global industrial landscape. On the international stage, star products such as the Tesla Optimus and Boston Dynamics Atlas continue to achieve technological breakthroughs, demonstrating the boundless potential of humanoid robots. In China, numerous startups and unicorns, such as Yushu Technology, Zhiyuan Robotics, and Fourier Intelligence, have realized numerous scenarios and examples of highly dynamic and intelligent humanoid robot operations.
[0003] The development of humanoid robots is receiving more and more attention. Their advantages in terrain adaptability, energy efficiency, wide range of workspace and walking efficiency give them broad application prospects in service, rescue and industrial fields.
[0004] Current research in the humanoid robotics field primarily encompasses upper limb manipulation (manipulation) and lower limb robust locomotion (locomotion). This project focuses on addressing the error between the simulation and real-world (SimToReal) gap in lower limb robust locomotion. Bipedal robots from different manufacturers vary in configuration, and factors such as mechanical structure, motor accuracy, and backlash increase the difficulty of transitioning from training to deployment. To reduce the gap, this project plans to reduce deployment difficulty through humanoid body design, refined model post-processing, pre-randomization, and increased backtracking in Sim2Sim and Real2Sim.
[0005] Furthermore, despite the enormous commercial and research value of bipedal robots, there are relatively few outstanding open source projects, making it difficult for the industry to form a collaborative effort. Therefore, the project plans to open source all detailed information, source code, and design files to establish a model and ecosystem for bipedal robots, making them easier to get started with.
[0006] In terms of technological progress, the research and development of bipedal robots has been ongoing for half a century, with its evolution divided into four phases. In recent years, the study of bipedal locomotion has become a hot topic in disciplines such as mechanics, mechanics, control, robotics, and biology. Bipedal robots are designed and manufactured to resemble and mimic the human body in form, function, and movement. The next generation of humanoid work robots will combine sensory awareness with mobile manipulation and dynamic locomotion to perform productive tasks previously reserved for biological humans.
[0007] In the development of robotics, bipedal robots, with their highly anthropomorphic locomotion and potential to adapt to complex environments, have become a hot research area. However, to transform bipedal robots from theoretical concepts into practical applications, real-world deployment is a critical and challenging step. Sim2Real (Simulation to Reality), a key research method, aims to transfer control strategies trained and verified in simulation environments to actual robotic hardware.
[0008] Sim2Real faces numerous challenges. Significant differences exist between the simulation environment and the real world, encompassing aspects such as physical properties, sensor noise, and subtle deviations in mechanical structure. Regarding physical properties, the modeling of plantar friction, joint force control, and other aspects in simulation is often simplified and deviates from actual conditions. This can cause a robot to perform well in simulation but struggle to walk stably in the real world. Sensor noise is also a significant issue. Actual sensor readings are subject to various interferences, while sensor data in simulation is typically idealized. This makes strategies trained on simulation data susceptible to failure in the face of real-world noise. Furthermore, factors such as manufacturing errors and assembly accuracy in the mechanical structure can affect the robot's actual motion performance, making these difficult to accurately simulate in simulation.
[0009] To overcome these difficulties, an effective deployment plan is crucial. First, the simulation environment needs to be optimized to approximate the physical properties and sensor noise of the real world as closely as possible. The realism of the simulation environment can be enhanced by introducing more accurate physical models and adding random interference. Secondly, in terms of control strategy, adaptive strategies and robust control methods are adopted to enable the robot to automatically adjust its motion mode in the face of environmental changes and uncertainties. Furthermore, the data collection and feedback mechanism during the actual machine deployment process is also very critical. By collecting data during actual operation and feeding it back to the simulation environment, the control strategy and simulation model can be continuously optimized. Summary of the Invention
[0010] The specific solution for the real-machine deployment (Sim2Real) of the bipedal robot of the present invention, including control frequency optimization, data processing and synchronization, real-machine testing and performance evaluation, aims to provide a feasible technical path for the practical application of bipedal robots.
[0011] The present invention provides the following technical solutions: A method for deploying a bipedal robot, comprising the following steps: Step 1: In Sim2Real deployment, the control frequency requirements of the simulation environment and the actual environment must reach the same control frequency; Step 2: When obtaining the position q and velocity dq information of the robot joint, a pseudo-parallel method is adopted through batch request sending and asynchronous response receiving; Step 3: In the Sim2Real process, control the synchronization between the IMU data acquisition frequency and the robot control frequency; Step 4: Improve the deployment of the actual bipedal robot by evaluating its performance in real-world scenarios and testing its ability to cope with various complex situations.
[0012] Preferably, the step 2 is specifically as follows: Pseudo-parallelism utilizes the connectionless feature of the UDP protocol to send requests to the IP addresses of all target joints at once, thus achieving "pseudo-parallel" packet sending; In the response reception phase, an asynchronous approach is adopted. All requests are received in a unified manner after they are sent. By utilizing the parallelism of the network layer and setting a timeout, unresponsive nodes are skipped to prevent the failure or delay of a single node from affecting the overall data acquisition efficiency. Use the thread pool to manage the creation and destruction of threads to avoid the overhead caused by frequent creation and destruction of threads; optimize the request format, reduce the size of request data, and improve network transmission efficiency; dynamically adjust the timeout period according to the actual network conditions.
[0013] Preferably, in Sim2Real deployment, the control frequency and control strategy are matched and optimized. For MPC-based strategies, when a control frequency of 400Hz cannot be achieved in the actual environment, the MPC algorithm is simplified or optimized, including reducing the prediction time domain and reducing the optimization accuracy, so that better control effects can still be achieved at a lower control frequency. For RL-based strategies, adjust training parameters based on the actual control frequency, including increasing the number of training rounds and adjusting the reward function, so that the RL strategy can better adapt to the control frequency in the actual environment; By adopting the online learning method, the strategy is continuously updated during the actual operation of the robot to improve its adaptability to environmental changes.
[0014] Preferably, the step 3 is specifically: Step 3.1: Based on the typical parameters of robot reinforcement learning (RL) control, whose control frequency is usually in the range of 50-100 Hz, a multi-level filtering system is constructed to address interference such as Gaussian noise and zero bias drift contained in the IMU raw signal. The extended Kalman filter (EKF) is used in conjunction with the robot dynamics model to perform fusion estimation on the measurement data of the accelerometer and gyroscope. Step 3.2: Define the robot's coordinate system as a right-handed system: the x-axis points straight ahead, the y-axis points to the left, and the z-axis points vertically upward. The physical installation position of the IMU must strictly coincide with the origin of this coordinate system. When importing the model into the simulation environment, verify the coordinate system configuration: ensure that the direction of gravity acceleration and the inertial measurement reference are exactly the same as those of the physical robot. Use a 3D calibration tool to measure the installation deviation angle between the IMU coordinate system and the body coordinate system, and use the rotation matrix to perform coordinate transformation compensation. Step 3.3: Perform systematic parameter calibration on the IMU, including bias error correction, sensitivity calibration, and dynamic characteristics matching.
[0015] Preferably, a nine-axis IMU module is used. The IMU module has a built-in extended Kalman filter algorithm, which automatically fuses the accelerometer, gyroscope, and magnetometer data. The original sensor data and the fused attitude information can be output via USB and is compatible with binary protocols. The IMU module directly outputs quaternions and Euler angles, allowing the carrier's attitude to be acquired without additional calculation. The quaternions are converted into Euler angles using the Northeast Celestial Coordinate System. A filtering algorithm suppresses noise, and the angle delay is ≤5ms in dynamic scenarios, ensuring real-time control accuracy.
[0016] Preferably, accelerometer data is used for long-term attitude trend calibration, gyroscope data ensures high-frequency dynamic response, and sensor noise covariance is estimated in real time through the state space model, and the attitude angle error after smoothing is outputted to be ≤0.5°.
[0017] Preferably, the step 4 is specifically as follows: After completing the Sim2Real deployment, Alexbotmini successfully implemented a variety of basic motion functions in real-machine testing, including: Stable walking movements, able to move forward at a certain speed and stride according to the set gait plan, and maintain body balance during walking without obvious shaking or falling; In terms of steering operation, precise steering is achieved according to the instructions, and the steering angle and speed are accurately controlled, showing good maneuverability; Robustness test,Multiple interference tests were conducted on Alexbotmini.
[0018] A bipedal robot real-machine deployment system, the system comprising: A frequency control module, wherein in Sim2Real deployment, the control frequency requirements of the frequency control module in the simulation environment and the actual environment are the same; A pseudo-parallel module, wherein the pseudo-parallel module acquires the position q and velocity dq information of the robot joints by sending batch requests and receiving asynchronous responses in a pseudo-parallel manner; A control module, which controls the synchronization between the IMU data acquisition frequency and the robot control frequency in the Sim2Real process; The deployment module improves the deployment of the actual bipedal robot by evaluating the robot's operating performance in actual scenarios and testing its ability to cope with various complex situations.
[0019] A computer-readable storage medium stores a computer program, which is executed by a processor to implement a method for deploying a bipedal robot.
[0020] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements a method for deploying a bipedal robot when executing the computer program.
[0021] The present invention has the following beneficial effects: During real-world testing and deployment, the Real2Sim technology process effectively verified the consistency between the hardware and the simulation environment, providing support for Sim2Real deployment. Through control frequency optimization, IMU optimization, and simulation alignment, the robot's performance in real-world environments was improved. In real-world testing, Alexbotmini demonstrated excellent basic motion capabilities and robustness. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0023] Figure 1 Shown is a schematic diagram of a bipedal robot of the present invention walking; Figure 2 Shown is a schematic diagram of the robustness test of the biped robot of the present invention; Figure 3 Shown is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0024] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0025] The present invention is described in detail below with reference to specific embodiments. Specific embodiment one: according to Figures 1 to 3 As shown, the specific optimization technical solution adopted by the present invention to solve the above technical problems is: the present invention relates to a real-machine deployment method of a bipedal robot.
[0027] The present invention provides a method for deploying a bipedal robot, the method comprising the following steps: Step 1: In Sim2Real deployment, the control frequency requirements of the simulation environment and the actual environment must reach the same control frequency; Step 2: When obtaining the position q and velocity dq information of the robot joint, a pseudo-parallel method is adopted through batch request sending and asynchronous response receiving; Step 3: In the Sim2Real process, control the synchronization between the IMU data acquisition frequency and the robot control frequency; Step 4: Improve the deployment of the actual bipedal robot by evaluating its performance in real-world scenarios and testing its ability to cope with various complex situations.
[0028] During real-world testing and deployment, the Real2Sim technology process effectively verified the consistency between the hardware and the simulation environment, providing support for Sim2Real deployment. Through control frequency optimization, IMU optimization, and simulation alignment, the robot's performance in real-world environments was improved. In real-world testing, Alexbotmini demonstrated excellent basic motion capabilities and robustness. Specific embodiment two: The difference between the second embodiment of the present application and the first embodiment is that: The step 2 is specifically as follows: Pseudo-parallelism utilizes the connectionless feature of the UDP protocol to send requests to the IP addresses of all target joints at once, thus achieving "pseudo-parallel" packet sending; In the response reception phase, an asynchronous approach is adopted. All requests are received in a unified manner after they are sent. By utilizing the parallelism of the network layer and setting a timeout, unresponsive nodes are skipped to prevent the failure or delay of a single node from affecting the overall data acquisition efficiency. Use the thread pool to manage the creation and destruction of threads to avoid the overhead caused by frequent creation and destruction of threads; optimize the request format, reduce the size of request data, and improve network transmission efficiency; dynamically adjust the timeout period according to the actual network conditions. Specific embodiment three: The only difference between the third embodiment of the present application and the second embodiment is that: In Sim2Real deployment, the control frequency and control strategy are matched and optimized. For MPC-based strategies, when a 400Hz control frequency cannot be achieved in the actual environment, the MPC algorithm is simplified or optimized, including reducing the prediction time domain and reducing the optimization accuracy, so that good control effects can still be achieved at a lower control frequency. For RL-based strategies, adjust training parameters based on the actual control frequency, including increasing the number of training rounds and adjusting the reward function, so that the RL strategy can better adapt to the control frequency in the actual environment; By adopting the online learning method, the strategy is continuously updated during the actual operation of the robot to improve its adaptability to environmental changes. Specific embodiment four: The only difference between the fourth embodiment of the present application and the third embodiment is that: The step 3 is specifically as follows: Step 3.1: Based on the typical parameters of robot reinforcement learning (RL) control, whose control frequency is usually in the range of 50-100 Hz, a multi-level filtering system is constructed to address interference such as Gaussian noise and zero bias drift contained in the IMU raw signal. The extended Kalman filter (EKF) is used in conjunction with the robot dynamics model to perform fusion estimation on the measurement data of the accelerometer and gyroscope. Step 3.2: Define the robot's coordinate system as a right-handed system: the x-axis points straight ahead, the y-axis points to the left, and the z-axis points vertically upward. The physical installation position of the IMU must strictly coincide with the origin of this coordinate system. When importing the model into the simulation environment, verify the coordinate system configuration: ensure that the direction of gravity acceleration and the inertial measurement reference are exactly the same as those of the physical robot. Use a 3D calibration tool to measure the installation deviation angle between the IMU coordinate system and the body coordinate system, and use the rotation matrix to perform coordinate transformation compensation. Step 3.3: Perform systematic parameter calibration on the IMU, including bias error correction, sensitivity calibration, and dynamic characteristics matching. Specific embodiment five: The only difference between the fifth embodiment of the present invention and the fourth embodiment is that: It uses a nine-axis IMU module with a built-in extended Kalman filter algorithm that automatically fuses accelerometer, gyroscope, and magnetometer data. It can output raw sensor data and fused attitude information via USB and is compatible with binary protocols. The IMU module directly outputs quaternions and Euler angles, allowing the carrier's attitude to be acquired without additional calculation. The quaternions are converted into Euler angles using the Northeast Celestial Coordinate System. A filtering algorithm suppresses noise, and the angle delay is ≤5ms in dynamic scenarios, ensuring real-time control accuracy. Specific embodiment six: The only difference between the sixth embodiment of the present invention and the fifth embodiment is that: Accelerometer data is used for long-term attitude trend calibration, gyroscope data ensures high-frequency dynamic response, and sensor noise covariance is estimated in real time through the state-space model. The attitude angle error after smoothing the output is ≤0.5°. Specific embodiment seven: The only difference between the seventh embodiment of the present invention and the sixth embodiment is that: The step 4 is specifically as follows: After completing the Sim2Real deployment, Alexbotmini successfully implemented a variety of basic motion functions in real-machine testing, including: Stable walking movements, able to move forward at a certain speed and stride according to the set gait plan, and maintain body balance during walking without obvious shaking or falling; In terms of steering operation, precise steering is achieved according to the instructions, and the steering angle and speed are accurately controlled, showing good maneuverability; Robustness test,Multiple interference tests were conducted on Alexbotmini. Specific embodiment eight: The only difference between the eighth embodiment of the present invention and the seventh embodiment is that: The present invention provides a bipedal robot real machine deployment system, the system comprising: A frequency control module, wherein in Sim2Real deployment, the control frequency requirements of the frequency control module in the simulation environment and the actual environment are the same; A pseudo-parallel module, wherein the pseudo-parallel module acquires the position q and velocity dq information of the robot joints by sending batch requests and receiving asynchronous responses in a pseudo-parallel manner; A control module, which controls the synchronization between the IMU data acquisition frequency and the robot control frequency in the Sim2Real process; The deployment module improves the deployment of the actual bipedal robot by evaluating the robot's operating performance in actual scenarios and testing its ability to cope with various complex situations.
[0036] During real-world testing and deployment, the Real2Sim technology process effectively verified the consistency between the hardware and the simulation environment, providing support for Sim2Real deployment. Through control frequency optimization, IMU optimization, and simulation alignment, the robot's performance in real-world environments was improved. In real-world testing, Alexbotmini demonstrated excellent basic motion capabilities and robustness. Specific embodiment nine: The only difference between the ninth embodiment of the present invention and the eighth embodiment is that: The present invention provides a computer-readable storage medium on which a computer program is stored. The program is executed by a processor to implement a method for deploying a bipedal robot.
[0038] The method comprises the following steps: Step 1: In Sim2Real deployment, the control frequency requirements of the simulation environment and the actual environment must reach the same control frequency; Step 2: When obtaining the position q and velocity dq information of the robot joint, a pseudo-parallel method is adopted through batch request sending and asynchronous response receiving; Step 3: In the Sim2Real process, control the synchronization between the IMU data acquisition frequency and the robot control frequency; Step 4: Improve the deployment of the actual bipedal robot by evaluating its performance in real-world scenarios and testing its ability to cope with various complex situations.
[0039] During real-world testing and deployment, the Real2Sim technology process effectively verified the consistency between the hardware and the simulation environment, providing support for Sim2Real deployment. Through control frequency optimization, IMU optimization, and simulation alignment, the robot's performance in real-world environments was improved. In real-world testing, Alexbotmini demonstrated excellent basic motion capabilities and robustness. Specific embodiment ten: The only difference between the tenth embodiment of the present invention and the ninth embodiment is that: The present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements a method for deploying a bipedal robot when executing the computer program.
[0041] The method comprises the following steps: Step 1: In Sim2Real deployment, the control frequency requirements of the simulation environment and the actual environment must reach the same control frequency; Step 2: When obtaining the position q and velocity dq information of the robot joint, a pseudo-parallel method is adopted through batch request sending and asynchronous response receiving; Step 3: In the Sim2Real process, control the synchronization between the IMU data acquisition frequency and the robot control frequency; Step 4: Improve the deployment of the actual bipedal robot by evaluating its performance in real-world scenarios and testing its ability to cope with various complex situations.
[0042] During real-world testing and deployment, the Real2Sim technology process effectively verified the consistency between the hardware and the simulation environment, providing support for Sim2Real deployment. Through control frequency optimization, IMU optimization, and simulation alignment, the robot's performance in real-world environments was improved. In real-world testing, Alexbotmini demonstrated excellent basic motion capabilities and robustness. Specific embodiment eleven: The only difference between the eleventh embodiment of the present invention and the tenth embodiment is that: A method for deploying a bipedal robot, comprising the following steps: Step 1: In Sim2Real deployments, the control frequency requirements for simulation and real-world environments differ significantly. In simulation, due to the idealized computing resources and simulation conditions, higher control frequencies are relatively easy to achieve. For example, certain high-precision dynamics simulation software can easily achieve control frequencies exceeding kilohertz, enabling detailed simulation of robot motion. However, achieving similarly high control frequencies on actual bipedal robot hardware platforms is extremely difficult due to limitations in hardware performance, communication bandwidth, and power consumption. Step 2: To increase control frequency in real-world environments, pseudo-parallel technology is an effective solution. When acquiring the position (q) and velocity (dq) information of robot joints, the traditional serial approach involves sending a request to each joint in sequence and waiting for a response. This approach results in high latency and limits the increase in control frequency. Pseudo-parallel technology, however, reduces latency and improves data acquisition efficiency by sending requests in batches and receiving responses asynchronously.
[0044] Specifically, pseudo-parallel technology leverages the connectionless nature of the UDP protocol to send requests to the IP addresses of all target joints simultaneously, achieving "pseudo-parallel" packet transmission. During the response reception phase, an asynchronous approach is employed, with responses received simultaneously after all requests have been sent. This leverages the parallelism of the network layer, avoiding the time accumulating by traditional methods of waiting for responses one by one. Furthermore, by properly setting timeouts, unresponsive joints can be quickly skipped, preventing individual joint failures or delays from impacting overall data acquisition efficiency.
[0045] In actual applications, the above code can be further optimized. For example, a thread pool can be used to manage thread creation and destruction to avoid the overhead of frequent thread creation and destruction; the request format can be optimized to reduce the size of request data and improve network transmission efficiency; and the timeout period can be dynamically adjusted according to actual network conditions to balance the accuracy and efficiency of data acquisition.
[0046] In Sim2Real deployment, the control frequency and control strategy must be optimized. For MPC-based strategies, if a 400Hz control frequency is unattainable in the actual environment, consider simplifying or optimizing the MPC algorithm. For example, by reducing the prediction time domain and lowering the optimization accuracy, this can help achieve good control results at a lower control frequency. Alternatively, other control strategies, such as PID control, can be combined to fine-tune the MPC output and improve the robot's motion stability.
[0047] For RL-based strategies, training parameters can be adjusted based on the actual control frequency. For example, increasing the number of training rounds and adjusting the reward function can help the RL strategy better adapt to the control frequency in the actual environment. Furthermore, online learning can be used to continuously update the strategy during the robot's actual operation to improve its adaptability to environmental changes.
[0048] Through the above control frequency optimization measures, the gap between the simulation environment and the actual environment can be effectively narrowed, and the performance and stability of the bipedal robot in Sim2Real deployment can be improved.
[0049] Step 3: In the Sim2Real technical process, synchronization between the IMU data acquisition frequency and the robot control frequency is fundamental to real-time attitude perception. Typical parameters for robot reinforcement learning (RL) control typically indicate a control frequency range of 50-100 Hz, and the IMU sampling frequency must strictly match this range. A mismatch between the two frequencies can result in frame drops or redundancy during data transmission, directly impacting the accuracy of the robot's real-time attitude calculations for pitch, roll, and yaw. Through hardware driver configuration and software synchronization algorithms, the IMU outputs data at a frequency that is an integer multiple of the control period, providing a stable inertial measurement data stream to the motion controller.
[20] To combat interference such as Gaussian noise and bias drift in the raw IMU signal, a multi-level filtering system must be constructed. An extended Kalman filter (EKF) or complementary filtering algorithm, combined with the robot's dynamics model, fuses and estimates the accelerometer and gyroscope data. Accelerometer data is used for long-term attitude trend calibration, while gyroscope data ensures high-frequency dynamic response. A state-space model is used to estimate sensor noise covariance in real time, and smoothed attitude angles (error ≤ 0.5°) are output. This process effectively suppresses environmental noise such as mechanical vibration and electromagnetic interference, providing reliable attitude feedback signals for the control algorithm.
[0050] Coordinate system consistency is a core principle of Sim2Real conversion. First, define the robot's body coordinate system (the baselink coordinate system in the URDF) as a right-handed system: the x-axis points straight ahead, the y-axis points to the left, and the z-axis points vertically upward. The physical installation position of the IMU must strictly coincide with the origin of this coordinate system. When importing the model into a simulation environment (such as Gazebo), the coordinate system configuration must be verified: ensure that the direction of gravity acceleration (negative z-axis) and the inertial measurement reference are exactly the same as those of the physical robot. Use a 3D calibration tool to measure the installation deviation angle between the IMU coordinate system and the body coordinate system. Use a rotation matrix to perform coordinate transformation compensation to eliminate attitude solution deviations caused by mechanical installation errors. This invention uses the nine-axis IMU module from HyperCore Technology (HIPNUC). The HyperCore IMU uses a USB interface design and supports plug-and-play. It can be connected to a computer or embedded device without additional hardware conversion. It is compatible with multiple systems and uses host software to monitor data and configure parameters in real time, significantly reducing the debugging threshold. On the hardware level, it has built-in high-precision MEMS sensors and is factory-calibrated over a full temperature range. The gyroscope has a zero bias stability as low as 1.76° / h, the accelerometer has a zero bias of 21μg, and supports 100-1000Hz data output (default 100Hz). IP68 waterproof and vibration-resistant design, combining miniaturization (<75g) with industrial-grade reliability, makes it suitable for rapid integration in scenarios such as robots and drones.
[0051] The built-in extended Kalman filter algorithm automatically fuses data from the accelerometer, gyroscope, and magnetometer, achieving static pitch / roll accuracy of ±0.15°. It also supports one-click leveling and geomagnetic "figure-8 calibration" to eliminate installation deviations and magnetic field interference. It outputs raw sensor data (acceleration / angular velocity) and fused attitude information (Euler angles and quaternions) via USB. It is compatible with binary protocols (HI91 / HI92 frames) and is compatible with development frameworks such as ROS and C. Users can flexibly adjust the coordinate system (e.g., rotate 90° around an axis) through commands and accommodate vertical and inverted mounting configurations to meet the attitude measurement requirements of different carriers.
[0052] The SuperCore IMU directly outputs quaternions (WXYZ) and Euler angles (heading / pitch / roll), enabling the determination of the vehicle's attitude without requiring additional calculations. Quaternions are converted to Euler angles using the Northeast Celestial coordinate system (ZXY rotation order). A filtering algorithm suppresses noise, ensuring angular latency of ≤5ms in dynamic scenarios, ensuring real-time control accuracy. When using the IMU, attention should be paid to geomagnetic calibration in 9-axis mode (to avoid interference from indoor magnetic fields). The host computer should also configure the sensor range (e.g., ±2000° / s for the gyroscope) to balance measurement range and resolution, achieving efficient Sim2Real coordinate alignment and attitude calculation.
[0053] In order to eliminate the difference in measurement characteristics between simulation and reality, the IMU needs to be systematically calibrated: Bias Error Correction: Static data is collected in a constant temperature environment. The accelerometer bias (the average output value in a non-motion state under a gravitational field) and gyroscope bias (drift noise at rest) are calculated. A temperature compensation model is established to improve bias stability to ±50 μg (accelerometer) and ±0.03° / s (gyroscope). Sensitivity calibration: Use a high-precision turntable to perform multi-axis motion testing, fit the scale factor error of the accelerometer and gyroscope (target accuracy ≤0.1% FS), and correct the nonlinear deviation of the sensor output. Dynamic characteristic matching: The amplitude-frequency response characteristics of the IMU are obtained through frequency sweep testing, and equivalent noise density and bandwidth parameters are added to the simulation model to ensure that the dynamic response of the virtual sensor is consistent with that of the real object. The integrated application of these technologies significantly improves Sim2Real's migration efficiency: gait control strategies trained in a simulation environment can be directly deployed to a physical robot after coordinate system alignment and parameter calibration, reducing on-site commissioning time by over 30%. Field data shows that the optimized IMU system achieves posture resolution latency of ≤5ms and heading angle drift rate of ≤0.1° / s when navigating complex terrain. This provides a high-precision inertial reference for the robot's dynamic balance control and adaptive gait planning, effectively shortening the technology transition cycle from simulation training to real-world application. Step 4: During the development of the bipedal robot Alexbot_mini, real-world deployment and robustness testing were key components in measuring its performance. Evaluating the robot's performance in real-world scenarios and testing its ability to handle various complex situations provided crucial insights for further optimizing the robot's design and control algorithms.
[0054] Demonstrating Basic Motion Capabilities: After completing the Sim2Real deployment, Alexbotmini successfully implemented various basic motion functions in real-world testing. These included stable walking, following a pre-programmed gait plan, moving forward at a specific speed and stride length, and maintaining balance without noticeable wobbling or falling.
[0055] In terms of steering operations, precise steering can be achieved according to instructions, and the steering angle and speed control are relatively accurate, showing good maneuverability.
[0056] Robustness Testing: Alexbotmini was subjected to various interference tests. When subjected to a certain external impact force, such as a gentle nudge on the robot's body, it responded quickly, adjusting its joint torque and posture to maintain its balance and avoid falls. It also maintained stable movement despite external impacts and uneven surfaces with small potholes or bumps, demonstrating strong terrain adaptability.
[0057] During real-world testing and deployment, the Real2Sim technology process effectively verified the consistency between the hardware and the simulation environment, providing support for Sim2Real deployment. Through control frequency optimization, IMU optimization, and simulation alignment, the robot's performance in real-world environments was improved. In real-world testing, Alexbotmini demonstrated excellent basic motion capabilities and robustness. Specific implementation method 12: The method comprises the following steps: Step 1: In Sim2Real deployment, the control frequency requirements of the simulation environment and the actual environment must reach the same control frequency; Step 2: When obtaining the position q and velocity dq information of the robot joint, a pseudo-parallel method is adopted through batch request sending and asynchronous response receiving; Step 3: In the Sim2Real process, control the synchronization between the IMU data acquisition frequency and the robot control frequency; Step 4: Improve the deployment of the actual bipedal robot by evaluating its performance in real-world scenarios and testing its ability to cope with various complex situations.
[0059] During real-world testing and deployment, the Real2Sim technology process effectively verified the consistency between the hardware and the simulation environment, providing support for Sim2Real deployment. Through control frequency optimization, IMU optimization, and simulation alignment, the robot's performance in real-world environments was improved. In real-world testing, Alexbotmini demonstrated excellent basic motion capabilities and robustness.
[0060] The above description is merely a preferred embodiment of a method for deploying a bipedal robot. The scope of protection for a method for deploying a bipedal robot is not limited to the aforementioned embodiment; all technical solutions based on this concept fall within the scope of protection of the present invention. It should be noted that improvements and variations that do not depart from the principles of the present invention, as readily apparent to those skilled in the art, should also be considered within the scope of protection of the present invention.
Claims
1. A method for deploying a bipedal robot, characterized by: The method comprises the following steps: Step 1: In Sim2Real deployment, the control frequency requirements of the simulation environment and the actual environment must be the same; Step 2: When obtaining the position q and velocity dq information of the robot joint, a pseudo-parallel method is adopted through batch request sending and asynchronous response receiving; Step 3: In the Sim2Real process, control the synchronization between the IMU data acquisition frequency and the robot control frequency; Step 4: Improve the deployment of the actual bipedal robot by evaluating its performance in real-world scenarios and testing its ability to cope with various complex situations.
2. The method according to claim 1, wherein: The step 2 is specifically as follows: Pseudo-parallelism utilizes the connectionless feature of the UDP protocol to send requests to the IP addresses of all target nodes at once, thus achieving "pseudo-parallel" packet sending. In the response reception phase, an asynchronous approach is adopted. All requests are received in a unified manner after they are sent. By utilizing the parallelism of the network layer and setting a timeout, unresponsive nodes are skipped to prevent the failure or delay of a single node from affecting the overall data acquisition efficiency. Use the thread pool to manage the creation and destruction of threads to avoid the overhead caused by frequent creation and destruction of threads; optimize the request format, reduce the size of request data, and improve network transmission efficiency; dynamically adjust the timeout period according to the actual network conditions.
3. The method according to claim 2, wherein: In Sim2Real deployment, the control frequency and control strategy are matched and optimized. For MPC-based strategies, when a control frequency of 400 Hz cannot be achieved in the actual environment, the MPC algorithm is simplified or optimized, including reducing the prediction time domain and lowering the optimization accuracy, so that good control effects can still be achieved at a lower control frequency. For RL-based strategies, adjust training parameters based on the actual control frequency, including increasing the number of training rounds and adjusting the reward function, so that the RL strategy can better adapt to the control frequency in the actual environment; By adopting the online learning method, the strategy is continuously updated during the actual operation of the robot to improve its adaptability to environmental changes.
4. The method according to claim 3, wherein: The step 3 is specifically as follows: Step 3.1: Based on the typical parameters of robot reinforcement learning (RL) control, whose control frequency is usually in the range of 50-100 Hz, a multi-level filtering system is constructed to address interference such as Gaussian noise and zero bias drift contained in the IMU raw signal. The extended Kalman filter (EKF) is used in combination with the robot dynamics model to perform fusion estimation on the measurement data of the accelerometer and gyroscope. Step 3.2: Define the robot's coordinate system as a right-handed system: the x-axis points straight ahead, the y-axis points to the left, and the z-axis points vertically upward. The physical installation position of the IMU must strictly coincide with the origin of this coordinate system. When importing the model into the simulation environment, the coordinate system configuration must be verified: ensure that the direction of gravity acceleration and the inertial measurement reference are completely consistent with those of the physical robot. Use a 3D calibration tool to measure the installation deviation angle between the IMU coordinate system and the body coordinate system, and use the rotation matrix to perform coordinate transformation compensation; Step 3.3: Perform systematic parameter calibration on the IMU, including zero bias error correction, sensitivity calibration, and dynamic characteristics matching.
5. The method according to claim 4, wherein: It uses a nine-axis IMU module with a built-in extended Kalman filter algorithm, which automatically fuses accelerometer, gyroscope, and magnetometer data. It can output raw sensor data and fused attitude information via USB and is compatible with binary protocols. The IMU module directly outputs quaternions and Euler angles, allowing the carrier's attitude to be acquired without additional calculation. The quaternions are converted into Euler angles using the Northeast Celestial Coordinate System. A filtering algorithm suppresses noise, and the angle delay is ≤5ms in dynamic scenarios, ensuring real-time control accuracy.
6. The method according to claim 4, wherein: Accelerometer data is used for long-term attitude trend calibration, gyroscope data ensures high-frequency dynamic response, and sensor noise covariance is estimated in real time through the state-space model. The attitude angle error after smoothing the output is ≤0.5°.
7. The method according to claim 6, wherein: The step 4 is specifically as follows: After completing the Sim2Real deployment, Alexbotmini successfully achieved a variety of basic motion functions in real-machine testing, including stable walking movements, following a set gait plan, moving forward at a certain speed and stride length, and maintaining body balance during walking without noticeable shaking or falling; In terms of steering operation, precise steering is achieved according to instructions, and the steering angle and speed are accurately controlled, showing good maneuverability; for robustness testing, various interference tests were conducted on Alexbotmini.
8. A bipedal robot real-machine deployment system, characterized by: The system comprises: A frequency control module, wherein in Sim2Real deployment, the control frequency requirements of the simulation environment and the actual environment are the same; A pseudo-parallel module, wherein the pseudo-parallel module acquires the position q and velocity dq information of the robot joints by sending batch requests and receiving asynchronous responses in a pseudo-parallel manner; A control module, which controls the synchronization of the IMU data acquisition frequency and the robot control frequency in the Sim2Real process; The deployment module improves the deployment of the actual bipedal robot by evaluating the robot's operating performance in actual scenarios and testing its ability to cope with various complex situations.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method according to claims 1 to 5.
10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method of claims 1 to 5 is implemented.