Method for combining humanoid robot with robotic dog and switching two modes of humanoid robot and robotic dog

Abstract

The inventor-shaped robot and robot dog combination relates to AI control of the inventor-shaped robot and the robot dog. The defects of the humanoid robot in calculation, structure, weight, power consumption, price and application scenes are overcome. The innovation and implementation of the combined application of the two break through the inertial thinking constraint of people that people and dogs cannot be interchanged; in the design and manufacturing process, the technical problems of compatibility, sharing, adaptation and the like of the two are solved; by means of the combined advantages of the humanoid robot and the robot dog, the humanoid robot mode and the robot dog mode can be switched for use according to different application scenes, and the endurance time can be prolonged; the scene application such as oil platform inspection (dog four-foot movement and human-shaped upper limb valve operation), battlefield rescue (two-foot obstacle crossing, two-hand handling and four-foot load-bearing transportation) and the like is realized; according to the invention, the operation and the structure of the humanoid robot are simplified, the weight (the average weight is more than ten kilograms) is reduced, the power consumption (the average endurance is more than several hours) is reduced, the selling price (the average unit price is tens of thousands of RMB) is reduced, and the commercialization process is accelerated.

Description

Technical Field The present invention relates to AI control of humanoid robots and robot dogs. Background Art A search of Chinese patent documents reveals no patents for similar inventions. A review of the 2023 annual report of the International Federation of Robotics (IFR) reveals no commercial applications for composite robots. Some experts predict that 2025 will be the year of the humanoid robot revolution. However, current conditions present complex computational and structural requirements, high weight (averaging tens of kilograms), high power consumption (average battery life of only two hours), high price tags (average price of several hundred thousand RMB), and limited application scenarios (not yet capable of replacing humans in rescue operations, etc.). Therefore, commercializing humanoid robots will require years of effort. Summary of the Invention The present invention aims to overcome the shortcomings of humanoid robots in terms of computing, structure, weight, power consumption, price, and application scenarios, thereby accelerating their commercialization.

[0004] 1. The present invention is a method (method of use) for a specific purpose of a humanoid robot. The technical features are, firstly, the embodiment of a humanoid robot combined with a robot dog:

[0005] 1. The integration of a humanoid robot and a robot dog is feasible, as the two share many similarities or similarities, such as core robotics theories (kinematics, cybernetics) and hardware (sensors, processors). To optimize computation, reduce weight, energy consumption, and lower costs, a compatible neural network control processor, such as the NVIDIA Jetson Xavier / Orin series and the HOVER neural network controller, can be used. Similarly, the integration should, to the greatest extent possible, feature unified design and shared monitoring and actuation components. This can include high-density battery packs (lithium polymer batteries, solid-state batteries), a unified power supply standard (e.g., 48V / DC), and shared motor drivers (e.g., the ElmoGold series, Copley AcceINet) to accommodate motor control of varying power levels. Furthermore, to ensure seamless interchange between humanoid and robot dog modes, the humanoid robot's upper limbs could be designed to be the same length as its lower limbs (for quadrupedal balance). Modular grippers / footpads (e.g., the SchunkWSG 50 and Boston Dynamics Spot footpads) could also be used to achieve optimal integration.

[0006] 2. Because the robot dog's structure and movements are simpler than those of a humanoid robot, with fewer than 23 joints (degrees of freedom) and motors, redundant structural components can be removed and optimized while maintaining the basic functions of the humanoid robot. This results in a compatible, shared, and adaptable combination of the humanoid robot and the robot dog. Using the AI ​​large-scale model DeepSeek and this invention, the parameters of the Yushu humanoid robot G1 were first optimized (battery life of 1-2 hours, standing height of 1.27 meters, dimensions of 1270×450×200 mm, calf + thigh length of 0.6 meters, weight of 35 kg, 23 joint motors, and price of 100,000 yuan). The AI ​​large-scale model DeepSeek, with its advanced, accurate, efficient, and convenient advantages, has already been widely used. According to Shenzhen News Network, on February 16, Shenzhen officially launched DeepSeek model application services for all districts and departments across the city, based on its government cloud environment. This enabled the integration and upgrade of DeepSeek-based AI government applications. Beijing Daily also reported that the Institute of Materia Medica, Chinese Academy of Medical Sciences and Peking Union Medical College, recently completed the deployment of the DeepSeek-R1 model and plans to apply it to high-precision drug target prediction, large-scale molecular library screening, and routine laboratory data analysis. To address this, DeepSeek was asked: If a humanoid robot has a complex structure, weighs 35 kilograms, has a two-hour flight time, and costs 100,000 yuan, while maintaining its basic and primary functions, and removing redundant components, how much would its structure, weight, flight time, and price change? DeepSeek's in-depth thinking and reasoning led to the following answers and actual results: a simplified structure, a weight of 24 kg (reduced by approximately 30% due to the removal of redundant joints, support structures, and decorative shells), a flight time of 3 hours (increased by 50% due to the removal of redundant motors and sensors, which reduces overall power consumption), and a price of 70,000 yuan (reduced by 30% due to the removal of high-cost components such as motors, sensors, and complex structures). This suggests that maintaining a flight time of 2 hours while reducing battery weight (capacity) would make the humanoid robot even lighter than the simplified 24 kg and more affordable than the simplified 70,000 yuan price.

[0007] 3. Experimental results of the humanoid robot and robot dog combination model (humanoid robot mode and robot dog mode can be switched between each other):

[0008] A. Experimental model parameters: The combined body of the humanoid robot (standing height 30 cm) and the robot dog (standing height 16 cm on four legs) is 28 cm long, 10 cm wide, and weighs 0.74 kg.

[0009] B. Experimental data of humanoid robot mode: walking distance 211 cm, walking time 12 seconds, and speed 17.6 cm / s.

[0010] C. Experimental data for robot dog mode: same distance as above, walking time 8 seconds, walking speed 26.4 cm / s.

[0011] D. Based on the analysis of experimental data, it was concluded that humanoid robot mode consumes 50% more energy than robot dog mode. Furthermore, the humanoid robot mode alone can last for 26 minutes. If the humanoid robot mode and robot dog mode are switched at equal intervals, after 26 minutes, both modes will last for 13 minutes. Furthermore, because robot dog mode consumes 50% less energy, the humanoid robot mode can last an additional 6.5 minutes (for a total of 19.5 minutes). The combined battery life of the two modes is 32.5 minutes, 6.5 minutes longer than the 26-minute humanoid robot mode alone. Furthermore, the addition of robot dog mode and its associated functionality expands the application scenarios.

[0012] Second, the technical features of this invention are also reflected in the integration of a humanoid robot and a robot dog, which can be automatically or manually controlled to switch from humanoid robot mode to robot dog mode and then back to humanoid robot mode. The core technical principles and hardware and software configurations for achieving dual-mode switching are as follows:

[0013] 1. Structural compatibility design (embodied module)

[0014] Principle: Reduce switching complexity by sharing physical structures.

[0015] hardware:

[0016] Core structure sharing: Design the same core module for the aforementioned humanoid robot and the robot dog combination. This module includes a main controller (such as NVIDIA Jetson Xavier / Orin), a high-density battery pack (such as lithium polymer batteries or solid-state batteries, all using the 48V DC standard), a core sensor suite (such as IMU, main vision camera), and a communication module, to ensure compatibility and sharing between the two modes.

[0017] Modular joints: Using compatible servo motors (such as the Maxon EC-i40 + Harmonic Drive reducer) to make the upper limb (humanoid robot) and forelimb (robot dog) joints interchangeable;

[0018] End-effector adaptation: Design or select a modular end-effector interface (e.g., based on the Schunk WSG 50 gripper or the Boston Dynamics Spot footpad). In robot dog mode, the upper limbs can be quickly replaced with functional dog-foot or claw-style end-effectors (emphasizing physical adaptation interface design); in humanoid robot mode, they can be replaced with multi-fingered dexterous hands or tool grippers.

[0019] 2. Environmental Perception (Module) and Decision-Making (Module)

[0020] Principle: Select the optimal mode in real time based on multi-source data.

[0021] hardware:

[0022] LiDAR: Velodyne VLP-16 (mapping and obstacle detection);

[0023] Depth camera: Intel RealSense D455 (object / terrain recognition);

[0024] IMU: ADIS16470 (attitude feedback);

[0025] Microphone array: ReSpeaker 6-Mic (voice command reception).

[0026] Switching command generation and execution: When the decision module determines that a combined mode switch is necessary (for example, when the environmental perception module detects long, flat terrain without hand control, it transmits information to the decision module to switch from humanoid walking mode to robot dog walking mode to save energy), the controller (decision-making, execution) generates specific switching commands based on the current relevant information. These include joint motion trajectories for a safe and smooth transition from the current posture (humanoid standing) to the target posture (robot dog standing on four legs), or vice versa. This includes executing the aforementioned motion plan by sending precise position and torque commands to the relevant joint motors via a shared motor driver (such as the Elmo Gold series or Copley AccelNet driver). This also includes replacing the end effector (if necessary), triggering the robotic arm or auxiliary mechanism to automatically remove and install the end effector.

[0027] 3. Motion planning and control (execution module)

[0028] Principle: Generate safe posture transition trajectories and execute them precisely.

[0029] hardware:

[0030] Main controller: NVIDIA Jetson AGX Orin (running ROS 2);

[0031] Joint driver: Elmo Gold Twitter (1000Hz refresh rate).

[0032] Core algorithm: Posture transformation planning uses the inverse kinematics (IK) optimization algorithm.

[0033] Control algorithm switching: The controller loads and switches to the motion control algorithm library for the corresponding mode (humanoid robot or robot dog). The humanoid mode algorithm library is responsible for bipedal gait balance and walking; the robot dog mode algorithm library is responsible for quadrupedal gait control (such as trotting and pacing).

[0034] Dynamic balance control: PD controller based on the zero torque point (ZMP) stability criterion.

[0035] Gait generation: The humanoid robot mode uses ZMP preview control, and the robot dog mode uses the MIT Cheetah single-leg spring model.

[0036] 4. Dynamic Energy Management (Module)

[0037] Principle: Use power consumption models to predict and optimize energy distribution.

[0038] hardware:

[0039] Current sensor: INA226 (accuracy ±0.1%);

[0040] Battery management chip: TIBQ40Z80

[0041] Establish a basic power consumption model (which can be based on measured data or optimized using AI such as DeepSeek): This model includes the power consumption characteristics of the two modes under different task loads.

[0042] Real-time battery life prediction and mission planning: The controller integrates this power consumption model. Before or during a mission, it predicts the remaining battery life based on the current battery level, the mission route plan (length, estimated terrain), the mission content (estimated modes and actions required for each route section), and the basic power consumption model.

[0043] Dynamic Mode Scheduling: To maximize battery life or meet mission deadlines, the controller dynamically schedules mode switching. During long-distance maneuvers, the controller proactively schedules the robot dog mode (which is approximately 50% more energy-efficient than humanoid mode and prioritizes it) and the humanoid mode to operate alternately at a preset or calculated time ratio, extending battery life.

[0044] 3. The integration of the humanoid robot and the robot dog expands the application scenarios (with unique application scenarios in two modes):

[0045] In particular, when operating in robot dog mode (stable center of gravity and strong load-bearing capacity), it saves approximately 50% more energy than operating in humanoid robot mode (wide field of view and capable of performing human-like tasks). For example, in oil platform inspections (dog's quadrupedal movement + humanoid's upper limbs operating valves) and battlefield rescue (legged obstacle crossing + two-handed handling + quadrupedal load-carrying), the combined humanoid robot and robot dog can switch between the two modes at equal intervals. After two hours (one hour each in humanoid robot mode and robot dog mode), the energy saved can power the combined robot for an additional 40 minutes of operation in these scenarios. This means the combined robot can last up to 2 hours and 40 minutes, compared to the two hours of humanoid robot mode alone. Furthermore, either humanoid robot mode or robot dog mode alone cannot perform tasks such as battlefield rescue. Maintaining a two-hour combined battery life would also reduce battery weight (capacity), further reducing the weight and price of the combined robot and robot dog.

[0046] The idea and implementation of this combined application of humanoid robots and robot dogs first breaks through the constraints of people's inertial thinking - it is generally believed that humans and dogs are two different species, and even robots and robot dogs cannot be interchangeable; secondly, because humanoid robots and robot dogs currently have different divisions of labor, in the design and manufacturing process of the combination of the two, it is necessary to first solve some technical problems such as how the two can be compatible, shared, and adapted; and then, by taking advantage of the advantages of robot dogs such as relatively simple design and manufacturing, stable center of gravity, large load rate, low price (about 10,000 yuan), and energy saving, it is possible to open up application scenarios for the combination of the two: for example, oil platform inspection (dog's four-legged movement + humanoid upper limbs operating valves), battlefield rescue (two feet crossing obstacles + two hands handling + four-legged load-bearing transportation), etc.; and it can simplify operations and structure, reduce weight (average weight of more than ten kilograms), reduce power consumption (average endurance of more than a few hours), and lower the price (average unit price of tens of thousands of yuan), thereby accelerating the commercialization process of humanoid robots. DETAILED DESCRIPTION 1. Use compatible neural network control computing devices, such as NVIDIA's Jetson Xavier / Orin series and HOVER neural network controller, in the humanoid robot and robot dog combination; share monitoring and execution components, such as high-density battery packs (lithium polymer batteries, solid-state batteries), unified power supply (such as 48V.DC); motor drivers (such as ElmoGold series, Copley AcceINet), etc.; adjust the fixed upper and lower limbs to the same length (to ensure stability during quadrupedal walking); remove non-compatible parts of the humanoid robot and robot dog, such as eliminating independent toe movement and finger joints; or install conversion module grippers / foot ends (such as SchunkWSG 50, Boston Dynamics Spot foot pads) to achieve the compatibility of the two combinations.

[0048] 2. Remove redundant joints, support structures, decorative shells, redundant motors and sensors of the humanoid robot to optimize its structure.

[0049] 3. Structural Compatibility Design (Embodied Module)

[0050] hardware:

[0051] Core structure sharing: Design the same core module for the aforementioned humanoid robot and the robot dog combination. This module includes a main controller (such as NVIDIA Jetson Xavier / Orin), a high-density battery pack (such as lithium polymer batteries or solid-state batteries, all using the 48V DC standard), a core sensor suite (such as IMU, main vision camera), and a communication module, to ensure compatibility and sharing between the two modes.

[0052] Modular joints: Using compatible servo motors (such as the Maxon EC-i40 + Harmonic Drive reducer) to make the upper limb (humanoid robot) and forelimb (robot dog) joints interchangeable;

[0053] End-effector adaptation: Design or select a modular end-effector interface (e.g., based on the Schunk WSG 50 gripper or the Boston Dynamics Spot footpad). In robot dog mode, the upper limbs can be quickly replaced with functional dog-foot or claw-style end-effectors (emphasizing physical adaptation interface design); in humanoid robot mode, they can be replaced with multi-fingered dexterous hands or tool grippers.

[0054] 4. Environmental Perception (Module) and Decision-Making (Module)

[0055] hardware:

[0056] LiDAR: Velodyne VLP-16 (mapping and obstacle detection);

[0057] Depth camera: Intel RealSense D455 (object / terrain recognition);

[0058] IMU: ADIS16470 (attitude feedback);

[0059] Microphone array: ReSpeaker 6-Mic (voice command reception).

[0060] ·Switching instruction generation and execution: When the decision module determines that it is necessary to switch the action mode (for example, when the environmental perception module detects a long distance of flat terrain and no hand operation is required, it uploads to the decision module to decide to switch from the humanoid robot walking mode to the robot dog walking mode to save energy), the controller (decision, execution) generates specific switching instructions based on the current relevant information, including the joint motion trajectory of safely and smoothly transitioning from the current posture (humanoid robot standing) to the target posture (robot dog standing on four legs), or vice versa; including sending precise position and torque instructions to the relevant joint motors through a shared motor driver (such as Elmo Gold series or CopleyAccelNet driver) to execute the above motion planning; also including the replacement of the end effector (if necessary), triggering the robotic arm or auxiliary mechanism to perform the automatic quick release and installation operations of the end effector, etc. 5. Motion planning and control (execution module)

[0061] hardware:

[0062] Main controller: NVIDIA Jetson AGX Orin (running ROS 2);

[0063] Joint driver: Elmo Gold Twitter (1000Hz refresh rate).

[0064] Core algorithm: Posture transformation planning uses the inverse kinematics (IK) optimization algorithm.

[0065] Control algorithm switching: The controller loads and switches to the motion control algorithm library for the corresponding mode (humanoid robot or robot dog). The humanoid mode algorithm library is responsible for bipedal gait balance and walking; the robot dog mode algorithm library is responsible for quadrupedal gait control (such as trotting and pacing).

[0066] Dynamic balance control: PD controller based on the zero torque point (ZMP) stability criterion.

[0067] Gait generation: The humanoid robot mode uses ZMP preview control, and the robot dog mode uses the MIT Cheetah single-leg spring model.

[0068] 6. Dynamic Energy Management (Module)

[0069] hardware:

[0070] Current sensor: INA226 (accuracy ±0.1%);

[0071] Battery management chip: TIBQ40Z80

[0072] Establish a basic power consumption model (which can be based on measured data or optimized using AI such as DeepSeek): This model includes the power consumption characteristics of the two modes under different task loads.

[0073] Real-time battery life prediction and mission planning: The controller integrates this power consumption model. Before or during a mission, it predicts the remaining battery life based on the current battery level, the mission route plan (length, estimated terrain), the mission content (estimated modes and actions required for each route section), and the basic power consumption model.

[0074] Dynamic Mode Scheduling: To maximize battery life or meet mission deadlines, the controller dynamically schedules mode switching. During long-distance maneuvers, the controller proactively schedules the robot dog mode (which is approximately 50% more energy-efficient than humanoid mode and prioritizes it) and the humanoid mode to operate alternately at a preset or calculated time ratio, extending battery life.

[0075] 7. Examples of switching between the two modes of a humanoid robot and a robot dog:

[0076] Application scenario: battlefield rescue.

[0077] Initial mode: Humanoid robot mode.

[0078] Mission: Carry the wounded and supplies back through flat grasslands and rubble.

[0079] Step 1: The perception module detects that the terrain ahead is flat grass and uploads the information. The decision module is energy-saving and automatic or manually controlled, triggering the execution module to switch modes. After the switch is successful, the combination walks in robot dog mode (the upper limbs are quickly replaced with functional dog-foot or claw-type end effectors).

[0080] Step 2: The perception module detects that the terrain has changed to rugged gravel. The decision module determines that the crossing efficiency of the robot dog mode may be lower than that of the humanoid robot mode, or the preset rules require the use of humanoid robot bipedalism to pass complex obstacles. The execution module is triggered to switch back to the humanoid robot mode (the upper limbs are quickly replaced with multi-fingered dexterous hands or tool grippers). After the switch is successful, the robot crosses the obstacle with two feet in humanoid robot mode.

[0081] Step 3: After arriving at the location of the injured, maintain the humanoid robot mode and perform tasks such as wounded inspection and treatment (visual observation, two-handed operation), and transportation (upper limb operation).

[0082] Step 4: When returning with the wounded and supplies, the decision module determines based on the information from the perception module that the combined load is heavy and the terrain is flat, and the robot dog mode may be more stable and energy-efficient. It then triggers the execution module to switch to the robot dog mode (the upper limbs are quickly replaced with functional dog-foot or claw-type end effectors), and finally the combined body returns with the load in the robot dog mode.

[0083] Energy Management: The range prediction module runs continuously throughout the mission. If the current battery consumption is too high to return to normal, the dispatcher can increase the proportion of robot dog mode on the remaining flat sections to save power or notify the operator.

Claims

1. The commercialization of humanoid robots is restricted by structure, weight, energy consumption, price, and application scenarios, while robot dogs have different characteristics. Combining a humanoid robot with a robot dog and switching them to humanoid robot mode or robot dog mode in different scenarios can accelerate the marketization of humanoid robots.