A dynamic posture smooth transition control method for a quadruped robot

Abstract

The application provides a kind of quadruped robot dynamic posture smooth transition control method, comprising the following steps: S1: transient snapshot and state freezing, including, response and posture intention instruction, execute freezing operation;S2: build non-impact time domain model, including, build core interpolation function;S3: multidimensional space mapping and inverse calculation, including, in each time period of action execution, according to time course deduce motion state;The application can produce the following technical effects: 1 non-zero position dynamic start: abandon forced reset mechanism, the physical state of robot current time (no matter in what distortion posture) is regarded as the legal motion starting point.2 Finite Jerk: adopt quintic polynomial model, ensure the continuity of position, velocity, acceleration in time domain from mathematical principle, eliminate rigid impact.3 full-dimensional coupling: interpolation factor is applied to Cartesian space coordinates and body attitude angle simultaneously, realize "body following leg movement" bionic coordination.

Description

Technical Field

[0001] This invention relates to the field of robotics, and more specifically to a method for smooth transition control of dynamic posture of a quadruped robot. Background Technology

[0002] In practical applications of quadruped robots (such as wilderness rescue and home care), the robots are often in unexpected postures (such as being pushed off balance by external force, being in a half-squatting state, or being in an interrupted state where the previous action was not completed).

[0003] The limitations of existing technologies include:

[0004] 1. The "Zeroing" Dependency Trap: Existing control algorithms typically follow a "static reset" logic, which assumes that all actions must start from a preset standard standing zero position. When the robot is tilted due to terrain or external forces, forcibly executing the reset logic will cause the robot to undergo a sudden and violent "jump". This non-physical displacement command can trigger a huge mechanical impact, and may even cause it to tip over due to a violent shift in the center of gravity.

[0005] 2. Rigid control damage: Traditional linear interpolation or trapezoidal velocity programming results in discontinuous acceleration at the start and end points of the action (i.e., the jerk acceleration is infinite). This mathematical discontinuity manifests physically as rigid impact, which can lead to pitting and tooth breakage in precision reduction gears over long-term operation, significantly reducing hardware lifespan.

[0006] 3. The contradiction between computing power and performance: Although high-order spline interpolation (such as B-Spline) is smooth, it involves complex matrix inverse operations, making it difficult to achieve kHz-level real-time computation on low-cost, low-power microcontrollers. Summary of the Invention

[0007] To address the above problems, this invention provides a method for smooth dynamic posture transition control of a quadruped robot, comprising the following steps:

[0008] S1: Transient snapshot and state freeze, including responding to attitude intention commands and performing freeze operations;

[0009] S2: Constructing an impact-free time-domain model, including constructing the core interpolation function. ;

[0010] S3: Multidimensional space mapping and inverse solution, including deducing the motion state based on the time process in each time period of action execution;

[0011] S4: Performs altitude compensation for special postures.

[0012] Furthermore, the S11 state capture operation includes reading and locking the current global attitude state vector, which includes the pitch angle, roll angle, altitude, and center of gravity offset of the aircraft.

[0013] The S12 reverse reconstruction operation includes, based on the captured body state, reversibly deriving the current three-dimensional coordinates of the four leg tips relative to the body center.

[0014] Furthermore, the interpolation function in step S2 Specifically:

[0015]

[0016] In the formula, Indicates normalized time. For instantaneous time, This represents the total action time.

[0017] Furthermore, step S3 includes,

[0018] S31 trajectory evolution: Calculate the interpolation ratio based on the current time progress. ;

[0019] S32 uses the following steps to calculate pose co-interpolation:

[0020] S321 calculates the limb trajectory, with the foot coordinates smoothly transitioning from the starting position to the target position, following... The evolutionary pattern, Indicates the instantaneous target pose. Indicates the starting target pose. Indicates the target pose;

[0021] S322 adjusts the trunk posture and uses the interpolation ratio to perform weighted evolution of the body posture angle, so that the trunk posture and limb movement achieve full-dimensional coupling of the interpolation factor.

[0022] The S33 real-time inverse kinematics solver inputs the instantaneous target pose into the inverse kinematics solver and uses the jerk continuity derivation formula to calculate the instantaneous target angle of each joint, which then drives the motor to execute.

[0023] Furthermore, step S4 includes dynamically adjusting the target height parameters of the relevant limbs in response to the asymmetric changes in the body structure caused by the target action, and smoothly transitioning through an S-curve.

[0024] This invention proposes a lightweight control strategy of "state freezing-quintic law programming-inverse solution mapping" to simulate the compliant characteristics of organisms during movement, which can produce the following technical effects:

[0025] 1. Non-zero position dynamic start: Abandoning the forced reset mechanism, the robot's current physical state (regardless of its twisted posture) is regarded as a legal starting point for motion.

[0026] 2. Finite Jerk: Employs a fifth-order polynomial model to mathematically ensure the continuity of position, velocity, and acceleration in the time domain, thus eliminating rigid impacts.

[0027] 3. Full-dimensional coupling: The interpolation factor is applied to both Cartesian space coordinates and body posture angles to achieve biomimetic coordination of "body follows leg movement". Attached Figure Description

[0028] Figure 1 For the control system flowchart, Figure 1 The complete closed loop is demonstrated from "receiving instructions" -> "determining current global variables (PIT_S / R_H)" -> "generating S-curve sequence" -> "IK calculation / angle interpolation" -> "PWM output".

[0029] Figure 2 This is a comparison graph of an S-curve and a linear curve. The coordinate system in the graph is: X-axis represents time, Y-axis represents velocity. Figure 2 The linear interpolation is shown to be a rectangular wave (abrupt change), while the S-curve of this invention is a bell wave (smooth transition).

[0030] Figure 3 This is a schematic diagram illustrating the principle of dynamic starting points. Figure 3 The robot demonstrates how it transitions directly from "posture A (tilted)" to "posture B (sitting)" without going through "posture 0 (standing)". Detailed Implementation

[0031] The method of this invention operates in the robot's motion control center, and its execution process simulates the planning and execution of actions by a biological nervous system, including the following steps:

[0032] Step S1: Transient Snapshot and State Freeze When a new postural intention (such as "sit down") is issued by a higher-level decision, the control center immediately executes a "freeze" operation:

[0033] State capture: The system instantly reads and locks the current global attitude state vector, including the pitch angle, roll angle, altitude and center of gravity offset of the aircraft.

[0034] Reverse reconstruction: Using a forward kinematics model, based on the captured body state, the current three-dimensional coordinates of the four legs relative to the body center are derived in reverse. This step establishes the physical starting point for trajectory planning, ensuring that new movements smoothly originate from the current real position.

[0035] Step S2: Constructing an impact-free time-domain model system by introducing normalized time variables. The core interpolation function is constructed using a fifth-order polynomial. Used to describe the evolution of motion processes:

[0036] Indicates normalized time. For instantaneous time, This represents the total action time.

[0037] set up Indicates the change in pose. This is the initial pose. The target pose.

[0038] Mathematical properties: This function at the initial time ( ) and the final moment ( The first derivative (velocity) and second derivative (acceleration) of ) are naturally zero.

[0039] Physical meaning: This means that the torque change rate is zero at the moment of motor start-up and stop, completely eliminating the "jerkiness" at start-up and the "overshoot oscillation" at stop.

[0040] Step S3: Multidimensional Space Mapping and Inverse Solving Within each tiny time slice of action execution, the system deduces the motion state based on the time progress:

[0041] Trajectory evolution: Calculate the interpolation ratio based on the current time progress. .

[0042] Pose-coordinated interpolation:

[0043] Limb trajectory: The foot coordinates smoothly transition from the starting position to the target position, following... The evolutionary pattern, It indicates the real-time pose.

[0044] Trunk posture: The body posture angles are simultaneously weighted and evolved to achieve a natural coupling between trunk posture and limb movements.

[0045] Real-time inverse kinematics solution: The evolved instantaneous target pose is input into the inverse kinematics solver to calculate the instantaneous target angle of each joint and drive the motor to execute.

[0046] Step S4: Adaptive Height Compensation (for Special Postures) For asymmetrical postures such as "sitting" or "lying down," the algorithm introduces a geometric compensation mechanism based on body structure:

[0047] When a target action is detected to involve asymmetrical changes in the body structure (such as changes in the height of the hind legs when sitting), the geometric compensation module is invoked to dynamically calculate the target height compensation value based on the body structure parameters, and the compensation amount is smoothly transitioned through the fifth-order polynomial curve.

[0048] The jerk continuity derivation of this invention uses the following formula.

[0049] Let the total action time be The instantaneous time is t, and the normalized time is... Position function It can be represented as:

[0050]

[0051] Regarding time Find the velocity by taking the first derivative. :

[0052]

[0053] It can be seen that when and hour, The speed must be continuous.

[0054] Regarding time Find the second derivative to obtain the acceleration. :

[0055]

[0056] It can be seen that when and hour, It satisfies the requirement of continuous acceleration, i.e., no rigid impact. It represents acceleration.

[0057] Regarding time Find the third derivative to obtain the accelerometer. :

[0058]

[0059] It is a continuous parabolic function, without any abrupt changes to infinity. This means that the torque change experienced by the servo motor is continuous and controllable.

[0060] The above formula does not require trigonometric functions or matrix inversion. On an ESP32 chip with a main frequency of 240MHz, the time for a single calculation is only in the microsecond range, which fully meets the real-time requirements of the 200Hz control loop.

[0061] Storage requirements: There is no need to store a complex trajectory point table; only the current starting state vector and target state vector need to be saved, resulting in extremely low RAM usage.

[0062] The present invention has the following technical advantages over the prior art:

[0063] This invention includes a transient state capture mechanism that uses the current real-time attitude parameters as the starting point for trajectory planning, rather than the preset zero-position parameters.

[0064] This invention uses a fifth-order polynomial model to generate normalized control signals, ensuring that the acceleration of the drive motor is continuous and bounded in the time domain.

[0065] This invention establishes a full-dimensional coupled mapping relationship between the Cartesian coordinates of the foot and the body attitude angles during the trajectory evolution process.

[0066] This invention significantly reduces the impact load on the servo gear set by eliminating abrupt acceleration changes, and theoretically can extend the service life of the actuator by more than 30%.

[0067] This invention allows robots to respond directly to new commands after being disturbed by external forces (such as being kicked out of place) without needing to straighten themselves, thus improving the robot's survivability.

[0068] The motion trajectory of this invention conforms to the laws of biological kinematics, eliminating the "mechanical stiffness" commonly found in robots.

Claims

1. A method for smooth dynamic posture transition control of a quadruped robot, comprising the following steps: S1: Transient snapshot and state freeze This includes responding to and executing gesture intent commands, and performing a freeze operation; S2: Constructing an impact-free time-domain model, including constructing the core interpolation function. ; S3: Multidimensional space mapping and inverse solution, including deducing the motion state based on the time process in each time period of action execution; S4: Performs altitude compensation for special postures.

2. The method for smooth transition of dynamic posture of a quadruped robot as described in claim 1, wherein the freezing operation in step S1 includes: The S11 state capture operation includes reading and locking the current global attitude state vector, which includes the pitch angle, roll angle, altitude, and center of gravity offset of the aircraft. The S12 reverse reconstruction operation includes, based on the captured body state, reversibly deriving the current three-dimensional coordinates of the four leg tips relative to the body center.

3. The method for smooth transition control of dynamic posture of a quadruped robot as described in claim 1, interpolation function in step S2 Specifically: In the formula, Indicates normalized time. For instantaneous time, This represents the total action time.

4. The method for smooth transition control of dynamic posture of a quadruped robot as described in claim 1, Step S3 includes... S31 trajectory evolution: Calculate the interpolation ratio based on the current time progress. ; S32 uses the following steps to calculate pose co-interpolation: S321 calculates the limb trajectory, with the foot coordinates smoothly transitioning from the starting position to the target position, following... The evolutionary pattern, Indicates the instantaneous target pose. Indicates the starting target pose. Indicates the target pose; S322 adjusts the trunk posture and uses the interpolation ratio to perform weighted evolution of the body posture angle, so that the trunk posture and limb movement achieve full-dimensional coupling of the interpolation factor. The S33 real-time inverse kinematics solver inputs the instantaneous target pose into the inverse kinematics solver and uses the jerk continuity derivation formula to calculate the instantaneous target angle of each joint, which then drives the motor to execute.

5. The method for smooth transition of dynamic posture of a quadruped robot as described in claim 1, Step S4 includes dynamically adjusting the target height parameters of the relevant limbs in response to the asymmetric changes in the body structure caused by the target action, and smoothly transitioning through an S-curve.

Images