Event-triggered method and system for detecting foot end phase of a leg-foot robot

Abstract

The application discloses a kind of event trigger-based leg-foot type robot foot end phase detection method and system, the method includes: obtaining the joint rotation angle and joint torque of leg-foot type robot at current time, and foot end speed and acceleration;Based on joint rotation angle, calculate foot end kinematics information, and input it into kinematics probability model, solve the foot end phase state under the model in combination with virtual body angle;Based on joint torque, calculate foot end three-dimensional observation force, and input it into dynamics probability model, solve the foot end phase state under the model in combination with foot end speed and acceleration;Based on foot end three-dimensional observation force, calculate and obtain model correction value;Combined with model correction value, construct phase detection pre-model based on reference time and friction, combined with the foot end phase state under two probability models, calculate actual foot end phase transition probability, judge the current occurrence of take-off or ground contact event, complete foot end phase detection, improve detection accuracy.

Description

Technical Field

[0001] This invention belongs to the field of legged robot control technology, and particularly relates to a method and system for detecting the foot phase of a legged robot based on event triggering. Background Technology

[0002] In nature, legged animals can easily move through complex terrain thanks to their efficient and stable abilities. Biomimetic legged robots, by mimicking these animals, achieve remarkable locomotion and possess a unique advantage in traversing complex, unstructured environments. However, this advantage is accompanied by significant challenges, primarily stemming from limitations in hardware technology and control methods. While existing motion controllers for legged robots are relatively mature and robust, many rely on predefined gait planning or heuristically predetermined foot contact points. These methods typically require support from robust sensory information or known terrain conditions. Although they perform well on relatively flat terrain when sensory information fails, they perform poorly in extremely unknown environments with varying altitudes. To address this issue, existing approaches to foot phase detection in legged robots primarily employ two methods.

[0003] One common method for leg state detection and control relies on force sensors fixed to the robot's feet. These sensors provide ground contact information, allowing for real-time monitoring of foot contact and lift-off states to improve the robustness of legged robots in unfamiliar terrain. However, this method faces two main problems. First, adding external devices increases inertial characteristics, contradicting the principle of lightweight design for legged mechanisms. Second, the legs of legged robots interact with the ground frequently, potentially damaging the fixed force sensors and causing detection failure. As an alternative to fixed force sensors, visual sensing devices (such as cameras or LiDAR) can be used as external sensors to provide the robot with a real-time mapping of its environment, which can be used to generate foot avoidance support points. However, in many complex situations, such as in dense smoke or noisy environments, the external visual sensors may fail to accurately reflect the current information due to sensor noise or lighting conditions, leading to the method's failure.

[0004] Another common approach is deep learning-based detection, but this method requires a large number of real-world ground samples with accurate labels and sufficient learning of terrain characteristics, which limits its application in legged robots to some extent. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention provides an event-triggered foot phase detection method for legged robots. By establishing kinematic and mechanical probability models of the foot, and with only an encoder and torque motor, the method uses the proprioceptors of the legged robot to sense feedback information and obtain the robot's leg state in real time, achieving more reliable leg state detection and thus completing accurate foot phase detection. This method exhibits high fault tolerance and robustness.

[0006] In a first aspect, the present invention provides a method for detecting the phase at the foot of a legged robot based on event triggering.

[0007] A method for phase detection at the foot of a legged robot based on event triggering, comprising:

[0008] The system obtains the observed values ​​of joint rotation angles and joint torques of the legged robot at the current moment, and calculates the robot's foot velocity and acceleration based on the joint rotation angles through kinematic calculations.

[0009] Based on joint rotation angles, foot kinematics information is calculated through forward kinematics. This foot kinematics information is then input into a kinematic probability model. Combined with virtual body angles, the foot phase state under the kinematic probability model is solved.

[0010] Based on joint torque, the three-dimensional observation force at the foot end is calculated using Jacobi. The three-dimensional observation force at the foot end is then input into the dynamic probability model. Combined with the foot end velocity and acceleration, the foot end phase state under the dynamic probability model is solved.

[0011] Based on the three-dimensional observation force at the foot, the model correction value is obtained by calculating the duration of the friction cone.

[0012] By combining model correction values, a phase detection pre-model based on reference time and friction is constructed. Combined with the foot phase state under the kinematic probability model and the dynamic probability model, the actual foot phase transition probability is calculated to determine whether the current event is a lift-off event or a touch-off event, thus completing the robot's foot phase detection.

[0013] Further technical solutions also include:

[0014] The trajectory of the foot swing phase is described by a Bézier curve generated by two curve segments and 14 control points, and the trajectory of the foot support phase is described by a cosine curve.

[0015] Based on the current foot phase state of the legged robot, the corresponding trajectory curve is selected so that the robot's foot moves according to the planned trajectory curve.

[0016] Secondly, the present invention provides an event-triggered foot phase detection system for a legged robot.

[0017] An event-triggered foot-end phase detection system for a legged robot includes:

[0018] The data acquisition module is used to acquire the observed values ​​of the joint angles and joint torques of the legged robot at the current moment, and to obtain the robot's foot velocity and acceleration based on the joint angles through kinematic calculations.

[0019] The initial foot phase detection module is used to calculate foot kinematic information based on joint rotation angles using forward kinematics, input the foot kinematic information into the kinematic probability model, and solve the foot phase state under the kinematic probability model by combining it with virtual body angles; based on joint torques, it calculates the three-dimensional observation force of the foot using Jacobi, inputs the three-dimensional observation force of the foot into the dynamic probability model, and solves the foot phase state under the dynamic probability model by combining it with foot velocity and acceleration;

[0020] The correction coefficient calculation module is used to obtain model correction values ​​based on the three-dimensional observation force at the foot end by calculating the duration of the friction cone.

[0021] The foot phase detection module is used to construct a phase detection pre-model based on reference time and friction by combining model correction values, and to calculate the actual foot phase transition probability by combining the foot phase state under the kinematic probability model and the dynamic probability model, to determine whether the current event is a lift-off event or a touch-off event, and to complete the robot's foot phase detection.

[0022] Further technical solutions also include:

[0023] The foot motion drive module is used to describe the foot swing phase trajectory using a Bezier curve generated by two curves and 14 control points, and to describe the foot support phase trajectory using a cosine curve. Based on the foot phase state of the legged robot at the current time point, the module selects the corresponding trajectory curve so that the robot's foot moves according to the planned trajectory curve.

[0024] The above one or more technical solutions have the following beneficial effects:

[0025] 1. This invention provides a method and system for foot phase detection of a legged robot based on event triggering. By establishing a kinematic probability model and a mechanical probability model of the foot, the system acquires the leg state of the legged robot in real time using only the encoder and the body perception feedback information of the legged robot, thereby achieving more reliable leg state detection and having higher fault tolerance. Based on the leg phase detection method, the body perception state detection method enables the robot to effectively change the leg state without the need for external sensors (such as lidar equipment), which extends the robot's applicability to low visibility environments, such as smoke and bright light environments, providing a more solid foundation for achieving stable movement in unknown environments.

[0026] 2. The detection method proposed in this invention has high robustness and high scalability. Integrating it with other leg state detection technologies further enhances the application potential of this method. This integration will significantly expand the application range of legged robots, enabling them to operate in various complex environments. Ultimately, it will enhance the stability, flexibility, and overall performance of quadruped robots, making them better able to navigate and complete complex tasks in challenging environments. Attached Figure Description

[0027] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0028] Figure 1 This is a flowchart of the event-triggered foot phase detection method for a legged robot in an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of the forward phase detection of the legged robot used in this embodiment of the invention;

[0030] Figure 3 This is a schematic diagram of reverse phase detection of a legged robot used in an embodiment of the present invention;

[0031] Figure 4 This is a schematic diagram of the virtual machine body angle in an embodiment of the present invention;

[0032] Figure 5 This is a schematic diagram of the friction constraint model in an embodiment of the present invention;

[0033] Figure 6 This is the optimized foot phase map used for phase detection in the legged robot embodiment of the present invention. Detailed Implementation

[0034] It should be noted that the following detailed descriptions are exemplary and are intended only to describe specific embodiments and to provide further explanation of the invention, and are not intended to limit the scope of exemplary embodiments of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0035] Example 1

[0036] Inspired by the strategies employed by legged animals to efficiently acquire foot position and contact force information through proprioceptors, this embodiment provides an event-triggered legged robot foot phase detection method. This method establishes kinematic and mechanical probability models of the foot and utilizes only encoders and proprioceptive feedback information from the legged robot to acquire the leg state in real time, thereby achieving more reliable leg state detection. With the aid of the adopted leg phase detection method, this proprioceptive state detection method enables the robot to effectively change its leg state without the need for external sensors (such as LiDAR devices), extending the robot's applicability to low-visibility environments, such as smoky and bright light environments, providing a more solid foundation for stable movement in unknown environments. Furthermore, this detection method, while possessing high robustness, also exhibits high scalability. Integrating it with other leg state detection technologies further enhances its application potential. This integration significantly expands the application range of legged robots, enabling them to operate in various complex environments, ultimately enhancing the stability, flexibility, and overall performance of quadruped robots, allowing them to navigate and complete complex tasks in challenging environments.

[0037] The event-triggered foot phase detection method for legged robots proposed in this embodiment utilizes a lightweight encoder located at the joint of the legged robot and forward kinematics to obtain the position of the foot in the hip joint coordinate system. By equipping the joint motor with torque feedback and dynamic information, the observed reaction force of the foot is obtained. Under the premise of obtaining accurate foot position and observed reaction force information, kinematic probability detection models and dynamic probability detection models are established respectively. These detection models use virtual body angles and bandpass filters to adjust the weights of the foot position and observed reaction force in the corresponding probability models in real time, so as to reflect the role that the current observation parameters should play in the corresponding models. Since the feedback information may fluctuate in the calculation of the probability model even after filtering due to terrain changes or poor control effect, a pre-model based on time and friction is introduced to ensure the accuracy of foot phase detection. In practical applications, in addition to the foot detection method mentioned above, a smooth foot swing trajectory composed of two Bezier curves is also proposed. This design ensures that while maintaining smooth motion and energy-saving characteristics, the leg and foot mechanism has the ability to wait for the detection phase change in the phase when it is more likely to touch the ground. Considering that there are fewer cases of sudden changes in the state of the foot during the support phase, the support phase trajectory adopts a simpler sine wave curve planning, and the foot position is accurately determined based on the constructed trajectory.

[0038] During the movement of a legged robot, the robot's foot undergoes a reciprocating periodic motion of swinging in the air, touching the ground for support, and then swinging in the air again. One swinging phase and one ground-touching phase constitute a complete legged robot gait cycle. In the gait cycle, the swinging phase is defined as the foot swing phase, and the ground-touching phase is defined as the foot support phase. These two phase states are represented by Boolean variables χ = {0 = swinging, 1 = support}. The event of switching from the swinging phase to the support phase is defined as the ground-touching event, and the event of switching from the support phase to the swinging phase is defined as the lift-off event.

[0039] The forward transition process of the foot-end phase detection method for legged robots is as follows: Figure 2 As shown, the novel reverse phase detection method proposed in this embodiment is as follows: Figure 3As shown, the specific scheme adopted includes: First, obtaining the feedback information required for robot foot phase detection in step S1; then, solving for the phase states under two models in probabilistic model-based calculation steps S2 and S3 respectively; and solving for the model correction value in step S4. Finally, combining the phase states obtained in steps S2 and S3 in step S5, and using the model correction value obtained in step S4 to calculate the probability of ground lift-off and ground contact events, thereby achieving robot phase detection. Furthermore, considering that ground contact events often occur in the last quarter of a gait cycle, a foot trajectory planning curve for waiting for ground contact events is proposed in step S6. This ensures better detection performance in steps S1-S5 while reducing lateral force interference in the plane.

[0040] The event-triggered foot phase detection method for legged robots proposed in this embodiment is as follows: Figure 1 As shown, the specific steps include:

[0041] Step S1: Obtain the observed values ​​of joint rotation angle and joint torque of the legged robot at the current moment, and obtain the robot's foot velocity and acceleration based on the joint rotation angle through kinematic calculation;

[0042] Step S2: Based on joint rotation angle, calculate foot kinematics information through forward kinematics, input the foot kinematics information into the kinematic probability model, and combine it with virtual body angle to solve the foot phase state under the kinematic probability model;

[0043] Step S3: Based on joint torque, calculate the three-dimensional observation force at the foot end using Jacobi, input the three-dimensional observation force at the foot end into the dynamic probability model, and combine the foot end velocity and acceleration to solve the foot end phase state under the dynamic probability model.

[0044] Step S4: Based on the three-dimensional observation force at the foot, obtain the model correction value by calculating the duration of the friction cone.

[0045] Step S5: Combine the model correction values ​​to construct a phase detection pre-model based on reference time and friction. Combine the foot phase state under the kinematic probability model and the dynamic probability model to calculate the actual foot phase transition probability, determine whether the current event is a lift-off event or a touch-off event, and complete the robot foot phase detection.

[0046] Furthermore, in step S6, the Bezier curve generated by two curve segments and 14 control points is used to describe the foot swing phase trajectory, and the cosine curve is used to describe the foot support phase trajectory; according to the foot phase state of the legged robot at the current time point, the corresponding trajectory curve is selected so that the robot's foot moves according to the planned trajectory curve.

[0047] The following content introduces the event-triggered foot phase detection method for legged robots proposed in this embodiment.

[0048] In step S1, the legged robot obtains joint feedback information from its leg mechanism via proprioceptors. In this embodiment, relevant information is obtained only through the robot's built-in component sensors. The robot's joint rotation angle θ can be obtained through encoders installed at the joints, and the joint angular velocity is also measured provided the encoder accuracy is high enough. Joint angles can be obtained by taking the first-order difference of the joint angles; joint acceleration can be obtained by taking the second-order difference of the joint angles. However, considering that noise in the joint angle information can lead to excessive calculation errors due to the second-order difference, joint acceleration information can be obtained based on feedback information from other aspects of the robot to ensure data accuracy. Given the known dynamic model of the legged robot, the generalized momentum of the robot system can be defined as:

[0049]

[0050] Where p is the generalized momentum of the robot system, and M(q) is the robot's inertia term. For the generalized velocity of the robot system.

[0051] By differentiating the generalized momentum and combining it with the dynamic expression, we can obtain the generalized acceleration that needs to be observed, as follows:

[0052]

[0053] in, Let g(q) represent the Coriolis force and centrifugal force terms of the robot system, g(q) represent the gravity term, and q represent the generalized coordinates (the generalized coordinates are a common term in this field; in this embodiment, the coordinates include the robot's center of mass XYZ coordinates, RPY angles, and the twelve joint rotation angles θ1...θ2 in the world coordinate system). 12 ), For the generalized acceleration of the robot system, τ tot M is the joint torque term. -1 This represents the inverse matrix of the robot's inertia term M(q). Joint acceleration. Included in generalized acceleration, i.e. The accuracy of the results obtained by the joint acceleration observation method described above is much higher than that obtained by the difference method. Based on the above calculation results, accurate information is provided for the dynamic threshold of the subsequent detection model, ensuring the accuracy of foot phase detection.

[0054] In addition to obtaining high-precision joint acceleration observations, generalized momentum can also be used to estimate joint torques. Specifically, this embodiment employs a first-order torque observer based on momentum, whose frequency domain expression is as follows:

[0055]

[0056] Where λ = 15Hz is the cutoff frequency of the observer; Let S be the actual observed value of the joint torque; S be the selection matrix; s be a complex variable; τ be the current torque fed back by the motor; and g be the gravity term. Since the above equation is a frequency domain expression, the gravity term g(q) is omitted and represented as g for convenience. Observed values ​​of joint torque. The contact Jacobi of the leg and foot mechanism can be converted into the foot-observed reaction force (i.e., the foot-observed contact force), that is:

[0057]

[0058] in, Let S be the observed contact force of the i-th leg. i To select the i-th leg, J i Let be the contact Jacobian matrix of the i-th leg. This is the symbol for the pseudo-inverse of a matrix.

[0059] In step S2, based on the joint rotation angle, the coordinate position of the foot under the shoulder joint (i.e., foot kinematic information) is obtained through forward kinematics calculation. The obtained foot kinematic information is used as input to the kinematic probability model. The kinematic probability model changes the weight of the current kinematic information in real time according to the current time point in the entire gait cycle, calculates the probability of the current foot phase change of the legged robot, and calculates the threshold probability of the current state phase change in combination with the virtual body angle, thus solving the foot phase state under the kinematic probability model.

[0060] After obtaining the required foot kinematic information through proprioceptors, a kinematic probability model suitable for legged robots is established, as shown in the following equation:

[0061]

[0062] in, Let be the foot kinematics information obtained through proprioceptive feedback at time k, and erf(·) be the error function formula. As a weighting adjustment factor for foot kinematics information, It is the average value of the foot kinematics information displayed within the currently running window. This represents the sample variance of foot kinematics information fed back within the current running window. The running window refers to the duration from the current time point to a previously preset time point. This duration is set according to the actual ground conditions; the duration of the running window increases when the standard deviation of the terrain height increases, and vice versa.

[0063] In an ideal scenario, during the swing phase, the probability of a change in foot phase is... The transition should proceed from 0 to 1. However, due to disturbances caused by sudden ground changes, achieving a complete conversion of the swing leg probability during this stage becomes extremely difficult. Therefore, an encoder embedded in each joint of the robot and a kinematic transformation matrix are used to determine the orientation of each leg. Furthermore, a virtual body angle is established based on the analysis. This virtual body angle represents the angle between the lower leg link and the horizontal direction of the robot body, such as... Figure 4 As shown. Based on the average value of the virtual machine's body angle, a reasonable setting can provide a reasonable and effective threshold range for the phase change probability of the legged robot, namely the upper limit T. PU and lower limit T PL Its expression is:

[0064]

[0065] Among them, A k w is the virtual machine body at the current moment. k N represents the weight of the virtual machine's body angle value at the current moment. c -N w The running window time before this moment; A b Additional angle values ​​are added to trigger the event; A std The judgment standard value is based on the standard deviation of the virtual machine body angle; A d It is the preset decay value for event triggering.

[0066] When kinematic probability Breaking the upper limit of threshold T PU When the kinematic probability is high, the model determines that the current leg or foot has made contact with the ground; when the kinematic probability is low... Breaking the lower limit of the threshold T PL When a foot is determined to have left the ground, the model determines that the foot has transitioned from the support phase to the swing phase. When a foot is determined to have left the ground, it means that the foot has transitioned from the swing phase to the support phase, and the phase can be determined to have entered the swing phase. When a foot is determined to have left the ground, it means that the foot has transitioned from the swing phase to the support phase, and the phase can be determined to have entered the support phase.

[0067] In step S3, based on the joint torque, the three-dimensional observed force at the foot end is calculated using Jacobi. The observed force at the foot end in the vertical direction is used as input to the dynamic probability model. The dynamic probability model changes the weight of the current joint torque in the model in real time according to the peak information after filtering the three-dimensional observed force at the foot end, and calculates the probability of the current foot end phase change of the legged robot. Combined with the observed foot end velocity and acceleration information, the threshold probability of the current state phase change is calculated, and the foot end phase state under the dynamic probability model is solved.

[0068] After obtaining the required dynamic information through proprioceptors, a dynamic probabilistic model suitable for legged robots is established, as shown in the following equation:

[0069]

[0070] in, Let represent the observed foot force information in the vertical direction at time k (i.e., the three-dimensional observed foot force), and erf(·) be the error function formula. The weighting adjustment factor for observing foot force information, This represents the average value of the estimated foot force within the current running window. This represents the sample variance of the estimated foot force within the current operating window. The operating window refers to the duration from the current time point to a previously preset time point. This duration is set according to the actual ground conditions; the operating window duration increases when the standard deviation of terrain height increases, and vice versa.

[0071] Because the dynamic probabilistic model is related to the dynamics of the robot system, it differs from the kinematic probabilistic model in that it lacks visual feedback information such as virtual body angles to provide a dynamic threshold for the probabilistic model. Under this judgment model, the probability of phase change in a legged robot requires additional information on the velocity and acceleration of the foot (obtained in step S1 above) to obtain a reasonable and effective threshold range, namely the upper limit T. FU and lower limit T FL Its expression is:

[0072]

[0073] Among them, F b Additional value for the observational power required to trigger the event. and These are the velocity and acceleration threshold coefficients, respectively, where v and a are the foot velocity and acceleration, respectively. When the dynamic probability... Breaking the upper limit of threshold T FU When the dynamic probability is high, the model determines that the current leg or foot has made contact with the ground; when the dynamic probability is low, the model determines that the current leg or foot has made contact with the ground. Breaking the lower limit of the threshold T FLWhen this happens, the model determines that the current leg or foot has left the ground.

[0074] In step S4, in steps S2 and S3 above, the robot may make incorrect judgments in the probability model due to the foot slipping on the ground. Therefore, a judgment correction model based on friction cone is introduced to improve the accuracy of detection in steps S2 and S3.

[0075] In the foot-support phase, the three-dimensional observation force at the robot's foot should satisfy the friction cone constraint, i.e.:

[0076]

[0077] Among them, F x F y F z μ represents the foot-level observation force in the x, y, and z directions, respectively. s The coefficient of friction between the foot and the ground.

[0078] When determining whether a lift-off or touch-off event has occurred, the possible states of the foot are as follows: Figure 5 As shown. By calculating the duration of the friction cone, the model correction value J at time k can be obtained. k Its definition is as follows:

[0079]

[0080] Where sgn(·) is the sign function, F ·，n Let n be the observation force at time n. The obtained model correction values ​​are used as input to the preceding model in step S5.

[0081] Furthermore, considering only whether the observed three-dimensional foot force conforms to the friction cone constraint often leads to confusion between the airborne and sliding states. Given that a high-frequency signal appears in the vertical foot force during a phase-switching event, but not during foot sliding, a low-pass filter needs to be applied to the observed foot force before calculating the correction value. This eliminates the high-frequency components of the foot force to correctly calculate the correction value J. k The low-pass filter uses a mean filter as shown in the following equation:

[0082]

[0083] in, They represent, This represents the foot force information after low-pass filtering, and the above formula (7) contains... Replace with This indicates that the filter has undergone low-pass filtering.

[0084] In step S5, a phase detection pre-model based on reference time and friction is constructed by combining the model correction values. The actual foot phase transition probability is calculated by combining the foot phase state under the kinematic probability model and the dynamic probability model, and it is determined whether the current event is a lift-off event or a touch-off event, thus completing the robot foot phase detection.

[0085] Specifically, by combining the model correction values, a phase detection pre-model P(χ) based on the reference time is established. k+1 |χ k ,φ lp ) and the friction-based phase detection pre-model P(χ) k+1 |χ k ,φ fc The phase detection pre-model based on the reference time is as follows:

[0086]

[0087] in, The normalized time is represented by the ratio of the current running time t to the total gait period T, adjusted by the weighting factor δ. φ Change its importance in the model; These are the expected values ​​of the probability that the robot will maintain the current phase in the next sub-phase under the current state, where subscript 0 represents the swing phase and subscript 1 represents the support phase; Let σ be the expected value of the probability that the robot's next sub-phase is a complementary phase in the current state; σ represents the standard deviation of the above probability.

[0088] The pre-model for phase detection based on friction is:

[0089]

[0090] in, For normalized time, J is the model calibration value. k The ratio of the gait period to the total gait period T is obtained by adjusting the weighting factor δ. φ Change the importance of it in the model.

[0091] After obtaining the two preceding models, the foot phase state χ obtained from the kinematic probability model and the dynamic probability model is input as a parameter into these two preceding models. That is, the foot phase state χ under the combined kinematic probability model and the dynamic probability model is used as the parameter input. k The actual (i.e., final) foot phase transition probability is calculated as follows:

[0092]

[0093] Where [w1 w2] is the weight vector, P(χ k+1 |χk ,φ lp P(χ) represents the phase transition probability calculated using a phase detection pre-model based on reference time. k+1 |χ k ,φ fc The phase transition probability P(χ) is calculated using a friction-based phase detection pre-model. k+1 |χ k Once a certain value is reached (this threshold is an empirical value and can be manually adjusted according to the complexity of the current terrain), it can be determined whether an event of leaving the ground or touching the ground has occurred, thus realizing the robot's foot phase detection.

[0094] Furthermore, in step S6, inspired by animal bionics, a Bézier curve generated by two curve segments and 14 control points is used to describe the foot's oscillating phase trajectory, and a cosine curve is used to describe the foot's support phase trajectory, such as... Figure 6 As shown. In the initial stage, the trajectory is defined by 13 control points, while in the last quarter of the swing phase, the trajectory shrinks to two control points, producing a distinct vertical downward motion until the ground contact event occurs. The trajectory of the foot swing phase is as follows:

[0095]

[0096] Where, p sw (t) represents the trajectory of the foot swing phase, S x,z To select the matrix, c ·，k For x or z axis curve control points, For combined calculation formulas, The normalized time is calculated by the current running time t and the oscillation phase period T. sw The ratio is obtained.

[0097] Due to the nature of the movement, it is rare for the foot to leave the ground before completing its trajectory during the support phase. Therefore, a simple cosine curve is used as the foot trajectory for the support phase, i.e., the foot support phase trajectory is:

[0098]

[0099] Where, p st (t) represents the foot support phase trajectory, L is the support phase step length, and λ is the step height adjustment factor. To support the trajectory planning in the x-direction at time t, p ·0 The position of the foot in the x and z axes at the time of the ground contact event.

[0100] Based on the current foot phase state of the legged robot, a corresponding trajectory curve is selected and transmitted to the robot so that the robot's foot moves according to the planned trajectory curve.

[0101] The above method enables foot phase detection in legged robots. It can detect foot phase using only encoders and torque motors, and only proprioceptors, which has high fault tolerance. Moreover, when moving on unstructured terrain, this method can help the robot detect the leg status in real time, effectively reducing the impact on the system and demonstrating strong adaptability to external environments.

[0102] Example 2

[0103] This embodiment provides an event-triggered foot phase detection system for a legged robot, including:

[0104] The data acquisition module is used to acquire the observed values ​​of the joint angles and joint torques of the legged robot at the current moment, and to obtain the robot's foot velocity and acceleration based on the joint angles through kinematic calculations.

[0105] The initial foot phase detection module is used to calculate foot kinematic information based on joint rotation angles using forward kinematics, input the foot kinematic information into the kinematic probability model, and solve the foot phase state under the kinematic probability model by combining virtual body angles; based on joint torques, it calculates the three-dimensional observation force of the foot using Jacobi, inputs the three-dimensional observation force of the foot into the dynamic probability model, and solves the foot phase state under the dynamic probability model by combining foot velocity and acceleration;

[0106] The correction coefficient calculation module is used to obtain model correction values ​​based on the three-dimensional observation force at the foot end by calculating the duration of the friction cone.

[0107] The foot phase detection module is used to construct a phase detection pre-model based on reference time and friction by combining model correction values, and to calculate the actual foot phase transition probability by combining the foot phase state under the kinematic probability model and the dynamic probability model, to determine whether the current event is a lift-off event or a touch-off event, and to complete the robot's foot phase detection.

[0108] Further technical solutions also include:

[0109] The foot motion drive module is used to describe the foot swing phase trajectory using a Bezier curve generated by two curves and 14 control points, and to describe the foot support phase trajectory using a cosine curve. Based on the foot phase state of the legged robot at the current time point, the module selects the corresponding trajectory curve so that the robot's foot moves according to the planned trajectory curve.

[0110] The steps and methods involved in the above embodiment two correspond to those in embodiment one. For specific implementation details, please refer to the relevant description section of embodiment one.

[0111] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0112] The above description is only a preferred embodiment of the present invention. Although the specific implementation of the present invention has been described in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative effort are still within the scope of protection of the present invention.

Claims

1. A method for phase detection at the foot of a legged robot based on event triggering, characterized in that, include: The system obtains the observed values ​​of joint rotation angles and joint torques of the legged robot at the current moment, and calculates the robot's foot velocity and acceleration based on the joint rotation angles through kinematic calculations. Based on joint rotation angles, foot kinematics information is calculated through forward kinematics. This foot kinematics information is then input into a kinematic probability model. Combined with virtual body angles, the foot phase state under the kinematic probability model is solved. Based on joint torque, the three-dimensional observation force at the foot end is calculated using Jacobi. The three-dimensional observation force at the foot end is then input into the dynamic probability model. Combined with the foot end velocity and acceleration, the foot end phase state under the dynamic probability model is solved. Based on the three-dimensional observation force at the foot, the model correction value is obtained by calculating the duration of the friction cone. By combining model correction values, a phase detection pre-model based on reference time and friction is constructed. Combined with the foot phase state under the kinematic probability model and the dynamic probability model, the actual foot phase transition probability is calculated to determine whether the current event is a lift-off event or a touch-off event, thus completing the robot's foot phase detection.

2. The event-triggered foot phase detection method for legged robots as described in claim 1, characterized in that it further... include: The trajectory of the foot swing phase is described by a Bézier curve generated by two curve segments and 14 control points, and the trajectory of the foot support phase is described by a cosine curve. Based on the current foot phase state of the legged robot, the corresponding trajectory curve is selected so that the robot's foot moves according to the planned trajectory curve.

3. The event-triggered foot phase detection method for legged robots as described in claim 1, characterized in that, The solution of the foot phase state under the kinematic probability model includes: Based on joint rotation angles, foot kinematics information is obtained through forward kinematics calculations; the foot kinematics information refers to the coordinate position of the foot below the shoulder joint. The obtained foot kinematics information is used as input to the kinematic probability model. The kinematic probability model changes the weight of the current kinematic information in real time according to the current time point in the entire gait cycle, and calculates the probability of the current foot phase change of the legged robot. By combining the virtual machine body angle, the threshold probability of the current state phase change is calculated, and the foot phase state under the kinematic probability model is solved.

4. The event-triggered foot phase detection method for legged robots as described in claim 3, characterized in that, The kinematic probability model is as follows: in, Let be the foot kinematics information obtained through proprioceptive feedback at time k, and erf(·) be the error function formula. As a weighting adjustment factor for foot kinematics information, It is the average value of the foot kinematics information displayed within the currently running window. It is the sample variance of the foot kinematics information fed back within the current running window.

5. The event-triggered foot phase detection method for legged robots as described in claim 1, characterized in that, The solution of the foot phase state under the dynamic probability model includes: Based on joint torque, the three-dimensional observation force at the foot end is obtained by Jacobi calculation; The observed force at the foot in the vertical direction is used as input to the dynamic probability model. The dynamic probability model changes the weight of the current joint torque in the model in real time based on the peak information of the filtered three-dimensional observed force at the foot and calculates the probability of the current foot phase change of the legged robot. By combining the observed foot velocity and acceleration information, the threshold probability of the current state phase change is calculated, and the foot phase state under the dynamic probability model is solved.

6. The event-triggered foot phase detection method for legged robots as described in claim 5, characterized in that, The dynamic probability model is as follows: in, Let represent the observed foot force information in the vertical direction at time k (i.e., the three-dimensional observed foot force), and erf(·) be the error function formula. The weighting adjustment factor for observing foot force information, This represents the average value of the estimated foot force within the current running window. This represents the sample variance of the estimated foot force within the current running window.

7. The event-triggered foot phase detection method for legged robots as described in claim 1, characterized in that, Combining model calibration values, a phase detection pre-model based on reference time and friction force is constructed, wherein the phase detection pre-model based on reference time is as follows: In the above formula, The normalized time is represented by δ, which is obtained by the ratio of the current running time t to the total gait period T. φ Indicates the weight adjustment factor; The pre-model for phase detection based on friction is: in, For normalized time, J is the model calibration value. k δ is obtained by calculating the ratio of δ to the total gait period T. φ Indicates the weight adjustment factor; These are the expected values ​​of the probability that the robot will maintain the current phase in the next sub-phase under the current state, where subscript 0 represents the swing phase and subscript 1 represents the support phase; Let σ be the expected value of the probability that the robot's next sub-phase is a complementary phase in the current state; σ represents the standard deviation of the probability.

8. The event-triggered foot phase detection method for legged robots as described in claim 7, characterized in that, Combining the foot phase state under the kinematic probability model and the dynamic probability model, the actual foot phase transition probability is calculated, including: inputting the foot phase state obtained from the kinematic probability model and the dynamic probability model as parameters into the phase detection pre-model based on reference time and friction, and calculating the actual foot phase transition probability, which is: Where [w1 w2] is the weight vector, P(χ k+1 |χ k ,φ lp P(χ) represents the phase transition probability calculated using a phase detection pre-model based on reference time. k+1 |χ k ,φ fc ) represents the phase transition probability calculated using a friction-based phase detection pre-model.

9. A foot phase detection system for a legged robot based on event triggering, characterized in that, include: The data acquisition module is used to acquire the observed values ​​of the joint angles and joint torques of the legged robot at the current moment, and to obtain the robot's foot velocity and acceleration based on the joint angles through kinematic calculations. The foot phase initial detection module is used to calculate foot kinematic information based on joint rotation angle through positive kinematics, input the foot kinematic information into the kinematic probability model, and solve the foot phase state under the kinematic probability model by combining the virtual body angle. Based on joint torque, the three-dimensional observation force at the foot end is calculated using Jacobi. The three-dimensional observation force at the foot end is then input into the dynamic probability model. Combined with the foot end velocity and acceleration, the foot end phase state under the dynamic probability model is solved. The correction coefficient calculation module is used to obtain model correction values ​​based on the three-dimensional observation force at the foot end by calculating the duration of the friction cone. The foot phase detection module is used to construct a phase detection pre-model based on reference time and friction by combining model correction values, and to calculate the actual foot phase transition probability by combining the foot phase state under the kinematic probability model and the dynamic probability model, to determine whether the current event is a lift-off event or a touch-off event, and to complete the robot's foot phase detection.

10. The event-triggered foot phase detection system for a legged robot as described in claim 9, characterized in that it further... include: The foot motion drive module is used to describe the foot swing phase trajectory using a Bezier curve generated by two curves and 14 control points, and to describe the foot support phase trajectory using a cosine curve. Based on the foot phase state of the legged robot at the current time point, the module selects the corresponding trajectory curve so that the robot's foot moves according to the planned trajectory curve.

Images