Control methods, devices, robots, and storage media for frog-inspired robots
By combining machine learning models with the state and environmental data of the frog-like robot, the problem of inaccurate movement in dynamic waters was solved, and precise control in complex waters was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TIANHE COLLEGE GUANGDONG POLYTECHNIC NORMAL UNIV
- Filing Date
- 2024-02-05
- Publication Date
- 2026-06-30
AI Technical Summary
Existing methods for controlling the movement of frog-like robots in water are unable to perceive environmental changes in dynamic water environments, resulting in low movement accuracy.
By employing a machine learning model, target parameters are determined by acquiring robot state data and environmental data, and then input into the machine learning model to output operating parameters, thereby controlling the robot's movement in water.
The accuracy of the frog-like robot's movement in dynamic waters has been improved. By controlling the operating parameters driven by environmental data, the robot's precise movement in complex waters has been achieved.
Smart Images
Figure CN117961899B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of robotics, and more particularly to a control method for a frog-like robot, the frog robot itself, and a storage medium thereof. Background Technology
[0002] The frog-like robot is an amphibious robot. Its movement control in water is more complex than that of propeller-driven machines. Currently, the driving method for frog-like robots in water involves setting a fixed motor control program based on the task requirements to enable the robot to move forward and turn in water. However, this driving method is limited to static water areas. In dynamic water areas, the inability to perceive changes in the environment leads to lower accuracy in movement.
[0003] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention
[0004] The main objective of this invention is to provide a control method, a control device, a robot, and a storage medium for a frog-like robot, aiming to improve the accuracy of controlling the robot's movement in water.
[0005] To achieve the above objectives, the present invention provides a control method for a frog-like robot, the control method comprising the following steps:
[0006] At the initial moment when the robot is detected to be in the running cycle, the robot's status data and the environmental data of the robot's location are acquired;
[0007] The target parameters for inputting into the machine learning model are determined based on the state data and the environment data, and the target parameters are input into the machine learning model to output the running parameters for the running cycle.
[0008] The robot is controlled according to the operating parameters.
[0009] Optionally, the machine learning model includes a neural network model, and the state data includes: current position information at the initial moment, historical position information, and historical operating parameters, wherein the historical position information and the historical operating parameters are respectively the current position information and operating parameters of adjacent operating cycles before the initial moment. Before the step of inputting the target parameters into the machine learning model and outputting the operating parameters of the operating cycle, the method further includes:
[0010] The historical location information, the current location information, and the environmental data are input into the neural network model, and calibration operation parameters are output.
[0011] The loss value is determined based on the historical operating parameters and the calibration operating parameters;
[0012] The neural network model is adjusted based on the loss value.
[0013] Optionally, the machine learning model is an ensemble learning model, which includes at least two individual learners and a combination strategy. The step of inputting the target parameters into the machine learning model and outputting the running parameters for the running cycle includes:
[0014] The target parameters are input into at least two individual learners to obtain at least two individual output results;
[0015] The operating parameters are determined based on the combination strategy and the output results of at least two individuals.
[0016] Optionally, the state data includes the current position information at the initial time, the first target position information, and the second target position information of the previous cycle of the running cycle corresponding to the initial time. Before the step of inputting the target parameters into the machine learning model and outputting the running parameters of the running cycle, the method further includes:
[0017] When the distance between the current location information and the second target location information is greater than a preset distance threshold, update the first target location, and / or adjust the combination strategy;
[0018] When the distance between the current location information and the second target location information is less than or equal to a preset distance threshold, the step of inputting the target parameters into the machine learning model and outputting the running parameters of the running cycle is determined.
[0019] Optionally, the combination strategy is a predicted value weighting method, and the step of adjusting the combination strategy includes:
[0020] Based on whether the environmental data is consistent with the historical environmental data of the previous cycle corresponding to the initial time;
[0021] When the environmental data and the historical environmental data are inconsistent, update the weighted weights of at least two of the individual learners.
[0022] When the environmental data and the historical environmental data are consistent, update the corresponding weighted weights of all individual learners.
[0023] Optionally, the state data includes the robot's moving speed, moving direction, current position information at the initial moment, and first target position information; the environmental data includes the water flow direction and water flow speed; and the step of determining the target parameters for inputting the machine learning model based on the state data and the environmental data includes:
[0024] Based on the status data and the environmental data, the target parameters are determined to be the moving speed, the moving direction, the current location information, the first target location information, the water flow direction, and the water flow speed.
[0025] Optionally, before the step of acquiring the robot's state data and the environmental parameter data of the robot's location at the initial moment of detecting that the robot is in the running cycle, the method further includes:
[0026] The number of operating cycles and the target position for each operating cycle are determined based on the robot's starting position, ending position, and initial environmental data.
[0027] Furthermore, to achieve the above objectives, the present invention also provides a control device for a frog-like robot, the control device comprising:
[0028] The detection module is used to acquire the robot's status data and the environmental data of the robot's location at the initial moment when the robot is detected to be in the running cycle.
[0029] The calculation module is used to determine the target parameters for inputting into the machine learning model based on the state data and the environment data, input the target parameters into the machine learning model, and output the running parameters for the running cycle;
[0030] A control module is used to control the robot according to the operating parameters.
[0031] Furthermore, to achieve the above objectives, the present invention also provides a robot, the robot comprising: a memory, a processor, and a control program for a frog-like robot stored in the memory and executable on the processor, the control program for the frog-like robot being configured to implement the steps of the control method for the frog-like robot described in any of the above claims.
[0032] In addition, to achieve the above objectives, the present invention also provides a storage medium storing a control program for a frog-like robot, wherein when the control program for the frog-like robot is executed by a processor, it implements the steps of the control method for the frog-like robot described in any of the above claims.
[0033] This invention proposes a control method for a frog-inspired robot. The method includes: acquiring the robot's state data and environmental data of its location at the initial moment of a running cycle; determining target parameters for inputting into a machine learning model based on the state data and environmental data, inputting the target parameters into the machine learning model, and outputting the running parameters for the running cycle; and controlling the robot according to the running parameters. By using the robot's state data and environmental data at the initial moment of the running cycle, the method identifies and perceives the environmental conditions and the robot's own state. By determining the target parameters for inputting into the machine learning model, the required running parameters for the current cycle can be determined. Since the running parameters are determined based on environmental data, the robot's operation can be accurately controlled even in dynamic water environments. Attached Figure Description
[0034] Figure 1 This is a schematic diagram of the structure of the robot in the hardware operating environment involved in the embodiments of the present invention;
[0035] Figure 2 This is a flowchart illustrating the first embodiment of the control method for the frog-like robot of the present invention;
[0036] Figure 3 This is a flowchart illustrating the second embodiment of the control method for the frog-like robot of the present invention.
[0037] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0038] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0039] Reference Figure 1 , Figure 1 This is a schematic diagram of the robot structure in the hardware operating environment involved in the embodiments of the present invention.
[0040] like Figure 1As shown, the robot may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a motion device 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The motion device 1003 may include motors, and optionally, the motion device 1003 may also be connected to the communication bus 1002 via a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.
[0041] Furthermore, the motor can be a synchronous motor, an asynchronous motor, a DC motor, etc., and the DC motor can be controlled using methods such as Proportional Integral Differential (PID) control and Pulse Width Modulation (PWM) control. PID control is the most commonly used method, which controls the motor's movement by adjusting the voltage and current of the motor power supply through feedback control of the motor's speed, position, and other states, thereby controlling the movement of the frog-like robot's thighs, calves, and other parts, enabling the robot to move in water. The operating parameters can be converted into specific control parameters for PID control to control the motor.
[0042] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the robot and may include more or fewer parts than shown, or combine certain parts, or have different arrangements of parts.
[0043] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and a control program for a frog-like robot.
[0044] exist Figure 1In the robot shown, the network interface 1004 is mainly used for data communication with other devices; the motion device 1003 is mainly used to realize the movement of the robot; the processor 1001 and the memory 1005 in the robot of the present invention can be set in the robot. The robot calls the control program of the frog-like robot stored in the memory 1005 through the processor 1001 and executes the control method of the frog-like robot provided in the embodiment of the present invention.
[0045] This invention provides a control method for a frog-like robot, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of a control method for a frog-like robot according to the present invention.
[0046] In this embodiment, the control method for the frog-like robot includes:
[0047] Step S10: At the initial moment when the robot is detected to be in the running cycle, acquire the robot's state data and the environmental data of the robot's location;
[0048] The robot in this embodiment refers to a frog-like robot or a robot with a similar movement pattern to a frog. The state data includes: the current position information at the initial moment, the state data of the running cycle before the initial moment, and the first target position within the running cycle. The current position information includes the position coordinates of the robot's torso, the direction of the torso, the robot's direction of movement, and its speed. The first target position is generally preset, but can also be updated after the end of the previous running cycle. Specifically, the state data of the running cycle before the initial moment can include the position coordinates of the robot's torso and the direction of the torso in the cycle preceding the initial moment. Specifically, when the robot moves in water, the environmental data here is the direction and speed of the water flow. Of course, it is not limited to the environmental data of the robot's movement in other scenarios, such as the tilt angle of the ground, the friction between the ground and the frog, etc. The initial moment of the running cycle here refers to the time after the frog-like robot has completed all the actions in one running cycle and before the actions in the next running cycle begin. The running cycle here can be preset by the user or controller.
[0049] Step S20: Determine the target parameters for inputting into the machine learning model based on the state data and the environment data, input the target parameters into the machine learning model, and output the running parameters of the running cycle;
[0050] The target parameters refer to the parameters input to the machine learning model, which can be an algorithm such as a decision tree, support vector machine, or neural network. The operating parameters are the rotation angle and speed of each motor on the robot at various moments during the operating cycle.
[0051] Step S30: Control the robot according to the operating parameters.
[0052] The operation of each motor of the robot is controlled according to the operating parameters to enable the robot to reach the first target position. For example, the robot includes two motors located at the hip joint, which control the movement of the left and right thighs respectively. Motors are also located at the left and right knees, controlling the movement of the left and right lower legs respectively. The rotation angle and speed of these four motors are controlled according to the operating parameters to drive the movement of the robot's torso. Of course, the more motors on the torso, the more degrees of freedom the thighs have. The above example is not intended to limit the number of motors on the robot.
[0053] In this embodiment, by determining the target parameters for inputting into the machine learning model and inputting the target parameters into the machine learning model, the operating parameters required for the current cycle can be determined. Since the operating parameters are determined based on environmental data, the robot's operation can be accurately controlled even in dynamic waters.
[0054] Furthermore, based on the first embodiment, a second embodiment of the control method for the frog-like robot of the present invention is proposed. In this embodiment, reference is made to... Figure 3 The machine learning model includes a neural network model, and the state data includes: the current position information at the initial moment, historical position information, and historical operating parameters. The historical position information and the historical operating parameters are respectively the current position information and operating parameters of the adjacent operating cycles before the initial moment. Before the step of inputting the target parameters into the machine learning model and outputting the operating parameters of the operating cycle, the method further includes:
[0055] Step S201: Input the historical location information, the current location information, and the environmental data into the neural network model, and output calibration operation parameters;
[0056] The neurons in the neural network are arranged in layers. Each neuron is connected only to the neurons in the layer above it, receives the output of the previous layer, and outputs it to the next layer. The neural network includes an input layer, a hidden layer, and an output layer. The neural network model is obtained by training a training dataset. The training dataset includes input features and output results. In this embodiment, specifically, the input features are: initial position, target position, water flow direction, water flow velocity, robot direction, and speed. The output results are the angles and speeds of the various motors set on the robot during the operating cycle. There is no limitation on the input features. Before step S201, the process also includes: training the neural network model. The steps of training the neural network model include: controlling the robot's motors to operate in different ways within an operating cycle; recording the angles and speeds of the motors, the robot's initial position, target position, water flow direction, water flow velocity, body direction, and body speed during the operating cycle as a training dataset; inputting the data into the neural network for training to obtain the neural network model.
[0057] In this embodiment, the historical location information includes not only the specific location but also the robot's direction of movement and speed at that location. The location in the historical location information is used as the initial location, the current location information as the target location, and the direction and speed of movement in the historical location information as the body direction and speed. These are input into the neural network model to output calibration operation parameters.
[0058] Step S202: Determine the loss value based on the historical operating parameters and the calibration operating parameters;
[0059] The historical operating parameters here refer to the parameters used to control the robot's operation in the previous cycle at the initial moment. These historical operating parameters control the robot to move to the historical target position corresponding to the previous cycle. It's easy to understand that when the neural network model outputs correct historical operating parameters, the current position information should be the historical target position. However, if the current position information is not the historical target position, it can be determined that the historical operating parameters output by the neural network model have an error, meaning the neural network model needs adjustment. In this embodiment, the historical operating parameters are not used as parameters to determine the error, but rather as parameters to verify against the calibration operating parameters. Since the historical operating parameters enable the robot to reach the current position, they are the correct parameters from the historical position information to the current position. Specifically, the loss value is determined by inputting the historical operating parameters and the calibration operating parameters into a loss function. The loss function can be the mean squared error or the cross-entropy error. In this embodiment, the loss function is:
[0060]
[0061] Here, E is the loss value, n is the number of motors, and k i M represents the error coefficient corresponding to the i-th motor. i Let be the error value of the i-th motor.
[0062] The M i The calculation method is as follows:
[0063]
[0064] Here, T represents the total duration within the robot's operating cycle, l it Let z be the operating parameter value of the i-th motor at time t in the historical operating parameters. it Let l be the operating parameter value of the i-th motor at time t in the calibration operating parameters. it and z it The parameter types are the same, and the calculation method for the operating parameter value is to multiply the angle value of the motor at a certain moment by the speed value of the motor. This embodiment does not limit the specific data type of the operating parameter value; different robots may have different motor control parameters. For example, when the motor parameter only has an angle value, the l... it and z it It can be the angle parameter of the motor, i.e., l it Let z be the angle value of the i-th motor at time t in the historical operating parameters. it Let be the angle value of the i-th motor at time t in the calibration operation parameters.
[0065] It should be noted that k i k is related to the distance between the motor and the robot's body. i The distance value is negatively correlated. Since the influence of motors at different positions on the robot is inconsistent, different k values are set for different motors. i This can reduce the impact of the above factors on robot movement. i The direction and velocity of the water flow are negatively correlated with the total duration of the robot's operation cycle; the lower the correlation, the greater the correlation. i The larger the value, the better. This helps to avoid the influence of environmental changes on the actual loss value and improves the accuracy of the loss value.
[0066] Step S203: Adjust the neural network model based on the loss value.
[0067] Specifically, the magnitude of the loss value determines whether the neural network model needs adjustment. When it is determined that the neural network model needs adjustment based on the loss value, the backpropagation algorithm is used to calculate the gradient of the loss function relative to the model parameters, and gradient descent is used to update the model's weights and biases, thereby adjusting the neural network model. The gradient descent can be stochastic gradient descent, batch gradient descent, mini-batch gradient descent, etc.
[0068] In this embodiment, by inputting the historical location information, the current location information, and the environmental data into the neural network model, outputting calibration operation parameters, and using the historical operation parameters to determine the loss value, it is possible to determine whether there is an error in the output result of the neural network model. The neural network model can be adjusted by the loss value to improve the accuracy of the output operation parameters, thereby improving the accuracy of robot movement.
[0069] Furthermore, based on the first or second embodiment, a third embodiment of the control method for the frog-like robot of the present invention is proposed. In this embodiment, the machine learning model is ensemble learning, which includes at least two individual learners and a combination strategy. The step of inputting the target parameters into the machine learning model and outputting the running parameters of the running cycle includes:
[0070] The target parameters are input into at least two individual learners to obtain at least two individual output results;
[0071] The operating parameters are determined based on the combination strategy and the output results of at least two individuals.
[0072] It should be noted that this embodiment does not conflict with the second embodiment. The individual learner in ensemble learning is a machine learning model. Different individual learners can be trained using the same neural network based on different training datasets. Specifically, the individual learner is the neural network model in the second embodiment, capable of both receiving the target parameters and outputting the corresponding results. However, in the case based on the first embodiment, the type of individual learner is not limited; it can be a decision tree, support vector machine, neural network, etc. The combination strategy here specifically processes the output results of at least two individuals to determine the accurate operating parameters.
[0073] In this embodiment, the operating parameters are determined by combining the results obtained from different individual learners through ensemble learning. This improves the accuracy of the operating parameters, thereby enhancing the accuracy of robot control.
[0074] Furthermore, based on the third embodiment, a fourth embodiment of the control method for the frog-like robot of the present invention is proposed. In this embodiment, the state data includes the current position information at the initial moment, the first target position, and the second target position of the previous cycle of the running cycle corresponding to the initial moment. Before the step of inputting the target parameters into the machine learning model and outputting the running parameters of the running cycle, the method further includes:
[0075] When the distance between the current location information and the second target location is greater than a preset distance threshold, update the first target location, and / or adjust the combination strategy;
[0076] When the distance between the current location information and the second target location information is less than or equal to a preset distance threshold, the step of inputting the target parameters into the machine learning model and outputting the running parameters of the running cycle is determined.
[0077] It should be noted that the difference between the first target position and the second target position is that the operating cycle corresponding to the first target position is the cycle that is adjacent to and follows the operating cycle corresponding to the second target position. The distance between the current position and the second target position refers to the straight-line distance. The step of updating the first target position includes: determining the updated first target position based on the current position information, environmental data, and the robot's mobility.
[0078] In this embodiment, the distance between the current position information and the second target position determines whether to change the first target position, so that a suitable target position can be obtained in each robot's operating cycle, thereby avoiding the position deviation at the end of the previous cycle from affecting the robot's operating parameters in the next cycle, thus improving the accuracy of robot operation.
[0079] Furthermore, based on the fourth embodiment, a fifth embodiment of the control method for the frog-like robot of the present invention is proposed. In this embodiment, the combination strategy is a predicted value weighted method, and the step of adjusting the combination strategy includes:
[0080] Based on whether the environmental data is consistent with the historical environmental data of the previous cycle corresponding to the initial time;
[0081] When the environmental data and the historical environmental data are inconsistent, update the weighted weights of at least two of the individual learners.
[0082] When the environmental data and the historical environmental data are consistent, update the corresponding weighted weights of all individual learners.
[0083] The weighted prediction method here multiplies the output of each model by its corresponding weight, and then sums the weighted predictions to obtain the final running parameters. The types of environmental data and historical environmental data are consistent. Specifically, environmental data includes the direction and velocity of water flow, while historical environmental data should also include the direction and velocity of water flow. The method determines whether the direction of water flow in the environmental data is consistent with that in the historical environmental data, and whether the velocity of water flow in the environmental data is equal to that in the historical environmental data. When adjusting the weights, cross-validation and hold-out methods are used to determine if the model's output is normal, and the corresponding weights are adjusted accordingly. In other embodiments, the combination strategy is a learning strategy. Specifically, the steps for adjusting the combination strategy include: determining the evaluation value of the previous running parameters for the period preceding the period corresponding to the initial time based on the distance between the current location information and the second target location information; using the evaluation value and the individual output of each individual learner for the period preceding the period corresponding to the initial time as new training samples and adding them to the training set of the learning strategy; and retraining the learning strategy based on the training set.
[0084] In this embodiment, the weighting of the individual learner is updated by comparing the environmental data with the historical environmental data, thereby improving the accuracy of the output operating parameters.
[0085] Furthermore, based on any of the above embodiments, a sixth embodiment of the control method for the frog-like robot of the present invention is proposed. In this embodiment, the state data includes the robot's moving speed, moving direction, current position information at the initial moment, and first target position information; the environmental data includes the water flow direction and water flow speed; and the step of determining the target parameters for inputting the machine learning model based on the state data and the environmental data includes:
[0086] Based on the status data and the environmental data, the target parameters are determined to be the moving speed, the moving direction, the current location information, the first target location information, the water flow direction, and the water flow speed.
[0087] In this embodiment, the target parameters are determined to correspond to the input parameters required by the machine learning model based on the state data and the environment data. Specifically, the state data and the environment data are preprocessed and normalized, for example, by determining the coordinates corresponding to the current position information and the first target position information based on the same coordinate system. The target parameters are input to the input layer of the neural network model. Since there are 6 neurons corresponding to the target parameters, the number of neurons in the input layer can be 6. The input layer passes the parameters to the hidden layer, which includes fully connected layers, convolutional layers, recurrent layers, etc. Finally, the hidden layer passes the parameters to the output layer, outputting the running parameters.
[0088] In this embodiment, determining the target parameters using the state data and the environmental data ensures the correctness of the target parameters and adapts them to the environmental conditions, thereby improving the accuracy of the output operating parameters.
[0089] Furthermore, based on any of the above embodiments, a seventh embodiment of the control method for the frog-like robot of the present invention is proposed, which further includes, before the step of acquiring the robot's state data and the environmental parameter data of the robot's location at the initial moment of detecting that the robot is in the running cycle, the method further includes:
[0090] The number of operating cycles and the target position for each operating cycle are determined based on the robot's starting position, ending position, and initial environmental data.
[0091] Specifically, the starting position, ending position, and initial environmental data determine the robot's running path. This running path is then divided into multiple sub-running paths, each corresponding to one running cycle. The endpoint of each sub-running path is designated as the target position, i.e., the first target position in the above embodiment. Further, after controlling the robot according to the running parameters, it is determined whether the endpoint position has been reached. If the endpoint position has not been reached, a new running cycle begins, and the process returns to step S10.
[0092] In this embodiment, the number of operating cycles and the target position of each operating cycle are determined based on the robot's starting position, ending position, and initial environmental data. The number of operating cycles is negatively correlated with the length of the operating cycle, which can effectively limit the length of the operating cycle. Since each cycle requires environmental data to determine the operating parameters of the operating cycle, the shorter the length of the operating cycle, the more times the robot utilizes the environmental data, thereby enabling timely and accurate control of the robot's operation in dynamic waters based on the water flow conditions.
[0093] Furthermore, this invention also proposes a control device for a frog-like robot, the control device comprising:
[0094] The detection module is used to acquire the robot's status data and the environmental data of the robot's location at the initial moment when the robot is detected to be in the running cycle.
[0095] The calculation module is used to determine the target parameters for inputting into the machine learning model based on the state data and the environment data, input the target parameters into the machine learning model, and output the running parameters for the running cycle;
[0096] A control module is used to control the robot according to the operating parameters.
[0097] Furthermore, this invention also proposes a robot, which includes: a memory, a processor, and a frog-like robot control program stored in the memory and executable on the processor. The frog-like robot control program is configured to implement the steps of any of the above-described frog-like robot control method embodiments.
[0098] Furthermore, this invention also proposes a storage medium storing a control program for a frog-like robot. When the control program for the frog-like robot is executed by a processor, it implements the steps of any of the above-described embodiments of the control method for the frog-like robot.
[0099] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0100] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.
[0102] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.
Claims
1. A control method for a frog-inspired robot, characterized in that, The control method for the frog-like robot includes the following steps: At the initial moment when the robot is detected to be in the running cycle, the robot's status data and the environmental data of the robot's location are acquired; The target parameters for inputting into the machine learning model are determined based on the state data and the environment data, and the target parameters are input into the machine learning model to output the running parameters for the running cycle. Control the robot according to the operating parameters; The machine learning model includes a neural network model, and the state data includes: current position information at the initial moment, historical position information, and historical running parameters. The historical position information and the historical running parameters are respectively the current position information and running parameters of adjacent running cycles before the initial moment. Before the step of inputting the target parameters into the machine learning model and outputting the running parameters of the running cycle, the method further includes: The historical location information, the current location information, and the environmental data are input into the neural network model, and calibration operation parameters are output. The loss value is determined based on the historical operating parameters and the calibration operating parameters; The neural network model is adjusted based on the loss value.
2. The control method for the frog-like robot as described in claim 1, characterized in that, The machine learning model is an ensemble learning model, which includes at least two individual learners and a combination strategy. The step of inputting the target parameters into the machine learning model and outputting the running parameters for the running cycle includes: The target parameters are input into at least two individual learners to obtain at least two individual output results; The operating parameters are determined based on the combination strategy and the output results of at least two individuals.
3. The control method for the frog-like robot as described in claim 2, characterized in that, The state data includes the current position information at the initial time, the first target position information, and the second target position information of the previous cycle of the running cycle corresponding to the initial time. Before the step of inputting the target parameters into the machine learning model and outputting the running parameters of the running cycle, the method further includes: When the distance between the current location information and the second target location information is greater than a preset distance threshold, update the first target location, and / or adjust the combination strategy; When the distance between the current location information and the second target location information is less than or equal to a preset distance threshold, the step of inputting the target parameters into the machine learning model and outputting the running parameters of the running cycle is determined.
4. The control method for the frog-like robot as described in claim 3, characterized in that, The combination strategy is a predicted value weighting method, and the step of adjusting the combination strategy includes: Based on whether the environmental data is consistent with the historical environmental data of the previous cycle corresponding to the initial time; When the environmental data and the historical environmental data are inconsistent, update the weighted weights of at least two of the individual learners. When the environmental data and the historical environmental data are consistent, update the corresponding weighted weights of all individual learners.
5. The control method for the frog-like robot as described in claim 1, characterized in that, The state data includes the robot's movement speed, movement direction, current position information at the initial moment, and first target position information. The environmental data includes the water flow direction and water flow speed. The step of determining the target parameters for inputting the machine learning model based on the state data and the environmental data includes: Based on the status data and the environmental data, the target parameters are determined to be the moving speed, the moving direction, the current location information, the first target location information, the water flow direction, and the water flow speed.
6. The control method for the frog-like robot as described in any one of claims 1 to 5, characterized in that, Before the step of acquiring the robot's state data and the environmental data of the robot's location at the initial moment of detecting that the robot is in the running cycle, the method further includes: The number of operating cycles and the target position for each operating cycle are determined based on the robot's starting position, ending position, and initial environmental data.
7. A control device for a frog-inspired robot, characterized in that, The control device for the frog-like robot includes: The detection module is used to acquire the robot's status data and the environmental data of the robot's location at the initial moment when the robot is detected to be in the running cycle. The calculation module is used to determine the target parameters for inputting into the machine learning model based on the state data and the environment data, input the target parameters into the machine learning model, and output the running parameters for the running cycle; The control module is used to control the robot according to the operating parameters; The machine learning model includes a neural network model, and the state data includes: current position information at the initial moment, historical position information, and historical running parameters. The historical position information and the historical running parameters are respectively the current position information and running parameters of adjacent running cycles before the initial moment. Before the step of inputting the target parameters into the machine learning model and outputting the running parameters of the running cycle, the method further includes: The historical location information, the current location information, and the environmental data are input into the neural network model, and calibration operation parameters are output. The loss value is determined based on the historical operating parameters and the calibration operating parameters; The neural network model is adjusted based on the loss value.
8. A robot, characterized in that, The robot includes: a memory, a processor, and a control program for the frog-like robot stored in the memory and executable on the processor, the control program being configured to implement the steps of the control method for the frog-like robot as described in any one of claims 1 to 6.
9. A storage medium, characterized in that, The storage medium stores a control program for a frog-like robot, which, when executed by a processor, implements the steps of the control method for a frog-like robot as described in any one of claims 1 to 6.