A method for intelligent agent motion control based on multi-objective, multi-step time-domain prediction

Abstract

This invention belongs to the technical field of intelligent agent control, and discloses an intelligent agent motion control method based on multi-objective, multi-step, variable time-domain prediction. The method includes: designing multi-objective optimization indices using the adjusted curvature and control period as preconditions; setting constraint information and constructing a velocity and prediction time-domain parameter library under different conditions; verifying the stability of the control system through the defined terminal domain and cost function constraints, and re-partitioning the terminal domain; designing controller parameter switching logic rules based on the position of the terminal domain where the system's final state is located; selecting appropriate parameters from the velocity and time-domain parameter library based on dynamic changes to achieve adaptive adjustment; and obtaining the control information at that moment using nonlinear model predictive control optimization to perform multi-step control of the intelligent agent control system. This invention aims to solve the motion control problem of intelligent agents, improving the accuracy and real-time performance of motion control.

Description

Technical Field

[0001] This invention belongs to the technical field of intelligent agent motion control, and more specifically, relates to an intelligent agent motion control method based on multi-target multi-step variable time-domain prediction. Background Technology

[0002] Motion control, as one of the core technologies in the field of intelligent agent research, requires an efficient control system to ensure stable motion and precise operation of the agent. Trajectory tracking control is a major research area in motion control. Model Predictive Control (MPC) is a feedback control strategy that has been widely studied and discussed in recent years. Its advantages include the ability to impose various constraints during the control process and its applicability to constrained and uncertain nonlinear multi-input multi-output systems. MPC methods obtain the optimal control sequence by solving a finite-time-domain constrained optimization problem, and then apply the first element of this sequence to the controller, repeating this process. It possesses advantages such as strong robustness and good control performance. With the development of MPC, the stability and feasibility of nonlinear MPC algorithms have also been gradually proven. This control algorithm is more suitable for systems with stronger nonlinear characteristics, such as intelligent agents, without reducing the accuracy of the system model, thus achieving higher control performance.

[0003] However, there are still some challenges in applying nonlinear model predictive control to intelligent agent control systems: 1) Since intelligent agent control systems are in a dynamic state, changes in system parameters and the environment will affect the control effect of nonlinear model predictive control; 2) Nonlinear model predictive control algorithms only use fixed time domain or single-objective optimization time domain methods, making it difficult to simultaneously meet the requirements of system stability, improved control accuracy, and real-time performance; 3) Due to the high computational complexity of nonlinear model predictive control, the solution time required is longer, making it difficult to guarantee real-time performance. Summary of the Invention

[0004] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides an intelligent agent motion control method based on multi-objective, multi-step variable time-domain prediction, which can efficiently adapt to the dynamic changes of the intelligent agent under multi-objective optimization conditions.

[0005] To achieve the above objectives, according to one aspect of the present invention, a method for intelligent agent motion control based on multi-objective multi-step variable time-domain prediction is provided, comprising: a parameter library acquisition stage: S11: designing a tracking error system for the intelligent agent motion control system, and constructing a cost function set for motion control based on different curvatures and control cycles; S12: offline training of the intelligent agent motion control system using a nonlinear model predictive controller with the goal of minimizing the cost function, to obtain an evaluation index of the multi-objective optimization algorithm; S13: setting constraints for the multi-objective optimization algorithm; S14: under the constraints, solving the evaluation index using the multi-objective optimization algorithm with the goal of minimizing the evaluation index, obtaining a Pareto set, and classifying the Pareto set to obtain a velocity parameter library and a prediction time-domain parameter library; a stability domain acquisition stage: S21: defining the... The terminal cost function of the nonlinear model predictive controller is used to obtain the terminal domain and terminal controller; S22: The terminal domain is narrowed according to the tracking error system to obtain the stable domain; Adjustment stage: S31: It is determined whether the end state of the agent is in the stable domain. If it is in the stable domain, the prediction time domain is reduced and the agent speed is increased; if it is not in the stable domain, the corresponding speed parameter library and prediction time domain parameter library are selected according to the current trajectory curvature and control cycle; S32: The weights of the multi-objective optimization evaluation index are allocated, and multi-objective optimization is performed to obtain the optimal target speed and target prediction time domain; S33: The target speed and target prediction time domain are introduced into the nonlinear model predictive controller to obtain the target control sequence and target state sequence of the agent motion control system. The agent is controlled by the target control sequence to obtain the end state of the agent. The loop step S31 is repeated to switch states.

[0006] Preferably, step S11 further includes presetting the desired trajectory of the intelligent agent, adjusting the curvature and control period of the desired trajectory, and obtaining the cost function of each trajectory; step S12 specifically involves using a nonlinear model predictive controller to perform offline training on the intelligent agent motion control system with the goal of minimizing the cost function, and obtaining the evaluation index of the multi-objective optimization algorithm.

[0007] Preferably, the cost function J(u,q0) is:

[0008]

[0009] Where u represents the control quantity sequence, u = {u e (0,q0),u e (1,q0),u e (2,q0)...u e (N-1, q0)}, where q0 represents the system state to be optimized, i∈[0,N-1] represents different times in the prediction time domain, and qe (i,q0) represents the error between the optimized trajectory and the reference trajectory at time i in the prediction time domain, and q e (0,q0)=q0,u e (i,q0) represents the control increment required at time i in the prediction time domain, W(q e (i,q e ),u e (i,q e W(q) is the stage indicator function. e ,u e )=q e T Qq e +u e T Ru e N is the prediction time domain, and F(·) is the terminal cost function, satisfying Q, R, and P are positive definite matrices of corresponding dimensions.

[0010] Preferably, obtaining the evaluation index of the multi-objective optimization algorithm in step S12 specifically involves evaluating the motion parameters during offline training to obtain the evaluation index of the multi-objective optimization algorithm.

[0011] Preferably, the evaluation index P is:

[0012] Π={f|f=[f1,f2,f3] T <F max}

[0013] in:

[0014]

[0015] f1 is the time-weighted integral absolute error index, f2 is the settling time index, f3 is the computation and solution time index, and t s Let e(t) be the time required for the output step response to enter the error band, e(t) be the error value at time t calculated offline, Δζ be the average time required for each optimization solution of the nonlinear model predictive controller, and F be the time required for the output step response to enter the error band. max The maximum value of the column vector consisting of the absolute error of the time-weighted integral error, the settling time, and the calculation and solution time is Δ, which is the preset error threshold.

[0016] Preferably, step S13 specifically involves setting constraints for a multi-objective optimization algorithm based on the control requirements of the intelligent agent motion control system.

[0017] Preferably, the constraint condition Ψ is:

[0018] Ψ={v,N|v min <v<v max (ρ),Nmin <N<N max (T)}

[0019] Where v is the speed of the agent, v min v is the minimum velocity of motion. max (ρ) represents the maximum velocity affected by the change in curvature ρ, N min Let N be the minimum value in the prediction time domain, and N be the prediction time domain. max (T) represents the predicted time-domain maximum value affected by the change in control period T. Preferably, step S21 further includes establishing a discrete polyhedral model of the agent tracking error system, using linear matrix inequalities to prove that the terminal cost function of the agent motion control system satisfies the stability condition, and satisfies the control constraint condition and state constraint condition in the terminal domain.

[0020] Preferably, in step S32, the multi-objective optimization evaluation index is weighted, specifically as follows:

[0021] Fuzzy logic weight allocation is performed on the multi-objective optimization evaluation indicators. The weight values ​​G of the multi-objective optimization evaluation indicators after allocation are:

[0022] G=σ1f1+σ2f2+σ3f3

[0023] Where (σ1,σ2,σ3) are weighting coefficients, f1 is the time-weighted integral absolute error index, f2 is the adjustment time index, and f3 is the calculation and solution time index.

[0024] Preferably, in step S33, the agent is controlled by the target control sequence to obtain the end state of the agent. Specifically, the rolling optimization principle is adopted, and the control parameters of the first preset number of items in the target control sequence are used as the control input of the agent in sequence. After the input is completed, the end state of the agent is obtained.

[0025] In summary, compared with the prior art, the intelligent agent motion control method based on multi-objective multi-step time-domain prediction provided by the present invention has the following beneficial effects:

[0026] 1. This application aims to minimize the cost function by conducting offline training on an intelligent agent motion control system to obtain evaluation indicators for multi-objective optimization algorithms. Combined with constraints, it obtains a velocity parameter library and a prediction time-domain parameter library, and further restricts the terminal domain to obtain a stable domain. Based on the stable domain, it judges the stability of the agent's end state and adjusts it to ensure that the agent works within the stable domain, thus guaranteeing stability under multi-objective operation and timeliness of changes over time.

[0027] 2. For multi-objective optimization algorithms, this method quantifies the control accuracy, stability, and real-time performance of the motion control system as multi-objective optimization evaluation indicators. Constraints are imposed on the evaluation indicators and independent variables according to the motion control requirements to accurately construct a parameter library for online adjustment.

[0028] 3. The designed switching logic rules can achieve adaptive adjustment of controller parameters, which ensures improved real-time performance and motion efficiency in a stable state of motion controller, and allows for the switching of appropriate motion speed and prediction time domain in an unstable state to ensure the agent tends to stabilize. At the same time, by using multiple terms of the control sequence solved by the controller as control inputs for agent motion control, the solution complexity caused by using only the first term at each time step is reduced. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the steps of the intelligent agent motion control method based on multi-objective multi-step variable time-domain prediction proposed in this application.

[0030] Figure 2 This is a flowchart of an intelligent agent motion control method based on multi-objective, multi-step, time-domain prediction;

[0031] Figure 3 This is a flowchart illustrating the construction of the velocity and prediction time-domain parameter library and the solution of optimal control parameters in an embodiment of this application.

[0032] Figure 4 This is a flowchart of parameter adjustment based on switching logic rules in an embodiment of this application. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0034] This invention provides an intelligent agent motion control method based on multi-objective, multi-step time-domain prediction, including a parameter library acquisition stage, a stability domain acquisition stage, and an adjustment stage, such as... Figure 1 and Figure 2 As shown.

[0035] The parameter library acquisition stage includes S11 to S14, and the specific steps are as follows.

[0036] S11: Design a tracking error system for an intelligent agent motion control system, and construct a cost function set for motion control based on different curvatures and control cycles.

[0037] Based on the intelligent agent motion control system, a tracking error system is designed to represent the error between the actual trajectory and the reference trajectory. The tracking error system model is discretized to obtain:

[0038] q e (k+1)=f(q e (k),u e (k))

[0039] Where, q e (k)∈R n Let u be the error between the current trajectory and the reference trajectory of the system at time k. e (k)∈R m Let f(·,·) be the control increment of the system at time k, and let f(·,·) be the control increment with respect to q. e (k) and u e A continuous function of (k) that simultaneously satisfies f(0,0)=0.

[0040] Based on the tracking error system, let T be the period, and define the time series as a given cost function J(u,q0):

[0041]

[0042] Where u represents the control quantity sequence, u = {u e (0,q0),u e (1,q0),u e (2,q0)...u e (N-1, q0)}, where q0 represents the system state to be optimized, i∈[0,N-1] represents different times in the prediction time domain, and q e (i,q0) represents the error between the optimized trajectory and the reference trajectory at time i in the prediction time domain, and q e (0,q0)=q0,u e (i,q0) represents the control increment required at time i in the prediction time domain, W(q e (i,q e ),u e (i,q e W(q) is the stage indicator function. e ,u e )=q e T Qq e +u e T Ru e N is the prediction time domain, and F(·) is the terminal cost function, satisfying Q, R, and P are positive definite matrices of corresponding dimensions.

[0043] In a further preferred embodiment, the desired trajectory of the intelligent agent is preset, for example, a trigonometric function curve is used as the desired trajectory, the curvature and control period of the desired trajectory are adjusted, and the cost function of each trajectory is obtained to form a cost function set S.

[0044]

[0045] Where ρ is the trajectory curvature, n is the number of elements in the cost function set, and J i For different cost functions.

[0046] S12: With the goal of minimizing the cost function, the nonlinear model predictive controller is used to perform offline training on the intelligent agent motion control system to obtain the evaluation index of the multi-objective optimization algorithm.

[0047] Nonlinear model predictive controllers are existing technologies and will not be elaborated here. In a further preferred embodiment, step S12 specifically involves using a nonlinear model predictive controller to perform offline training on the intelligent agent motion control system with the goal of minimizing the cost function, thereby obtaining the evaluation index Π of the multi-objective optimization algorithm.

[0048] The evaluation index Π is:

[0049] Π={f|f=[f1,f2,f3] T <F max}

[0050] in:

[0051]

[0052] f1 is the time-weighted integral absolute error index, f2 is the settling time index, f3 is the computation and solution time index, and t s The time required for the output step response to enter the error band (|e(t)|≤Δ) is given, where e(t) is the error value at time t calculated offline, Δζ is the average time required for each optimization solution of the nonlinear model predictive controller, and F is the time required for the step response to enter the error band (|e(t)|≤Δ). max This represents the maximum value of the column vector formed by the time-weighted integral error absolute error, the settling time, and the computation time. Preferably, in this application, Δ = 0.02.

[0053] S13: Set the constraints for the multi-objective optimization algorithm.

[0054] Constraints for a multi-objective optimization algorithm are set based on the control requirements of an intelligent agent motion control system.

[0055] The constraint condition Ψ is:

[0056] Ψ={v,N|v min <v<vmax (ρ),N min <N<N max (T)}

[0057] Where v is the speed of the agent, v min v is the minimum velocity of motion. max (ρ) represents the maximum velocity affected by the change in curvature ρ, N min Let N be the minimum value in the prediction time domain, and N be the prediction time domain. max (T) represents the maximum value in the prediction time domain affected by the change in the control period T.

[0058] S14: Under the constraints, the multi-objective optimization algorithm is used to solve for the evaluation index with the goal of minimizing the evaluation index, obtaining the Pareto set, and classifying the Pareto set to obtain a velocity parameter library and a prediction time-domain parameter library. For example... Figure 3 As shown, specifically, based on different curvatures and periods, corresponding velocity parameter libraries and prediction time-domain parameter libraries are obtained.

[0059] Stability domain acquisition phase:

[0060] S21: Define the terminal cost function of the nonlinear model predictive controller, and obtain the terminal domain and terminal controller.

[0061] Suppose there exists a positive definite symmetric matrix X∈R 3×3 and Y∈R 2×3 And the coefficient α,K>0.

[0062] Define the terminal cost function as follows:

[0063] F(q e (k))=q e (k) T Pq e (k)

[0064] The terminal domain Ω is:

[0065] Ω={q e ∈R n |q e (k) T Pq e (k)≤α}

[0066] The terminal controller is:

[0067] κ(q e (k))=Kq e (k)

[0068] Establish a discrete polyhedral model for the intelligent agent tracking error system:

[0069]

[0070] in, It is the tracking error system with respect to q e (k) and u e (k) Polyhedral set.

[0071] The linear matrix inequality is used to prove that the terminal cost function of the intelligent agent motion control system satisfies the stability condition, and also satisfies the control constraints and state constraints in the terminal domain:

[0072]

[0073]

[0074] Where, q e,max u represents the limit value of the system state. e,max This represents the limit value of the system control quantity.

[0075] S22: Based on the tracking error system, the terminal domain is narrowed to obtain a stable domain.

[0076] Since the terminal domain is the neighborhood of the origin, and considering the characteristics of the error tracking system, its state needs to gradually approach the reference trajectory. Therefore, the smaller the error value, the closer its state is to the origin of the terminal domain. Based on this characteristic, the range of the terminal domain is defined and divided as follows: a scaling factor λ is introduced for the terminal domain, and the set of λΩ contained in the terminal domain is set as the stable region, while the rest are set as the unstable region.

[0077] Adjustment phase:

[0078] S31: Determine whether the end state of the agent is in the stable domain. If it is in the stable domain, reduce the prediction time domain and increase the agent's speed. If it is not in the stable domain, select the corresponding speed parameter library and prediction time domain parameter library according to the current trajectory curvature and control cycle.

[0079] When the terminal state q e If (N,q0) is within the set λΩ, then the actual trajectory is defined as being sufficiently close to the reference trajectory, resulting in stable trajectory tracking; when the terminal state q e If (N,q0) is within the set Ω but not within the set λΩ, then there is an error between the actual value and the expected value, resulting in unstable trajectory tracking. By dividing the terminal domain, the position of the terminal domain in which the system solves for the terminal state is located is used as the criterion for judging the unstable signal M, as shown in the following equation.

[0080]

[0081] By evaluating the controller's terminal state at time k, the unstable signal value is determined, and the parameters of the nonlinear model predictive controller at the next time step are adjusted.

[0082] The system state is used to determine whether the trajectory tracking effect is stable. If the trajectory tracking effect is stable (M=0), the time domain of the next moment is reduced and the agent speed is increased; if the trajectory tracking effect is unstable (M=1), the corresponding velocity parameter library and prediction time domain parameter library are selected according to the current curvature and the current control cycle.

[0083] The stability of the trajectory tracking effect is determined based on the system status, i.e., the state of the unstable signal M, such as... Figure 4 As shown. If M=0, the trajectory tracking effect is stable, and then the prediction time domain of the next nonlinear model predictive control is reduced, while the speed of the agent is improved. The specific expression is:

[0084]

[0085] Where v′ and N′ represent the motion velocity and prediction time domain of the improved next nonlinear model predictive control, and Round represents the rounding function.

[0086] S32: Assign weights to the multi-objective optimization evaluation indexes and perform multi-objective optimization to obtain the optimal target speed and target prediction time domain.

[0087] Furthermore, based on the current system's requirements for accuracy, stability, and computational complexity, fuzzy logic weights (σ1, σ2, σ3) are applied to the multi-objective optimization evaluation indices f1, f2, and f3 to obtain the optimal target velocity and target prediction time domain.

[0088]

[0089] Among them, v best For the optimal solution of speed, N best To predict the optimal solution in the time domain, we define the weight values ​​of the multi-objective optimization evaluation index.

[0090] The obtained Pareto set and parameter library are initially selected to obtain multiple optimal solutions under the balance of multiple indicators. Simultaneously, the error, error rate of change, and current controller solution speed are used as inputs, and the weights (σ1, σ2, σ3) of f1, f2, and f3 are used as outputs. A fuzzy feature set of {small, medium-small, medium, medium-large, large} is selected, and the preference weights (σ1, σ2, σ3) of the corresponding indicators are allocated through triangular membership functions. Finally, based on the weight values ​​of the multi-objective optimization evaluation indicators, the most suitable motion speed and prediction time domain are selected.

[0091] S33: Input the target velocity and target prediction time domain into the nonlinear model predictive controller to obtain the target control sequence and target state sequence of the intelligent agent motion control system. Use the target control sequence to control the intelligent agent and obtain the end state of the intelligent agent. Repeat step S31 to switch states.

[0092] The calculated optimal target velocity and target prediction time domain value are substituted into the nonlinear model predictive controller, and the minimum value of the optimization problem is solved to obtain the control sequence and state sequence of the system.

[0093] In a further preferred embodiment, the agent is controlled by the target control sequence to obtain the agent's end state, specifically as follows:

[0094] Using the rolling optimization principle, the target control sequence u is... * (q0)={u * (0,q0),u * (1,q0),…,u e The preset number of control parameters in (N-1,q0)} are used as the control inputs of the agent in sequence, and the end state of the agent is obtained after the input is completed.

[0095] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent agent motion control based on multi-objective, multi-step, time-domain variable prediction, characterized in that, include: Parameter library acquisition stage: S11: Design a tracking error system for an intelligent agent motion control system, and construct a cost function set for motion control based on different curvatures and control cycles; S12: With the goal of minimizing the cost function, the nonlinear model predictive controller is used to perform offline training on the intelligent agent motion control system to obtain the evaluation index of the multi-objective optimization algorithm. S13: Set the constraints for the multi-objective optimization algorithm; S14: Under the constraints, the multi-objective optimization algorithm is used to solve the evaluation index with the goal of minimizing the evaluation index, to obtain the Pareto set, and the Pareto set is classified to obtain the velocity parameter library and the prediction time domain parameter library. Stability domain acquisition phase: S21: Define the terminal cost function of the nonlinear model predictive controller, and obtain the terminal domain and terminal controller; S22: Based on the tracking error system, the terminal domain is narrowed to obtain a stable domain; Adjustment phase: S31: Determine whether the end state of the agent is in the stable domain. If it is in the stable domain, reduce the prediction time domain and increase the agent speed. If it is not in the stable domain, select the corresponding speed parameter library and prediction time domain parameter library according to the current trajectory curvature and control cycle. S32: Assign weights to the multi-objective optimization evaluation indexes and perform multi-objective optimization to obtain the optimal target velocity and target prediction time domain; S33: Input the target velocity and target prediction time domain into the nonlinear model predictive controller to obtain the target control sequence and target state sequence of the intelligent agent motion control system. Use the target control sequence to control the intelligent agent and obtain the end state of the intelligent agent. Repeat step S31 to switch states.

2. The intelligent agent motion control method according to claim 1, characterized in that, Step S11 further includes presetting the desired trajectory of the intelligent agent, adjusting the curvature and control period of the desired trajectory, and obtaining the cost function of each trajectory; Step S12 specifically involves using a nonlinear model predictive controller to perform offline training on the intelligent agent motion control system with the goal of minimizing the cost function, and obtaining the evaluation index of the multi-objective optimization algorithm.

3. The intelligent agent motion control method according to claim 1 or 2, characterized in that, The desired trajectory of the intelligent agent is preset, the curvature and control period of the desired trajectory are adjusted, and the cost function of each trajectory is obtained, forming a cost function set S: Where ρ is the trajectory curvature, n is the number of elements in the cost function set, and J i For different cost functions.

4. The intelligent agent motion control method according to any one of claims 1 to 3, characterized in that, The evaluation metrics for the multi-objective optimization algorithm obtained in step S12 are as follows: Evaluation metrics for multi-objective optimization algorithms are obtained by assessing motion parameters during offline training.

5. The intelligent agent motion control method according to claim 1 or 4, characterized in that, The evaluation index P is: P={f|f=[f1,f2,f3] T <F max } in: f1 is the time-weighted integral absolute error index, f2 is the settling time index, f3 is the computation and solution time index, and t s The time required for the output step response to enter the error band is given by the error range, where e(t) is the error value at time t calculated offline, Δζ is the average time required for each optimization solution of the nonlinear model predictive controller, and F is the error range. max D is the maximum value of the column vector consisting of the absolute error of the time-weighted integral error, the adjustment time, and the calculation and solution time, and D is the preset error threshold.

6. The intelligent agent motion control method according to claim 1, characterized in that, Step S13 specifically involves setting constraints for a multi-objective optimization algorithm based on the control requirements of the intelligent agent motion control system.

7. The intelligent agent motion control method according to claim 6, characterized in that, The constraint condition Y is: Y={v,N|v min <v<v max (ρ),N min <N<N max (T)} Where v is the speed of the agent, v min v is the minimum velocity of motion. max (ρ) represents the maximum velocity affected by the change in curvature ρ, N min Let N be the minimum value in the prediction time domain, and N be the prediction time domain. max (T) represents the maximum value in the prediction time domain affected by the change in the control period T.

8. The intelligent agent motion control method according to claim 1, characterized in that, Step S21 further includes establishing a discrete polyhedral model of the intelligent agent tracking error system, using linear matrix inequalities to prove that the terminal cost function of the intelligent agent motion control system satisfies the stability condition, and satisfies the control constraint condition and state constraint condition in the terminal domain; more preferably, step S22 specifically involves limiting the terminal domain based on the tracking error system to obtain the stable domain by setting a scaling factor λ, then the stable domain is λW, where W is the terminal domain.

9. The intelligent agent motion control method according to claim 1, characterized in that, In step S32, the weights of the multi-objective optimization evaluation indicators are assigned, specifically as follows: Fuzzy logic weight allocation is performed on the multi-objective optimization evaluation indicators. The weight values ​​G of the multi-objective optimization evaluation indicators after allocation are: G=σ1f1+σ2f2+σ3f3 Where (σ1,σ2,σ3) are weighting coefficients, f1 is the time-weighted integral absolute error index, f2 is the adjustment time index, and f3 is the calculation and solution time index.

10. The intelligent agent motion control method according to claim 1, characterized in that, In step S33, the agent is controlled using the target control sequence to obtain the agent's terminal state, specifically as follows: Using the principle of rolling optimization, the control parameters of the first preset number of items in the target control sequence are sequentially used as the control input of the agent, and the end state of the agent is obtained after the input is completed.

Images