A cascaded large model robot motion planning method based on timing logic constraints

Abstract

The application discloses a cascaded large model robot motion planning method based on timing logic constraints, which comprises the following steps: a task planner receives user instructions and corresponding STL expressions, environment information, decomposes the user instructions and corresponding STL expressions, and obtains sub-tasks and corresponding STL expressions; whether the decomposed sub-tasks meet the logic constraints, time constraints and space constraints is detected by an STL checker; if yes, correct controllers are generated according to the sub-tasks; the generated correct controllers are integrated with a robot model, a Monte Carlo method is used to generate multiple groups of random parameters based on uncertainty factors, real environment disturbances are simulated, an Euler method is used to discretize and iteratively solve trajectories of the robot continuous kinematics equation, the trajectories are checked by the STL checker to make the trajectories meet the requirements of the user instructions, and finally correct trajectories are obtained. The application realizes user instructions containing space, time and kinematics constraints.

Description

Technical Field

[0001] This application relates to the technical field of robot task and motion planning, and in particular to a motion planning method for cascaded large model robots based on temporal logic constraints. Background Technology

[0002] In the field of task and motion planning, existing LLM (Large Language Model)-based frameworks cannot effectively solve planning tasks with complex time and space constraints. When faced with temporal and spatial constraints, the output sub-objectives of high-level planning are not necessarily optimal, and may not even be feasible for low-level planning. Especially for time-related multi-step action sequences, optimizing both task logic and motion planning becomes crucial. Summary of the Invention

[0003] In view of this, this application provides a motion planning method for cascaded large-scale robot based on temporal logic constraints, which includes:

[0004] The task planner receives user instructions and their corresponding STL representations, along with environmental information. It decomposes the user instructions and their corresponding STL representations into subtasks and their corresponding STL representations. An STL checker verifies whether the decomposed subtasks satisfy logical, temporal, and spatial constraints. If they do, a correct controller is generated based on the subtasks. The user instructions instruct the robot to execute the specified task according to the STL representation within a specified time frame, within the environmental information provided. The environmental information includes the environment in which the robot is performing the specified task and the roads within that environment. The STL representation is the machine representation of the user instructions.

[0005] The generated correct controller is integrated with the robot model. The Monte Carlo method is used to generate multiple sets of random parameters based on uncertainties to simulate real environmental disturbances. The Euler method is used to discretize and iteratively solve the robot's continuous kinematic equations to obtain the trajectory. The trajectory is checked by the STL checker to ensure that it meets the user's instructions and does not violate time and space constraints, thus obtaining the final correct trajectory.

[0006] Furthermore, user instructions contain logical descriptions; sequential logic is signal-timing logic (STL), and a corresponding STL expression is obtained based on the user instructions. The STL expression has logical constraints; the logical constraints enable the robot to execute tasks according to the user instructions.

[0007] The time constraint is that the user instruction includes a time description;

[0008] The spatial constraint is that the user instruction contains a spatial location;

[0009] The kinematic constraint is that the final trajectory that satisfies the user's instructions must satisfy the kinematic model.

[0010] Furthermore, generating the correct controller based on the subtask includes:

[0011] The algorithm selector selects the appropriate algorithm from the knowledge base to implement each subtask based on the subtask and its corresponding STL expression; the knowledge base includes multiple algorithms to implement each subtask.

[0012] The motion planner generates a controller, or executable Python code, based on the selected algorithm.

[0013] The controller undergoes a compilation check. After the Python interpreter performs the compilation check, a functional check is performed. If the functional check passes, the correct controller is obtained.

[0014] Furthermore, the compilation check of the controller includes:

[0015] The controller is compiled. If no errors are found, the Python interpreter is used to check the compilation process. Otherwise, it indicates that there are related errors in the executable Python code, which need to be regenerated. Related errors include syntax errors.

[0016] Furthermore, a functional check of the controller is performed, including:

[0017] The controller tests each case in the case library in turn, and the STL checker verifies the test results. If each case in the case library passes the verification, it means that the controller has passed the functional check; otherwise, it means that the controller has a functional error, and the motion planner regenerates the controller according to the selected algorithm. Cases include edge cases.

[0018] Furthermore, the STL checker verifies the test results, including:

[0019] The STL checker calculates a robustness score by comparing the test results with the corresponding STL expression.

[0020] If the STL checker detects a non-negative robustness score, the validation passes; otherwise, the validation fails.

[0021] Furthermore, the STL checker calculates a robustness score by comparing the test results with the corresponding STL expression, including:

[0022] Given an STL formula ,Signal and time Quantitative semantics Calculate recursively according to the following rules:

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032] in, To express "True", the predicate form is: , yes arrive The function, by state It takes as input and returns a scalar value; The quantitative robustness function represents the signal. At the point of time For the formula The degree of satisfaction is indicated by the return value, which is a real number. Time series function s : T →R n , for The state vector at any given time; For the time domain T Reference time in t ∈ T , T ⊆R≥0; Formulas that are logically true and always hold true The robustness is always 1; For predicate functions, mapping functions μ : → Input status Output scalar values; atomic predicate form is: ; For predicates in The value at time is used to directly calculate the signal state. pass The result of the function; For logical NOT, take the negative value of the robustness of the subformulation; For logical AND, take the minimum robustness of the two sub-formulas; The result is a logical OR operation, taking the maximum robustness of the two sub-formulas. For logical implication, it is equivalent to Is it false or True; For the bounded Until operator, exist[ t + , t + This is the first time it has been true, and previously... Continuing to be true; For the bounded Always operator, in [ t + , t + ]Memory at a certain moment True; For the bounded Always operator, in [ t + , t + All moments within ] All are true; and Define the relative time interval as the time limit parameter. t + , t + The offset of ]; and To help with time variables, Traversal [ t + , t + ], Traversal [ t , ], used for U Operator; For supremum, take the least upper bound of the set; To determine the infimum, take the largest lower bound of the set; It is a minimum value function. It is a function for maximizing the value.

[0033] Furthermore, the step of checking the trajectory using an STL checker to ensure that the trajectory conforms to the user instruction requirements includes:

[0034] The trajectory is compared with the STL expression output by the STL checker. If they match, the trajectory meets the user instruction requirements; otherwise, the trajectory does not meet the user instruction requirements.

[0035] Furthermore, the task planner, the algorithm selector, and the motion planner are Large Language Models (LLMs).

[0036] Furthermore, it also includes:

[0037] If a subtask does not meet the logical, temporal, and spatial constraints, the user instructions and their corresponding STL expressions are re-decomposed by the task planner.

[0038] Due to the adoption of the above technical solution, this application has the following advantages: This solution can implement user instructions that include spatial, temporal, and kinematic constraints. It satisfies the constraints contained in the instructions in a hierarchical manner and designs a feedback mechanism based on STL (SignalTemporal Logic) to effectively verify the planning results and provide error feedback. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0040] Figure 1 This is a schematic diagram of a motion planning method for a cascaded large-scale robot based on temporal logic constraints, according to an embodiment of this application. Detailed Implementation

[0041] The present application will be further described in conjunction with the accompanying drawings and embodiments. The described embodiments are only some, not all, of the embodiments of the present application. All other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of the present application.

[0042] See Figure 1 This application provides an embodiment of a motion planning method for a cascaded large-scale robot based on temporal logic constraints, which includes:

[0043] The task planner receives user instructions and their corresponding STL representations, along with environmental information. It decomposes the user instructions and their corresponding STL representations into subtasks and their corresponding STL representations. An STL checker verifies whether the decomposed subtasks satisfy logical, temporal, and spatial constraints. If they do, a correct controller is generated based on the subtasks. The user instructions instruct the robot to execute the specified task according to the STL representation within a specified time frame, within the environmental information provided. The environmental information includes the environment in which the robot is performing the specified task and the roads within that environment. The STL representation is the machine representation of the user instructions.

[0044] The generated correct controller is integrated with the robot model. The Monte Carlo method is used to generate multiple sets of random parameters based on uncertainties to simulate real environmental disturbances. The Euler method is used to discretize and iteratively solve the robot's continuous kinematic equations to obtain the trajectory. The trajectory is checked by the STL checker to ensure that it meets the user's instructions and does not violate time and space constraints, thus obtaining the final correct trajectory.

[0045] Optionally, the user instructions include a logical description; the timing logic is signal-timing logic (STL), and the corresponding STL expression is obtained according to the user instructions. The STL expression has logical constraints; the logical constraints are the robot's ability to execute tasks according to the user instructions.

[0046] The time constraint is that the user instruction includes a time description;

[0047] The spatial constraint is that the user instruction contains a spatial location;

[0048] The kinematic constraint is that the final trajectory that satisfies the user's instructions must satisfy the kinematic model.

[0049] Optionally, generating the correct controller based on the subtask includes:

[0050] The algorithm selector selects the appropriate algorithm from the knowledge base to implement each subtask based on the subtask and its corresponding STL expression; the knowledge base includes multiple algorithms to implement each subtask.

[0051] The motion planner generates a controller, or executable Python code, based on the selected algorithm.

[0052] The controller undergoes a compilation check. After the Python interpreter performs the compilation check, a functional check is performed. If the functional check passes, the correct controller is obtained.

[0053] Optionally, the compilation check of the controller includes:

[0054] The controller is compiled. If no errors are found, the Python interpreter is used to check the compilation process. Otherwise, it indicates that there are related errors in the executable Python code, which need to be regenerated. Related errors include syntax errors.

[0055] Optionally, a functional check of the controller may be performed, including:

[0056] The controller tests each case in the case library in turn, and the STL checker verifies the test results. If each case in the case library passes the verification, it means that the controller has passed the functional check; otherwise, it means that the controller has a functional error, and the motion planner regenerates the controller according to the selected algorithm. Cases include edge cases.

[0057] Optionally, the STL checker verifies the test results, including:

[0058] The STL checker calculates a robustness score by comparing the test results with the corresponding STL expression.

[0059] If the STL checker detects a non-negative robustness score, the validation passes; otherwise, the validation fails.

[0060] Optionally, the STL checker calculates a robustness score between the test results and the corresponding STL expression, including:

[0061] Given an STL formula ,Signal and time Quantitative semantics Calculate recursively according to the following rules:

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071] in, To express "True", the predicate form is: , yes arrive The function, by state It takes as input and returns a scalar value; The quantitative robustness function represents the signal. At the point of time For the formula The degree of satisfaction is indicated by the return value, which is a real number. Time series function s : T →R n , for The state vector at any given time; For the time domain T Reference time in t ∈ T , T ⊆R≥0; Formulas that are logically true and always hold true The robustness is always 1; For predicate functions, mapping functions μ : → Input status Output scalar values; atomic predicate form is: ; For predicates in The value at time is used to directly calculate the signal state. pass The result of the function; For logical NOT, take the negative value of the robustness of the subformulation; For logical AND, take the minimum robustness of the two sub-formulas; The result is a logical OR operation, taking the maximum robustness of the two sub-formulas. For logical implication, it is equivalent to Is it false or True; For the bounded Until operator, exist[ t + , t + This is the first time it has been true, and previously... Continuing to be true; For the bounded Always operator, in [ t + , t + ]Memory at a certain moment True; For the bounded Always operator, in [ t + , t + All moments within ] All are true; and Define the relative time interval as the time limit parameter. t + , t + The offset of ]; and To help with time variables, Traversal [ t + , t + ], Traversal [ t , ], used for U Operator; For supremum, take the least upper bound of the set; To determine the infimum, take the largest lower bound of the set; It is a minimum value function. It is a function for maximizing the value.

[0072] Optionally, the step of checking the trajectory using an STL checker to ensure that the trajectory conforms to user instructions includes:

[0073] The trajectory is compared with the STL expression output by the STL checker. If they match, the trajectory meets the user instruction requirements; otherwise, the trajectory does not meet the user instruction requirements.

[0074] Optionally, the task planner, the algorithm selector, and the motion planner are Large Language Models (LLMs).

[0075] Optionally, it also includes:

[0076] If a subtask does not meet the logical, temporal, and spatial constraints, the user instructions and their corresponding STL expressions are re-decomposed by the task planner.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of the claims of this application.

Claims

1. A motion planning method for a cascaded large-scale robot based on temporal logic constraints, characterized by: The task planner receives user instructions and their corresponding STL expressions, as well as environment information. It decomposes the user instructions and their corresponding STL expressions to obtain subtasks and their corresponding STL expressions. The STL checker checks whether the decomposed subtasks meet the logical, temporal, and spatial constraints. If they do, the correct controller is generated based on the subtasks. User instructions are used to instruct the robot to perform the specified task according to the STL expression within the specified time. The environmental information includes the environment in which the robot is performing the specified task and the roads in the environment. The STL expression is the machine expression of the user instructions. The generated correct controller is integrated with the robot model. The Monte Carlo method is used to generate multiple sets of random parameters based on uncertainties to simulate real environmental disturbances. The Euler method is used to discretize and iteratively solve the robot's continuous kinematic equations to obtain the trajectory. The trajectory is checked by the STL checker to ensure that it meets the user's instructions and does not violate time and space constraints, thus obtaining the final correct trajectory.

2. The method according to claim 1, characterized in that, User instructions contain logical descriptions; sequential logic is signal-time logic (STL), and the corresponding STL expression is obtained according to the user instructions. The STL expression has logical constraints; the logical constraints enable the robot to execute tasks according to the user instructions. The time constraint is that the user instruction includes a time description; The spatial constraint is that the user instruction contains a spatial location; The kinematic constraint is that the final trajectory that satisfies the user's instructions must satisfy the kinematic model.

3. The method according to claim 1, characterized in that, The step of generating the correct controller based on the subtasks specifically includes: The algorithm selector selects the appropriate algorithm from the knowledge base to implement each subtask based on the subtask and its corresponding STL expression; the knowledge base includes multiple algorithms to implement each subtask. The motion planner generates a controller, or executable Python code, based on the selected algorithm. The controller undergoes a compilation check. After the Python interpreter performs the compilation check, a functional check is performed. If the functional check passes, the correct controller is obtained.

4. The method according to claim 3, characterized in that, The compilation check of the controller includes: The controller is compiled. If the Python interpreter does not report any errors, the compilation check passes. Otherwise, it means that there are related errors in the executable Python code, which need to be regenerated. Related errors include syntax errors.

5. The method according to claim 3, characterized in that, Perform a functional check on the controller, including: The controller tests each case in the case library in turn, and the STL checker verifies the test results. If each case in the case library passes the verification, it means that the controller has passed the functional check; otherwise, it means that the controller has a functional error, and the motion planner regenerates the controller according to the selected algorithm. Cases include edge cases.

6. The method according to claim 5, characterized in that, The STL checker verifies the test results, including: The STL checker calculates a robustness score by comparing the test results with the corresponding STL expression. If the STL checker detects a non-negative robustness score, the validation passes; otherwise, the validation fails.

7. The method according to claim 6, characterized in that, The STL checker calculates a robustness score by comparing the test results with the corresponding STL expression, including: Given an STL formula ,Signal and time Quantitative semantics Calculate recursively according to the following rules: in, To express "True", the predicate form is: , yes arrive The function, by state It takes as input and returns a scalar value; The quantitative robustness function represents the signal. At the point of time For the formula The degree of satisfaction is indicated by the return value, which is a real number. Time series function s : T →R n , for The state vector at any given time; For the time domain T Reference time in t ∈ T , T R≥0; Formulas that are always true for logical truth values The robustness is always 1; For predicate functions, mapping functions μ : → Input status Output scalar values; atomic predicate form is: ; For predicates in The value at time is used to directly calculate the signal state. pass The result of the function; For logical NOT, take the negative value of the robustness of the subformulation; For logical AND, take the minimum robustness of the two sub-formulas; The result is a logical OR operation, taking the maximum robustness of the two sub-formulas. For logical implication, it is equivalent to Is it false or True; For the bounded Until operator, exist[ t + , t + This is the first time it has been true, and previously... Continuing to be true; For the bounded Always operator, in [ t + , t + ]Memory at a certain moment True; For the bounded Always operator, in [ t + , t + All moments within ] All are true; and Define the relative time interval as the time limit parameter. t + , t + The offset of ]; and To help with time variables, Traversal [ t + , t + ], Traversal [ t , ], used for U Operator; For supremum, take the least upper bound of the set; To determine the infimum, take the largest lower bound of the set; It is a minimum value function. It is a function for maximizing the value.

8. The method according to claim 6, characterized in that, The step of checking the trajectory using an STL checker to ensure that the trajectory conforms to the user's instructions includes: The trajectory is compared with the STL expression output by the STL checker. If they match, the trajectory meets the user instruction requirements; otherwise, the trajectory does not meet the user instruction requirements.

9. The method according to claim 3, characterized in that, The task planner, the algorithm selector, and the motion planner are Large Language Models (LLMs).

10. The method according to claim 1, characterized in that, Also includes: If a subtask does not meet the logical, temporal, and spatial constraints, the user instructions and their corresponding STL expressions are re-decomposed by the task planner.

Images