A mechanical arm double-layer collaborative trajectory planning method under a confined space man-machine co-domain

Abstract

The application discloses a mechanical arm double-layer collaborative trajectory planning method under a man-machine co-domain in a confined space, and first, establishes a work environment model according to mechanical arm structure parameters, end tool parameters, target pose, a human body geometric model, device space boundaries and cable constraint information; second, establishes a mechanical arm forward and inverse kinematics model, and introduces an end redundancy angle to solve a joint configuration set corresponding to a candidate base-target point combination; then, constructs an end stiffness prediction model fused with physical priori; next, constructs a base optimal pose planning model with the upper layer base pose and the end redundancy bias angle as optimization variables; finally, collaboratively optimizes the mechanical arm speed, acceleration, motion efficiency, interactive compliance and joint torque burden, and outputs the optimal base pose and the mechanical arm execution trajectory. The application improves the safety, stability and execution efficiency in the man-machine co-domain operation in the confined space.

Description

Technical Field

[0001] This invention relates to the field of optimization of transcranial magnetic stimulation (TMS) treatment systems for medical robots, specifically to a two-layer collaborative trajectory planning method for robotic arms in a confined space human-machine co-domain environment. Background Technology

[0002] With the development of medical robots, neuromodulation equipment, and intelligent treatment systems, clinical treatment tasks place higher demands on the precise positioning, safe access, and stable operation capabilities of robotic end-effectors. These tasks are typically performed within a confined space where the patient, treatment chair, main unit, connecting cables, and auxiliary equipment coexist. The robot needs to achieve end-effector positioning, posture adjustment, and trajectory movement in a near-human environment. Therefore, trajectory planning for medical robot treatment systems not only needs to meet the end-effector pose accuracy requirements but also needs to comprehensively consider factors such as human safety distance, obstacle avoidance, cable constraints, posture accessibility, end-effector stiffness, motion smoothness, and execution efficiency.

[0003] In close-contact, contact-based, or directional therapeutic scenarios, significant coupling relationships exist between robot base position, robotic arm redundancy configuration, end-effector direction, and joint trajectories. Base layout affects target accessibility and joint posture distribution; redundancy configuration affects the end-effector's direction of action, obstacle avoidance margin, and stiffness performance; and trajectory parameters affect dynamic safety margin and motion stability. Existing trajectory planning methods are mostly designed for general working environments, typically handling base configuration, end-effector pose determination, and joint trajectory generation in separate stages, making it difficult to uniformly describe the coupling relationships between multiple variables. Furthermore, feasibility assessments based on static safety distances or discrete collision detection fail to fully reflect the dynamic safety changes during robot trajectory movement, easily leading to situations where local configurations are feasible but the overall trajectory safety margin is insufficient.

[0004] Furthermore, end-effector stiffness is a crucial indicator affecting the stability of therapeutic tools, closely related to robot joint configuration, base pose, end-effector application direction, joint stiffness parameters, and load transfer path. While physical calculations can be performed based on kinematic models, posture Jacobians, and stiffness transfer relationships, the stiffness performance of numerous candidate configurations needs to be repeatedly evaluated during bi-level programming and multi-objective optimization. Directly calling the physical stiffness model increases iterative computational overhead and reduces planning efficiency. Therefore, a sample set can be constructed using offline physical stiffness calculation results, and a neural network can be trained to establish a nonlinear mapping relationship between configuration parameters and end-effector stiffness indices, enabling rapid stiffness prediction and efficient candidate configuration selection during online planning.

[0005] Therefore, for human-robot co-operation in confined spaces in medical robot treatment systems, there is an urgent need to establish a two-layer collaborative trajectory planning method that integrates neural network stiffness prediction. This method would incorporate base pose optimization, end-effector configuration selection, rapid stiffness evaluation, and joint trajectory generation into a unified planning framework. While meeting the requirements for treatment target localization, it would also improve the robot's obstacle avoidance safety, physical constraint adaptability, end-effector operation stability, and trajectory execution reliability during motion, providing a safe, accurate, and stable motion planning solution for medical robot treatment systems, including robot-assisted TMS systems. Summary of the Invention

[0006] To address the needs of existing technologies, this invention provides a two-layer collaborative trajectory planning method for robotic arms in a confined space human-machine co-domain environment. The aim is to transform the decentralized processing of base posture optimization, end effector configuration selection, rapid stiffness evaluation, and joint trajectory generation into collaborative planning, providing technical support for the safe, accurate, and stable operation of medical robot treatment systems in complex clinical environments.

[0007] Step 1: Based on the parameters of the human-machine co-operation scenario in a confined space, establish a robotic arm operation environment model; the scenario parameters include the robotic arm structural parameters, joint range of motion, end effector size, treatment target pose, patient human geometric model, treatment seat model, treatment host position, connecting cable parameters, and spatial pose information of surrounding auxiliary equipment;

[0008] Step 2: Based on the structural diagram, establish the forward and inverse kinematics model of the robotic arm, and determine the end-effector pose mapping relationship and candidate configuration set;

[0009] Step 3: Considering that the end-effector pose is affected by joint flexibility, gravity and external load when the robotic arm is carrying the end-effector treatment tool to perform the task, an end-effector stiffness prediction model is established to quickly evaluate the end-effector stiffness performance under different candidate configurations.

[0010] Step 4: Construct the optimal pose planning model for the upper base, using the base spatial translation parameters and end-effector redundancy offset angle as optimization variables, and combining target reachability, operability, flexibility, end-effector stiffness, joint limit, cable constraint and obstacle avoidance verification conditions to screen feasible base poses and end-effector working configurations.

[0011] Step 5: The PIBF-MODE algorithm is proposed to solve the upper-level optimization model. Through parent initialization, dual-channel offspring generation, interference slip repair, low-fidelity pre-screening, high-fidelity evaluation, Pareto update and Nash decision, the optimal base configuration that satisfies the performance and safety constraints is obtained.

[0012] Step 6: Construct a lower-level robotic arm trajectory planning model. Based on the optimal base configuration, joint space geometric path, and dynamic safety envelope determined in the upper level, introduce pseudo-velocity variables to transform the trajectory time optimization into a state-dependent optimization problem. Then, perform constraint optimization on the robotic arm speed, acceleration, motion efficiency, interaction compliance, and joint torque burden to obtain a time-parameterized trajectory that meets the safety requirements of the confined space.

[0013] Step 7: Output the two-layer collaborative trajectory planning results of the robotic arm, perform final verification on the optimal base configuration of the upper layer and the motion trajectory of the lower layer, and output the optimal base pose and the execution trajectory of the robotic arm after meeting the requirements of target pose, joint limit, end effector stiffness, dynamic safety envelope, cable constraint and obstacle avoidance distance.

[0014] The beneficial effects of this invention are as follows: Addressing the problem of the coupling and difficulty in unified optimization of the robotic arm's base posture, end-effector configuration, end-effector stiffness, and motion trajectory in confined space human-robot co-domain treatment scenarios, a two-layer collaborative trajectory planning method integrating neural network stiffness prediction is constructed. By combining upper-layer base posture optimization with lower-layer time-parameterized trajectory optimization, collaborative planning is achieved for the arrangement of the gripping device base, selection of redundant end-effector postures, stiffness performance evaluation, and joint trajectory generation. The physically driven PIBF-MODE algorithm is used to perform Pareto pre-screening, interference slip repair, and high-fidelity comprehensive evaluation of candidate configurations, improving optimization efficiency and feasible solution quality under complex constraints. Simultaneously, the dynamic safety envelope is used to adaptively constrain the robotic arm's motion speed and acceleration, enabling the robotic arm to ensure obstacle avoidance safety for the human body, seat, and treatment equipment while also considering motion efficiency, interaction compliance, and end-effector stability. This provides an effective trajectory planning scheme for the safe, accurate, and stable operation of confined space medical robot treatment systems. Attached Figure Description

[0015] Figure 1 This is a flowchart of the present invention;

[0016] Figure 2 This is a schematic diagram of the clamping device in the TMS medical robot treatment system;

[0017] Figure 3 This is a neural network prediction block diagram for the end-effector stiffness of the clamping device in a TMS medical robot treatment system. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings. Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The directional terms such as left, center, right, top, and bottom in the embodiments of the present invention are only relative concepts or referenced to the normal use state of the product, and should not be considered restrictive.

[0019] A two-layer collaborative trajectory planning method for robotic arms in a confined space with human-machine co-domain environment, such as Figure 1 As shown, it includes the following steps:

[0020] Step 1: Based on the parameters of the human-machine co-operation scenario in a confined space, establish a robotic arm operation environment model; the scenario parameters include the robotic arm structural parameters, joint range of motion, end effector size, treatment target pose, patient human geometric model, treatment seat model, treatment host position, connecting cable parameters, and spatial pose information of surrounding auxiliary equipment;

[0021] Step 2: Taking the UR10 robotic arm as an example, based on the structural diagram, as follows... Figure 2 As shown, a forward and inverse kinematics model of the robotic arm is established to determine the end-effector pose mapping relationship and the candidate configuration set;

[0022] Considering the structural composition of the TMS robotic treatment system, the TMS treatment coil is usually held by a gripper installed at the end of a collaborative robot and moved to the vicinity of the human head. Ideally, it should be close to the skin of the head. Assuming that the coil is held perpendicular to the gripper, the motion parameters of the gripping device are shown in Table 1.

[0023] Table 1. Motion parameters of the clamping device

[0024] value / mm 127.3 176 612 127.8 572.3 115.7 115.7 92.2 128.5 80 2

[0025] l1-l8 are the kinematic structural parameters of the robot body, l9 is the gripper length, and l10 represents the structural offset distance between the gripper's point of action and the coil's stimulation center. This parameter is mainly determined by the coil's geometric dimensions and the gripping structure layout. Considering that the diameter of a single loop of a commonly used figure-eight coil is 70–100 mm, and the gripping position is usually set in the reinforced area of ​​the coil shell, l10 is taken as 80 mm to balance structural stability and avoidance of the magnetic field core area. l11 represents the gap between the coil shell and the scalp. Ideally, the coil should be tangent to the scalp (gap of 0 mm), but considering practical factors such as hair thickness, head curvature error, and positioning error, the gap is usually controlled within the range of 0–5 mm. To balance stimulation effect and practical feasibility, l11 is taken as 2 mm.

[0026] Step 2.1: Establish the forward kinematics model of the robotic arm based on screw theory;

[0027] Step 2.2: Considering the influence of the redundant angle of the TMS coil, establish the inverse kinematics model of the robotic arm, specifically as follows:

[0028] Considering the rotational invariance of the TMS treatment coil about its axis of symmetry, redundant angular parameters are introduced and an end-effector pose matrix is ​​established. As shown in Equation 1;

[0029]

[0030] in, Based on the rotation matrix, Let be the rotation transformation matrix about the axis containing the end point. The coordinates of the target point are given; then the Paden-Kahan subproblem method is used to solve the inverse kinematics equations analytically, thus completing the inverse kinematics modeling.

[0031] Step 3: Considering that the end-effector pose is affected by joint flexibility, gravity, and external load when the robotic arm is carrying the end-effector treatment tool to perform tasks, an end-effector stiffness prediction model is established to quickly evaluate the end-effector stiffness performance under different candidate configurations.

[0032] Step 3.1: Analyze the mapping relationship between base, target point, joint angle, and end effector stiffness to construct a neural network stiffness prediction training set, specifically as follows:

[0033] The permissible arrangement space of the robotic arm is traversed to generate different candidate base positions. Multiple target points and their corresponding normal vectors are randomly sampled within the permissible range of the human head target point. The candidate base positions and head target point positions are paired, and based on the kinematic model established in step 2, the joint angles of the robotic arm satisfying the target pose constraints under each base-target point combination are solved. For joint angle samples that satisfy joint limits and preliminary feasibility constraints, the corresponding end effector stiffness values ​​are calculated offline using the physical stiffness model as shown in Equation 2. , as training labels for neural networks;

[0034]

[0035] in, Here is the joint space inherent stiffness matrix. The geometric stiffness matrix is... The torque of the robot arm's own weight. Let Jacobian matrix be the current configuration of the robotic arm. The external force rotation is the weight of the coil itself.

[0036] Step 3.2: Construct a neural network stiffness prediction model based on the stiffness training set and incorporating physical priors, specifically as follows:

[0037] Based on the training samples obtained in step 3.1, the relative position vector from the clamping device base to the human target point is... The vector is composed of the six joint angles of the robotic arm obtained from inverse kinematics. Using the physical prior of basic stiffness as input to the neural network, and the total end-effector stiffness obtained offline as the network output label, an end-effector stiffness prediction neural network is constructed. The neural network block diagram is as follows: Figure 3 As shown; where, After sine-cosine encoding, the data is input into a residual multilayer perceptron to extract the robotic arm configuration features; the basic stiffness physical prior end-effector load force... and the mass of the robotic arm link Stiffness prior features are extracted by a multilayer perceptron; then the configuration features and stiffness prior features are concatenated and fused, and the end stiffness prediction results are output through a multilayer perceptron and a linear mapping layer. Finally, the output results are subjected to non-negativity constraints by the Softplus function.

[0038] Step 4: Construct the optimal pose planning model for the upper base, using the base spatial translation parameters and end-effector redundancy offset angle as optimization variables, and combining target reachability, operability, flexibility, end-effector stiffness, joint limit, cable constraint and obstacle avoidance verification conditions to screen feasible base poses and end-effector working configurations.

[0039] Step 4.1: Using the upper-level base pose and end-effector redundancy angle as optimization variables, establish the optimal pose search space for the clamping device base, specifically as follows:

[0040] After completing the modeling of the work scenario, the modeling of forward and inverse kinematics, and the construction of the end effector stiffness prediction model, the above models are used as the basis for calculation. The three-dimensional translation parameters of the clamping device base in the confined space and the redundant offset angle of the end effector around the action axis are used as the upper-level optimization variables. Since the position of the base directly affects the reachability of the robotic arm to the human target, the joint posture distribution, the obstacle avoidance margin, and the end effector stiffness performance, while the redundant offset angle affects the action direction of the end effector, the joint configuration selection, and the trajectory sweep range, it is necessary to perform collaborative optimization of the two at the upper level. The upper-level optimization variables are shown in Equation 3.

[0041]

[0042] in, These represent the spatial offset of the clamping device base relative to the reference coordinate system. This represents the redundant offset angle of the end-effector about its target direction;

[0043] Step 4.2: Construct the upper-level multi-objective optimization function, specifically as follows:

[0044] In the upper-level optimization process, it is necessary not only to determine whether the robotic arm can reach the target point, but also to evaluate whether the robotic arm has good operational performance under this configuration. Postural flexibility and end stiffness performance Therefore, the robotic arm's operability, dexterity, and end-effector static stiffness are taken as the upper-level optimization objectives, as shown in Equations 4-6 respectively:

[0045]

[0046]

[0047]

[0048] in, This is the set of angle values ​​for each joint in the robotic arm's working state; to ensure the principle of minimizing the joint sweep volume in a single operation, the gripping device uses a linear geometric path in joint space when moving from the initial state to the working state. The angles of each joint during the movement are calculated as shown in Equation 7.

[0049]

[0050] in, This indicates the angle values ​​for each joint. This represents the joint values ​​in the initial state. This represents the joint value when the target point is reached. Normalized geometric path parameters, , Represents the path evaluation node; in calculating the distance between the clamping device and the human body, seat, and TMS host. Simultaneously construct dynamic safety envelopes for velocity and acceleration decay. As shown in Equation 8;

[0051]

[0052] The upper-level path verification constraint is shown in Equation 9;

[0053]

[0054] Step 5: The PIBF-MODE algorithm is proposed to solve the upper-level optimization model. Through parent initialization, dual-channel offspring generation, interference slip repair, low-fidelity pre-screening, high-fidelity evaluation, Pareto update and Nash decision, the optimal base configuration that satisfies the performance and safety constraints is obtained.

[0055] Step 5.1: Generate the initial parent population, specifically:

[0056] Based on the human target location, the maximum physical reach of the clamping device, and the allowable placement area of ​​the base, several candidate base locations and end-effector redundant offset angles are randomly generated to form the initial parent population; the initial parent individuals are represented as... ,in, Candidate base locations, This is the end-redundancy offset angle. Representing different bases, The population size is defined as follows: each parent individual represents a possible base layout scheme and end-effector redundancy attitude scheme; after generating parent individuals, it is necessary to check whether they meet the base layout range, cable length, and initial obstacle avoidance constraints in turn; if not, resampling and generation are performed to ensure that all individuals in the initial population have basic physical feasibility; after initialization, the iteration generation is set to... and will As the first The parent generation participates in the generation of subsequent offspring;

[0057] Step 5.2: Generate candidate offspring individuals based on the current parent population, specifically as follows:

[0058] In the During the generational search process, the PIBF-MODE algorithm generates offspring individuals in the spatial location channel and the redundant spin channel, respectively. Spatial location channel Redundant spin channels are used to search for optimal base layout locations. The calculations are shown in Equations 10-11, which are used to search for a better end-point attitude redundancy angle.

[0059]

[0060]

[0061] in, The adaptive convergence factor. The minimum value of the adaptive convergence factor represents the minimum search step size allowed in the later stages of the algorithm iteration, used to enhance local convergence capability. The maximum value of the adaptive convergence factor represents the maximum allowed search step size in the early stages of the algorithm iteration, used to enhance global search capabilities. As the current globally optimal individual, and To randomly select except The difference perturbation brought about by two individuals outside the body The current parent spin angle, Refers to the Gaussian adjustment factor. The standard deviation parameter of the Gaussian perturbation represents the redundant spin angle. The scale of random perturbations during offspring generation;

[0062] Step 5.3: Conduct preliminary feasibility assessment and interference slip repair for candidate offspring individuals, specifically as follows:

[0063] For offspring candidate individuals The constraints of cable length, base obstacle avoidance, inverse kinematics solvability, joint limit constraint and path safety constraint are judged in sequence. If the candidate offspring directly meets the above constraints, the offspring is retained for subsequent evaluation. If the candidate offspring has local collision or insufficient safety distance, but is still within the repairable range, interference slip repair is performed, as shown in Equation 12-13.

[0064]

[0065]

[0066] in, Represents the number of times interference repairs are performed. , The step size is finely adjusted for the angle and the base position respectively; after each sliding repair, the inverse kinematics solution, robot geometry and obstacle avoidance distance of the candidate offspring are recalculated; if the safety distance constraint is met after repair, the offspring is retained; if the constraint is not met after the maximum number of repairs, the offspring is removed.

[0067] Step 5.4: Pareto-based multi-objective pre-screening and high-fidelity comprehensive scoring, specifically as follows:

[0068] A multi-objective pre-screening index is constructed for the retained candidate offspring set, using the minimum cumulative joint angle change and the maximum end effector stiffness as pre-screening objectives. The candidate offspring are then initially ranked in a non-dominated order to establish a local Pareto first front. The multi-objective pre-screening function... As shown in Equation 14;

[0069]

[0070] Since the local Pareto first front contains multiple non-dominant candidate offspring, and there is no absolute order of superiority within the front, a further crowding distance is introduced. As a secondary screening criterion, a unique deterministic trial gene is determined from it, and the calculation formulas are shown in Equations 15-16 respectively;

[0071]

[0072]

[0073] Deterministically test offspring Substitute the values ​​into the upper-level optimization model to calculate the high-fidelity objective function values ​​of its operability, flexibility, and end stiffness, and simultaneously perform upper-level constraint verification, as shown in Equation 17.

[0074]

[0075] in, This is a high-fidelity comprehensive score for deterministic genes, and this score is used for subsequent parent-offspring selection.

[0076] Step 5.5: Perform parent-child generation selection and update the parent population, specifically as follows:

[0077] To ensure that parental individuals and deterministic trial genes are compared on the same evaluation scale, the first generation will be... The generation individual parents Substituting the results into the high-fidelity comprehensive scoring function, we obtain the evaluation results of its upper-level objective function. and with deterministic probing genes High fidelity comprehensive score results A comparison is made between the two, examining their overall performance under the given objective function and constraints. It satisfies the upper-level constraints, and its overall performance is superior to that of other systems. Then by Alternative Otherwise, the original parent individual is retained; the parent update relationship is represented as shown in Equation 18;

[0078]

[0079] After completing the above updates for all parent individuals, the next generation of parent population is formed. Then ordered And continue to execute the processes of offspring generation, interference slip repair, Pareto pre-screening, high-fidelity comprehensive scoring and parent-child generation update until the maximum number of iterations is reached or the convergence condition is met;

[0080] Step 5.6: Determine the optimal base configuration based on the final Pareto front and Nash decision, specifically as follows:

[0081] After the upper-level iterations are completed, non-dominated solutions that satisfy the constraints are extracted from the final parent population to form the upper-level Pareto front set. Each candidate solution in this set corresponds to a feasible base pose and end-work configuration. To determine the unique optimal solution from multiple Pareto candidate solutions, a Nash game decision criterion is used for comprehensive evaluation to obtain the solution corresponding to the optimal base configuration. As shown in Equation 19;

[0082]

[0083] Step 6: Construct a lower-level robotic arm trajectory planning model. Based on the optimal base configuration, joint space geometric path, and dynamic safety envelope determined in the upper level, introduce pseudo-velocity variables to transform the trajectory time optimization into a state-dependent optimization problem. Then, perform constraint optimization on the robotic arm speed, acceleration, motion efficiency, interaction compliance, and joint torque burden to obtain a time-parameterized trajectory that meets the safety requirements of the confined space.

[0084] Step 6.1: Based on the optimization results of the upper layer, generate the joint space path and dynamic safety envelope, specifically as follows:

[0085] The optimal base configuration obtained from the upper-level planning As boundary conditions, the joint path transformation law constructed in step 4.2 is invoked. and dynamic security envelope This serves as the path basis and safety constraint for lower-level trajectory time parameterization optimization;

[0086] Step 6.2: Establish the relationship between joint motion states and path parameters, specifically as follows:

[0087] The time optimization problem is transformed into a state-dependent optimization problem, and the parameters of the robot arm joint velocity, joint acceleration, and path are established. The relationship between them; for the first For each path evaluation node, the joint velocity and acceleration of the robotic arm are represented as shown in Equations 20-21, respectively.

[0088]

[0089]

[0090] Wherein, pseudo-velocity is defined as , ;

[0091] Step 6.3: Establish velocity and acceleration constraints based on dynamic safety envelopes, specifically as follows:

[0092] based on Adaptive constraints are applied to the speed and acceleration of the robotic arm during its movement along the path, so that the robotic arm maintains high motion efficiency in areas far from the human body, seat and TMS host, and automatically reduces motion speed and acceleration in areas close to the human target point or in areas with small spatial gaps. The calculation formulas for speed and acceleration are shown in Equations 22-23, respectively.

[0093]

[0094]

[0095] in, and The first The maximum allowable speed and maximum acceleration of each joint; through the above constraints, the movement speed and acceleration of the robotic arm can be adaptively changed with the dynamic safety envelope, thereby meeting the safety requirements of human-machine co-operation in confined spaces;

[0096] Step 6.4: Construct the lower-level trajectory optimization objective, specifically:

[0097] Lower-level trajectory planning uses state functions To optimize the object, with motion efficiency, interaction compliance and joint torque load as optimization objectives, a set of lower-level objective functions is established, as shown in Equation 24.

[0098]

[0099] in, The motion efficiency target is used to shorten the time it takes for the robotic arm to move from the initial state to the target working state, as shown in Equation 25. The interaction compliance target is used to reduce abrupt changes in joint velocity and acceleration, and improve the motion stability of the robotic arm in the human-machine co-domain environment, as shown in Equation 26. The target for joint torque load is used to reduce joint drive load and long-term kinematic wear, as shown in Equation 27.

[0100]

[0101]

[0102]

[0103] in, The inertia matrix is ​​used. Since the three types of objectives have different dimensions and numerical ranges, direct weighting or comparison may cause one objective to have an excessive influence on the optimization result. Therefore, the objective functions are normalized and a lower-level comprehensive optimization objective is constructed, as shown in Equation 28.

[0104]

[0105] in, For the goal of comprehensive optimization at the lower level, and The first The minimum and maximum values ​​of the objective function in the current set of candidate trajectories.

[0106] Step 6.5: Solve the lower-level trajectory optimization model, specifically:

[0107] Under the conditions of satisfying dynamic safety envelope constraints, joint velocity constraints, joint acceleration constraints, and path endpoint constraints, the state function is... Optimize the solution to obtain the optimal path state function. ;according to Further calculation of the robotic arm's geometric path along the joint space The velocity, acceleration, and time parameterized trajectory are used to obtain the optimal robotic arm execution trajectory that balances motion efficiency, interaction compliance, and joint torque burden.

[0108] Step 7: Output the two-layer collaborative trajectory planning results of the robotic arm, and perform final verification on the optimal base configuration of the upper layer and the motion trajectory of the lower layer. After meeting the requirements of target pose, joint limit, end stiffness, dynamic safety envelope, cable constraint and obstacle avoidance distance, output the optimal base pose and the execution trajectory of the robotic arm. The final planning result is shown in Equation 29.

[0109]

[0110] in, For optimal base configuration, This represents the final trajectory of the robotic arm.

Claims

1. A two-layer cooperative trajectory planning method for a robotic arm in a confined space with human-machine co-domain conditions, characterized in that, Includes the following steps: Step 1: Based on the parameters of the human-machine co-operation scenario in a confined space, establish a robotic arm operation environment model; the scenario parameters include the robotic arm structural parameters, joint range of motion, end effector size, treatment target pose, patient human geometric model, treatment seat model, treatment host position, connecting cable parameters, and spatial pose information of surrounding auxiliary equipment; Step 2: Based on the structural diagram, establish the forward and inverse kinematics models of the robotic arm, determine the end-effector pose mapping relationship and candidate configuration set; among them, the inverse kinematics model considers the influence of the redundant angle of the end-effector treatment tool around its axis of action, constructs the end-effector target pose by introducing the end-effector redundant angle parameter, and solves the candidate joint configuration that satisfies the constraints; Step 3: Establish an end-effector stiffness prediction model to quickly evaluate the end-effector stiffness performance under different candidate configurations; the stiffness prediction model is based on a neural network, and its training samples are obtained by traversing the allowable arrangement space of the robotic arm and the human target sampling combination, and the labels are calculated offline by the physical stiffness model. Step 4: Construct the optimal pose planning model for the upper base, using the base spatial translation parameters and end-effector redundancy offset angle as optimization variables. Combine target accessibility, operability, flexibility, end-effector stiffness, joint limit, cable constraint and obstacle avoidance composite conditions to screen feasible base poses and end-effector working configurations. Step 5: Solve the upper-level optimization model based on the PIBF-MODE algorithm. The optimal base configuration that satisfies the performance and safety constraints is obtained through parent initialization, dual-channel offspring generation, interference slip repair, Pareto-based pre-screening and high-fidelity evaluation, population update and Nash decision. Step 6: Construct a lower-level robotic arm trajectory planning model. Based on the optimal base configuration, joint space geometric path, and dynamic safety envelope determined in the upper level, introduce pseudo-velocity variables to transform the trajectory time optimization into a state-dependent optimization problem. Then, perform constraint optimization on the robotic arm speed, acceleration, motion efficiency, interaction compliance, and joint torque burden to obtain a time-parameterized trajectory that meets the safety requirements of the confined space. Step 7: Output the two-layer collaborative trajectory planning results of the robotic arm, perform final verification on the optimal base configuration of the upper layer and the motion trajectory of the lower layer, and output the optimal base pose and the execution trajectory of the robotic arm after meeting all requirements.

2. The method for planning the two-layer collaborative trajectory of a robotic arm in a confined space with human-machine co-domain as described in claim 1, characterized in that, In step 3, the physical stiffness model considers the inherent stiffness of the robot arm joint space, geometric stiffness, the robot arm's own weight torque, the robot arm's Jacobian matrix, and the influence of the end-effector load on the end-effector stiffness. The neural network takes the relative position vector from the clamping device base to the target point, the joint angle vector, and the basic stiffness physical prior as input, and the total end-effector stiffness as output, and applies non-negative constraints to the output through the Softplus function.

3. The method for planning the two-layer collaborative trajectory of a robotic arm in a confined space human-machine co-domain as described in claim 1, characterized in that, In step 4, the upper-level optimization variables include the translation parameters of the base in three-dimensional space. , , and the redundant offset angle of the end-effector around the target direction. The upper-level optimization variables are represented as follows: 。 4. The method for planning the two-layer collaborative trajectory of a robotic arm in a confined space with human-machine co-domain as described in claim 1, characterized in that, In step 4, a linear geometric path in joint space is established based on the joint angles of the robotic arm in its initial and target states, and evaluation nodes are set along the path. At each node, the obstacle avoidance distance between the gripping device and the human body, the treatment chair, and the treatment host is calculated, and a dynamic safety envelope for velocity and acceleration decay is constructed based on this distance.

5. The method for planning the two-layer collaborative trajectory of a robotic arm in a confined space human-machine co-domain as described in claim 1, characterized in that, In step 5, the parent initialization includes: randomly generating candidate base positions and end redundancy offset angles to form an initial parent population based on the human target location, the maximum reachable range of the equipment, and the allowable layout area of ​​the base, and resampling individuals that do not meet the base layout range, cable length, or preliminary obstacle avoidance constraints.

6. The method for planning the two-layer cooperative trajectory of a robotic arm in a confined space human-machine co-domain as described in claim 1, characterized in that, In step 5, the dual-channel offspring generation includes a spatial position channel and a redundant spin channel; The spatial location channel generates candidate base position offspring based on the parent individual, the globally optimal individual, and random difference perturbation, while the redundant spin channel generates candidate redundant bias angle offspring through Gaussian perturbation.

7. The method for planning the two-layer cooperative trajectory of a robotic arm in a confined space with human-machine co-domain as described in claim 1, characterized in that, In step 5, the interference slip repair includes: sequentially judging the cable length, base obstacle avoidance, inverse kinematics solvability, joint limit and path safety constraints of the offspring candidate individuals; if there is a local collision or insufficient safety distance and it is within the repairable range, fine adjustment is made along the direction of increasing obstacle avoidance distance and recalculated; if the constraints are satisfied after repair, it is retained, otherwise it is eliminated.

8. The method for planning the two-layer collaborative trajectory of a robotic arm in a confined space human-machine co-domain as described in claim 1, characterized in that, In step 5, the Pareto-based pre-screening includes: taking the minimum cumulative joint angle change and the maximum end stiffness as the pre-screening targets, performing non-dominated sorting on candidate offspring and constructing a local Pareto first front, and then introducing crowding distance to determine a unique deterministic trial gene; subsequently, substituting the trial gene into the upper-level optimization model for high-fidelity comprehensive scoring, and comparing it with the parent individuals to update the population.

9. The method for planning the two-layer collaborative trajectory of a robotic arm in a confined space with human-machine co-domain as described in claim 1, characterized in that, In step 6, the speed and acceleration of the robotic arm during its path movement are adaptively constrained based on the dynamic safety envelope, so that the robotic arm maintains high movement efficiency in areas far from the human body, treatment seat and treatment host, and reduces movement speed and acceleration in areas close to the target or in areas with small spatial gaps. The lower-level trajectory planning takes motion efficiency, interaction compliance, and joint torque load as optimization objectives, and constructs a comprehensive optimization objective after normalization.

10. The method for planning the two-layer collaborative trajectory of a robotic arm in a confined space human-machine co-domain as described in claim 1, characterized in that, In step 7, the final verification includes target pose verification, joint limit verification, end effector stiffness verification, dynamic safety envelope verification, cable constraint verification, and obstacle avoidance distance verification. Once all verifications meet the requirements, the optimal base pose and the robotic arm's execution trajectory will be output.

Images