Spacecraft obstacle avoidance control method combining model predictive control with incremental safety constraints

Abstract

This invention discloses a spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints. The method includes: establishing a relative motion dynamics model of the spacecraft and obtaining current state information; generating nominal control inputs based on the current state information using model predictive control; acquiring environmental information within the spacecraft's current local perception range and constructing a local safe state region; generating an expert sample set within the local safe state region through safety reachability analysis and learning local control obstacle functions online; incrementally combining the newly learned local control obstacle functions with existing functions to form a global safety function; and constructing a safety filter based on the global safety function to correct the nominal control inputs, obtaining actual safety control inputs to drive the spacecraft's motion. This invention improves the robustness and real-time performance of obstacle avoidance control by generating dynamic safety constraints in real-time in unknown and complex environments through online learning and incremental combination of local control obstacle functions.

Description

Technical Field

[0001] This invention relates to the field of spacecraft control technology, and in particular to a spacecraft obstacle avoidance control method that integrates model predictive control and incremental safety constraints. Background Technology

[0002] In missions such as on-orbit servicing, space station inspection, approaching non-cooperative targets, and space debris removal, spacecraft need to perform operations such as approaching, circling, and observing near space stations or large spacecraft structures. Unlike conventional free-space flight, space stations typically consist of main modules, solar panels, trusses, connecting mechanisms, and external payloads, with complex geometry and issues of obstruction and limited local field of view. When flying in such an environment, spacecraft not only need to track target positions or reference trajectories but also must avoid collisions in real time under limited thrust, external disturbances, and localized sensing conditions.

[0003] Existing spacecraft obstacle avoidance control methods mainly include path planning methods, artificial potential field methods, model predictive control methods, control obstacle function methods, and learning-based control methods. Path planning and artificial potential field methods are relatively simple to implement and can generate obstacle avoidance trajectories in partially known environments. Model predictive control methods can simultaneously handle target tracking, control input constraints, and state constraints within a rolling optimization framework and have been widely used for close-range spacecraft operation and obstacle avoidance control. Control obstacle function methods can transform safety requirements into control constraints in real time and perform minimum corrections to nominal control inputs through quadratic programming, exhibiting good real-time performance and closed-loop safety control capabilities. In recent years, learning-based methods such as neural networks, Gaussian processes, and reinforcement learning have also been used for safety boundary learning to reduce reliance on accurate environment models and artificial safety function designs. However, the above methods still have certain limitations. Path planning, artificial potential fields, and some model predictive control methods typically rely on pre-obtained obstacle geometry information or global environment models. However, targets such as space stations have complex non-convex structures like solar panels, trusses, and module connection mechanisms. Using spheres or ellipsoids to enclose obstacles results in conservative depictions of safe zones, making it difficult to reflect the traversable areas near real space targets. Near complex space structures, collision avoidance constraints often exhibit strong non-convexity. Directly embedding them into model predictive control problems requires linearization, increasing online computation. For spacecraft with limited onboard computing resources and high control requirements, this is insufficient to meet real-time control needs. While control obstacle function methods are suitable for real-time safety filtering, their effectiveness depends on the construction of the safety function. For known obstacles with simple geometry, safety functions can be manually designed based on distance functions. However, in the unknown and complex environment of a space station, spacecraft typically rely only on onboard sensors to obtain local environmental information, making it difficult to construct globally effective control obstacle functions before the mission begins. Although learning-based methods have some adaptability, their safety typically relies on extensive data training and empirical verification, making it difficult to directly provide strict real-time safety guarantees.

[0004] A search revealed that Chinese Patent Publication No. CN121697882A discloses a spacecraft cooperative formation safety control method and system based on an adjustable adaptive control barrier function. This scheme establishes a six-degree-of-freedom relative motion model for spacecraft formation, defines multiple key safety constraints such as obstacle avoidance, collision prevention, and attitude exclusion zones, designs a disturbance observer to estimate and compensate for unknown lumped disturbances online, and constructs an adjustable adaptive control barrier function. Finally, it solves for the optimal control command that satisfies the safety constraints through quadratic programming. However, the obstacle avoidance constraint of this scheme is based on a pre-defined spherical envelope geometry model, and its control barrier function is a pre-defined fixed mathematical form rather than being learned and generated online. Furthermore, it relies on prior information such as known obstacle positions and attitude exclusion zone directions, making it difficult to learn the control barrier function in real time and incrementally construct safety constraints for unknown complex non-convex spatial structures with only local perception information.

[0005] Therefore, how to generate and update safety constraints in real time for unknown and complex space station structures under the condition that spacecraft can only obtain local environmental perception information, and achieve robust and safe obstacle avoidance control while ensuring target tracking performance, is a technical problem that needs to be solved. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a spacecraft obstacle avoidance control method that integrates model predictive control and incremental safety constraints.

[0007] The objective of this invention can be achieved through the following technical solutions: According to a first aspect of the present invention, a spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints is provided, comprising: Establish a dynamic model of the relative motion of the spacecraft and obtain the current state information of the spacecraft; A model predictive control method is used to generate nominal control inputs based on the current state information; Acquire environmental information within the current local sensing range of the spacecraft, and construct a local safe state region based on the environmental information; Within the local safe state area, an expert sample set is generated through safe reachability analysis, and the local control obstacle function is obtained through online learning using the expert sample set; The newly learned local control obstacle function is incrementally combined with the previously learned local control obstacle function to form a global safety function; A safety filter is constructed based on the global safety function, and the nominal control input is corrected using the safety filter to obtain the actual safety control input, thereby driving the spacecraft's motion.

[0008] As a preferred technical solution, the safety reachability analysis adopts the Hamilton-Jacobi-Isaacs reachability equation. By solving the equation, a local safety value function is obtained, and safe samples, buffer samples, and unsafe samples are divided according to the local safety value function. For the safe samples, expert control inputs are obtained according to the Hamiltonian maximization condition to form the expert sample set.

[0009] As a preferred technical solution, the local control barrier function is parameterized using a compactly supported radial basis function. By applying classification constraints to the safe samples, buffer samples, and unsafe samples in the expert sample set, and combining the expert control input, the learned local control barrier function can simultaneously satisfy boundary safety and control feasibility.

[0010] As a preferred technical solution, for each local control barrier function, if the centers of all tightly supported radial basis functions with positive coefficients are located within the effective region of the local control barrier function, and the distance from the region boundary is not less than the support radius of the radial basis function, then the local control barrier function takes a negative value outside its effective region.

[0011] As a preferred technical solution, the incremental combination adopts a point-by-point maximum value approach, defining the global security function as the maximum value of all learned local control barrier functions: ,in K This is the set of indices for the learned local control barrier functions. For the first k A local control barrier function.

[0012] As a preferred technical solution, the incremental combination further introduces an approximate activation set, which is defined as follows: ,in ϵ >0 is the preset threshold.

[0013] As a preferred technical solution, the safety filter is a quadratic programming safety filter based on the control barrier function. Its optimization objective is to minimize the deviation between the actual safety control input and the nominal control input. The constraints include: satisfying the control barrier function safety constraint condition for all local control barrier functions in the approximate activation set, and control input amplitude constraints.

[0014] As a preferred technical solution, the safety constraint condition of the control barrier function is: , in, d To disturb the upper bound, α (⋅) is a class K function.

[0015] As a preferred technical solution, the relative motion dynamics model of the spacecraft is described by the HCW equation, and the environmental information within the local sensing range is acquired by a LiDAR sensor.

[0016] As a preferred technical solution, the model predictive control adopts a predictive-correction iterative solution method. At each sampling time, a predicted trajectory is generated based on the current control sequence, and the nominal prediction model is linearized to the first order at the prediction point. The control correction is obtained by solving the local quadratic programming subproblem. The control sequence is iteratively updated until convergence, and the first term of the optimal control sequence is taken as the nominal control input.

[0017] Compared with the prior art, the present invention has the following advantages: 1. This invention solves the problem of real-time generation and online updating of safety constraints driven by local perception, combined with a hierarchical control architecture of model predictive control and safety filtering of control obstacle functions. This solves the problem of difficulty in generating and updating safety constraints in real time when spacecraft can only obtain local environmental information and face unknown and complex space station structures. It achieves robust and safe obstacle avoidance control while ensuring target tracking performance.

[0018] 2. This invention generates an expert sample set through safety reachability analysis and learns local control obstacle functions, so that safety constraints reflect the geometric distance of obstacles and also include spacecraft velocity constraints, control input capability constraints and external disturbance effects, thereby improving the dynamic feasibility and control effectiveness of safety constraints.

[0019] 3. This invention uses a point-by-point maximum method to incrementally combine the newly learned local control obstacle function with the learned local control obstacle function, and introduces an approximate activation set to process the non-smooth boundary region of the combined function. It does not require the prior acquisition of a complete global map, and can gradually expand the safety coverage as the spacecraft moves, adapting to complex non-convex space structures, thus improving the versatility and scalability of the method.

[0020] 4. This invention uses model predictive control to generate nominal control inputs and combines them with a quadratic programming safety filter based on control obstacle functions to obtain actual safety control inputs in a minimum correction manner. While ensuring safety, it preserves target tracking performance and decouples complex non-convex obstacle constraints from the model predictive control problem, reducing the online computational burden and making it suitable for spaceborne real-time control.

[0021] 5. This invention explicitly incorporates an upper bound on external disturbances into the safety constraints of the obstacle avoidance function, so that even under the worst bounded disturbance, the system state can still remain within the forward-invariant safety set, thus enhancing the robustness of obstacle avoidance control. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a closed-loop obstacle avoidance trajectory of a spacecraft in a complex space station environment, as described in this embodiment of the invention. Figure 3 This is the incremental expansion process of the local control barrier function in the embodiments of the present invention; Figure 4 This illustrates the variation of the global security function H(x) along the closed-loop trajectory in this embodiment of the invention. Figure 5 This is a comparison of the safety residuals of the symbolic distance function method and the incremental CBF method in this embodiment of the invention; Figure 6This is the closed-loop obstacle avoidance trajectory of the method in a complex space station scenario according to an embodiment of the present invention; wherein, Figure 6 (a) in the diagram represents the solar panel connection area; Figure 6 (b) in the diagram represents the compartment interface area; Figure 7 This is an online expansion process for the coverage area of ​​local control obstacle functions in complex scenarios, as described in this embodiment of the invention. Figure 8 This invention provides a comparison of closed-loop trajectories using two baseline methods in complex scenarios; wherein, Figure 8 (a) in the figure represents the closed-loop trajectory of the MPC-only method; Figure 8 (b) in the figure represents the closed-loop trajectory of the MPC+SDF baseline method; Figure 9 This presents a comparison of the safety performance of different control methods in a complex space station scenario according to embodiments of the present invention; wherein, Figure 9 (a) in the figure represents the change in the closest distance between the spacecraft and the space station structure; Figure 9 (b) in the figure represents the change in safety residuals for different methods. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] Example 1: like Figure 1 As shown, this invention provides a spacecraft obstacle avoidance control method that integrates model predictive control and incremental safety constraints. In unknown and complex space environments, it can progressively construct safety constraints usable for closed-loop control based solely on local perception information, avoiding reliance on global environment models and artificial safety boundaries. While ensuring spacecraft target tracking performance, it improves obstacle avoidance safety and real-time control capabilities near complex non-convex space targets. It is suitable for autonomous approach, fly-around, inspection, and obstacle avoidance missions near space stations, large spacecraft structures, non-cooperative targets, and space debris. This method executes the following steps in each control cycle: Step S1: Establish a dynamic model of the relative motion of the spacecraft and obtain the current state information of the spacecraft; Step S2: Using model predictive control, generate nominal control inputs based on the current state information; Step S3: Obtain environmental information within the current local sensing range of the spacecraft, and construct a local safe state region based on the environmental information; Step S4: Within the local safe state area, an expert sample set is generated through safe reachability analysis, and the local control obstacle function is obtained through online learning using the expert sample set; Step S5: Incrementally combine the newly learned local control barrier function with the already learned local control barrier function to form a global safety function; Step S6: Construct a safety filter based on the global safety function, and use the safety filter to correct the nominal control input to obtain the actual safety control input, so as to drive the spacecraft to move.

[0025] The above steps are executed online in real time during each control cycle, enabling the spacecraft to gradually learn and expand safety constraints in an unknown and complex space station environment, relying only on local LiDAR perception information, and achieve coordinated and unified obstacle avoidance and target tracking.

[0026] The steps of this invention will be described in detail below.

[0027] Step S101: Modeling the relative motion of the spacecraft.

[0028] This invention uses the LVLH coordinate system to describe the translational motion between the spacecraft and the space station reference point, assuming the relative position of the spacecraft is... The relative speed is The state vector is ,in, These represent the relative position components of the spacecraft in the three directions of the LVLH coordinate system. These represent the relative velocity components in the corresponding directions. Let represent an n-dimensional real space, with the superscript T indicating the transpose of a matrix or vector. Spacecraft control input is denoted as . ,in, These represent the control accelerations acting in the three coordinate directions. The external disturbance is... ,in, This represents a bounded perturbation acting on the spacecraft's acceleration channel. Unless otherwise specified, This represents the Euclidean norm, also known as the 2-norm.

[0029] Under the assumptions of small relative distances and near-circular reference orbits, the relative motion of the spacecraft can be approximately described by the Hill-Clohessy-Wiltshire (HCW) equations: Where n is the average angular velocity of the reference orbit. These represent the relative accelerations in the three directions, respectively.

[0030] Furthermore, the above dynamic model can be written in a control affine form: , in, This represents the derivative of the state vector with respect to time. Represents the drift dynamics term of the system. Let represent the control input matrix, and d represent the lumped bounded perturbation term. Specifically, , ; Disturbance satisfies ,in, As the upper bound of the disturbance, the set of control inputs is ,in, This represents the upper limit of the spacecraft control input amplitude; the disturbance set is... ,in, express Zero-dimensional vector; initial time step denoted as The initial state is denoted as This model provides a precise dynamic basis for subsequent controller design.

[0031] Step S201: MPC generates nominal control input.

[0032] The main objectives of upper-level model predictive control (MPC) are to achieve target tracking, control input smoothing, and control energy optimization. To avoid excessive online computational burden caused by directly embedding complex non-convex obstacle avoidance constraints into MPC, this invention delegates the obstacle avoidance task to the lower-level safety filter. Therefore, MPC adopts a nominal dynamics model without disturbance terms. , At sampling time t, let the prediction time domain length be N and the sampling period be... Define the predictive control sequence as The predicted state sequence is ,in, This represents the k-th step control input predicted at the current sampling time t. This represents the k-th step state predicted at the current sampling time t.

[0033] The nominal dynamic model is discretized using the Euler discretization method, resulting in: For ease of representation, the discrete nominal prediction model is defined as follows: Then the prediction model can be written as .

[0034] Let the reference state be ,in, For target reference position, Let the target reference velocity be the reference state, which is selected from the currently constructed combined safe region. This aims to ensure that the nominal predicted trajectory moves towards the target direction while remaining within the safe region as much as possible. The model predictive control problem is written as follows: st in, This represents the obtained optimal predictive control sequence. This represents the terminal state error weight matrix. This represents the state error weight matrix during the prediction process. This represents the control input weight matrix. This represents the control input increment weight matrix, when k=0. It can be taken as the actual control input applied at the previous sampling time.

[0035] To avoid repeatedly solving the complete nonlinear finite-time optimization problem at each sampling time, this embodiment employs a prediction-correction iterative method, in the first... In the next iteration, based on the current control sequence Generate the corresponding predicted trajectory Define the state correction and control correction as follows: , At the prediction point Nearby, for discrete nominal prediction models After performing first-order linearization, we get: , in, , respectively, represent the discrete prediction model's state at the current prediction point. and control input The Jacobian matrix.

[0036] The state corrections over the entire prediction time domain are stacked as follows: Control correction amount stacked as Then there is ,in, Depend on and A linear mapping matrix constructed based on the recursive relationship in the prediction time domain.

[0037] The control correction is then obtained by solving the local quadratic programming subproblem. and in accordance with Update the control sequence, where, For the first The step size of the next iteration satisfies After iteration, the optimal control sequence is obtained. And take the first item as the nominal control input at the current moment. The purpose of this prediction-correction iteration is to reduce the online computational burden caused by repeatedly solving the complete finite-time constrained optimization problem.

[0038] Step S301: Local perception and safe zone construction.

[0039] At each control moment, the spacecraft acquires local environmental information within its current visible range via its onboard LiDAR. Let the spacecraft's relative position at the current moment be denoted as . If the LiDAR sensing radius is R, then the local sensing area is: , in, Indicated by A three-dimensional spherical sensing area centered on a point with a radius of R.

[0040] The set of local obstacles currently sensed by LiDAR is denoted as . The locally unobstructed area is , where the symbol " It represents the difference between sets.

[0041] Considering spacecraft velocity constraints, the velocity set is defined as follows: ,in, This represents the upper limit of spacecraft speed, thereby constructing a local state region. and local safe zone ,in, This represents the set of all states that satisfy the velocity constraints within the current local sensing range. This represents the set of safe states where a location is within a locally unobstructed area.

[0042] Step S401: Local CBF learning based on HJI analysis.

[0043] Within a small local area, safe expert data is generated using Hamilton-Jacobi-Isaacs (HJI) reachability analysis. Then, a lightweight local control barrier function (CBF) is learned in a data-driven manner. This CBF not only reflects geometric distance but also incorporates spacecraft dynamics, control capabilities, and disturbance effects.

[0044] Step S4011: HJI reachability analysis.

[0045] Within a local state region, this embodiment generates local security expert data through Hamilton-Jacobi-Isaacs secure reachability analysis. First, based on the current local obstacle set... Define terminal security functions ,in, Represents the distance from position r to the local obstacle set. The minimum distance, Indicates the preset safe distance. When When, it indicates that the current position is not less than the preset safe distance from the obstacle; when When the current position is within a safe distance, it indicates that the current position has entered the safe distance range.

[0046] Solve the local HJI equations within a finite time interval: , And adopt terminal conditions ,in, Represents a local safety value function. express Regarding the status The gradient of the safety reachability analysis is given by T, which represents the terminal time of the safety reachability analysis. This safety value function is used to determine whether a spacecraft can remain safe under conditions of limited control inputs and worst-case disturbances.

[0047] Based on the local security value function, the sampled state set Divided into safe sample set, buffer sample set and unsafe sample set }, , in, This represents the set of sampling points selected within the local state region. This represents the safety margin threshold. For safe samples, the expert control input is calculated based on the Hamiltonian maximization condition. , in, Indicates the first Each sampling state, This represents the corresponding security expert control input. When the denominator is not zero, the expert control input takes values ​​along the Hamiltonian maximization direction; when the denominator is zero, the nominal control input can be used, thus obtaining the local security expert dataset. , This indicates the number of expert samples in the current local area.

[0048] Step S4012: Parameterized learning of local CBF.

[0049] After obtaining local security expert data, this embodiment learns local control barrier functions. For the first... For a local region, the local control barrier function is defined as follows: ,in, Indicates the first A local control barrier function, Represents a basis function vector. Represents the vector of parameters to be learned. This is a safety bias term. The local safety set is defined as follows: ,in, Indicates the first The effective domain of a local CBF.

[0050] This embodiment uses a compactly supported radial basis function (CS-RBF) for construction. Let the selected basis function centers within the local region be... The support radius is Then the radial variable is defined as For the six-dimensional spacecraft state space, we select the Wendland compactly supported radial basis functions: , in, This represents the normalized distance variable.

[0051] The basis function vector can be written as ,in, This indicates the number of basis function centers selected within the current local region. (Parameter) The following optimization problem was solved to obtain the following result: , , , in, Indicates the safety margin. Indicates the margin of insecurity. This represents the dynamic safety constraint margin. (Function) For extended classes The function, in this embodiment, can be taken as... .symbol express Regarding the status gradient, express Regarding velocity components The gradient, the first type of constraint mentioned above, makes the safe sample at... If positive, the second type of constraint makes the unsafe sample location... If the value is negative, the third type of constraint ensures that the dynamic conditions of the control barrier function considering the effects of disturbances are satisfied at the safe and buffered samples.

[0052] Step S501: Incremental global combination of local CBF.

[0053] Since each local control barrier function is only effective near the current LiDAR sensing range and cannot directly cover the complete and complex space station environment, this embodiment uses an incremental combination method to construct a global safety function.

[0054] Step S5011: Combining the maximum values ​​point by point.

[0055] Let the set of indices of the local control barrier functions that have been learned and retained be . Then the combined security function is , represents a global non-smooth safety function obtained by combining multiple local control barrier functions, and the corresponding combined safety set is . ,when Time indicates state It must be located within at least one local safe set; when Time indicates state It does not belong to the currently constructed safe set.

[0056] Step S5012: Approximate activation set processing.

[0057] During online operation, a local safety constraint update is triggered when any of the following conditions occur: the current local CBF set is empty; the preset refresh cycle is reached; or the current combined safety function value is lower than the preset refresh threshold. Because... This is obtained by maximizing multiple local control barrier functions, and non-smoothness may exist at the boundaries of different local safety functions. To address this problem, an approximate activation set is defined. ,in, To approximate the activation threshold, this set represents the set of local CBF indices near the current state that play a major role in the global security function. During online security filtering, for... All local CBFs within the area are simultaneously subject to safety constraints to prevent constraint failure at the local safety zone switching point.

[0058] Step S5013: Negative conditions outside the domain.

[0059] To ensure that the safe region is not erroneously expanded after the local CBF combination, an extra-domain negativity condition for the local CBF is further introduced, that is, for the ... k For each local CBF, if all satisfy... CS-RBF Center Located in the local effective region of CBF Inside, and satisfy Then we get This condition ensures that the local CBF provides the correct safety decision only within its effective area.

[0060] Step S601: CBF-QP security filter.

[0061] The combined global security function is embedded in a quadratic programming (QP)-based security filter. At each control time step, the nominal control input is first obtained from the upper-level model predictive control. The lower-level CBF-QP safety filter then performs minimal corrections to obtain the actual safety control input applied to the spacecraft. The security filter is written as: , , , in, express mechanics of drifting Lie derivative, express Along the control matrix Lie derivative, This represents robust compensation to external bounded disturbances; the objective of this quadratic programming is to optimize the actual control input. As close as possible to the nominal control input Simultaneously satisfying both the control input constraints and the security constraints corresponding to the current combined security function, when When the safety constraints are already met, the output of the safety filter is and Approaching; when When a spacecraft approaches an obstacle or enters an unsafe area, the safety filter automatically corrects the control input through secondary programming to keep the closed-loop trajectory within the combined safety set.

[0062] The online execution process is as follows: Input the initial state of the spacecraft. Target area LiDAR sensing radius Refresh cycle Refresh threshold Approximate activation threshold Control input upper limit and safe distance Initialize the local CBF index set Local CBF serial number Control Step Current state At each control moment, the spacecraft first acquires local LiDAR observation information and constructs a local state region. .like Then perform a local safety constraint update; if Then first combine the current global security functions. And determine whether to update the local CBF based on the refresh cycle and refresh threshold. When the update conditions are met, Internal sampling Solve for the local HJI safety value function and calculate the expert control input. Divide the sampling state into , and Then learn new local control barrier functions. If the local CBF meets the feasibility requirements for online security filtering, then the local CBF is accepted and updated. and order Then, the current global security functions are recombined. Select reference state Solving the upper-level MPC yields Construct an approximate activation set Solving the CBF-QP security filter yields... and will Acting on the spacecraft dynamics system, updating the state and causing The above process is repeated until the spacecraft enters the target area. .

[0063] The method of this invention does not rely on a complete global environment model and can generate and update safety constraints in real time and online in an unknown and complex space station environment. This improves the reliability of spacecraft obstacle avoidance and robustness to disturbances near non-convex obstacles, while ensuring target tracking performance and real-time computation.

[0064] Example 2: To verify the effectiveness of the method of the present invention, a complex obstacle avoidance simulation scenario near a space station was constructed based on the spacecraft HCW relative motion model.

[0065] The simulation employs a six-dimensional spacecraft relative motion model. The space station consists of a main module, lateral structures, and multiple solar panels, exhibiting a complex non-convex geometry. The initial position of the spacecraft in the simulation is set to... The target location is set to The control input constraints are The sampling period is taken The average angular velocity of the reference orbit is taken as Local sensing uses LiDAR, with a sensing radius of [missing information]. Safe distance is taken The threshold for the approximate activation set is taken as follows Partial updates employ a dual-trigger mechanism: every Each control step calls a local HJI security analysis once, or when the current global security function is invoked. Below the refresh threshold At this time, an additional local update is triggered. The local CBF learning problem is solved using numerical optimization methods.

[0066] Figure 2 This is a closed-loop obstacle avoidance trajectory in the complex space station environment. The space station consists of main modules, lateral structures, and multiple solar panels, exhibiting complex non-convex geometry. (Spacecraft initial position) The target location is set to The control input constraints are The sampling period is The average angular velocity of the reference orbit is The spacecraft from its initial position Starting from the nominal input generated by model predictive control and the incremental CBF safety filter correction, the system bypasses the main body of the space station, the solar panels and connecting structures, and finally reaches the vicinity of the target area with a position error of approximately 0.804m. This demonstrates that even without a pre-prepared complete global obstacle map, the present invention can still complete closed-loop obstacle avoidance flight in a complex space station environment through local perception and incremental safety constraints.

[0067] Figure 3 This represents the incremental expansion process of the local control obstacle function. The red curve indicates the closed-loop trajectory executed by the spacecraft from the starting point to the current refresh phase, the yellow marker indicates the spacecraft state that triggers the local safety constraint update, and the blue circular area represents the local safety coverage area accumulated over time. The local sensing radius is R=8.0m, set in the local safety analysis. As the spacecraft moves along its closed-loop trajectory, LiDAR continuously acquires new local environmental information, triggering local HJI safety analysis and local CBF learning. This occurs every [period]. Each control step invokes a local HJI update, triggering a total of 68 local refreshes throughout the flight. Some local models that made safety filtering infeasible or unsuitable for global composition were discarded, ultimately retaining 38 local CBFs for constructing the global safety function. Each subgraph represents the gradual expansion of the local safety coverage area as the spacecraft moves. This result demonstrates that the present invention can gradually expand the safety coverage area as the spacecraft explores, without needing to obtain the complete global safety function before the mission begins.

[0068] Figure 4 For global security functions The curve representing the change along the closed-loop trajectory shows the horizontal axis representing the spacecraft's flight time from its initial position to its target position, and the vertical axis representing the value of H(x) along the closed-loop trajectory. In this embodiment, the global safety function is derived from multiple local CBFs. The combination yields the desired result. Figure 4 The security boundary in is , The corresponding safe state region indicates that the spacecraft is within the safe set. Simulation results show that during closed-loop flight... The fact that it remains positive throughout indicates that the spacecraft's state is always within the combined safety set. Within the system, as local CBFs are continuously added, the safety coverage gradually expands, and online CBF-QP can continuously provide effective safety corrections.

[0069] Figure 5 The diagram compares the safety residuals obtained using the Signed Distance Function (SDF) method and the incremental CBF method. The horizontal axis represents the runtime of the spacecraft's closed-loop obstacle avoidance process, and the vertical axis represents the degree to which the current state and control inputs satisfy the safety constraints. The safety residual is defined as... The safety boundary is r=0, where, This indicates that the current control inputs meet the safety constraints. This indicates that some safety constraints are not met. Figure 5 It can be seen that the SDF method exhibits negative residuals at certain times. When t≈26s, the residuals of the SDF method drop to negative values, with the lowest residual being approximately -0.23. This indicates that the geometric safety constraints constructed solely based on the signed distance function cannot continuously satisfy the dynamic safety conditions. In contrast, the safety residuals corresponding to the incremental CBF method of this invention remain above r=0, indicating that the control barrier function obtained through incremental learning can more stably support online safety filtering.

[0070] Figure 6 This demonstrates a closed-loop obstacle avoidance trajectory in a complex local structure scenario. The scenario guides the spacecraft through the solar panel connection area and the module interface area to test the obstacle avoidance capabilities of this invention near complex local structures and narrow areas. Figure 6 (a) in the diagram represents the solar array connection area, a complex local structural region that the spacecraft passes through, where local safety constraints are frequently updated. Figure 6 In diagram (b), the module interface area represents a local region of the space station's main body around which the spacecraft orbits. The red trajectory indicates the actual flight path of the spacecraft after applying MPC and incremental CBF safety filtering. The black dots represent the locations where local HJI safety analysis and local CBF learning are triggered. The blue spherical area represents the LiDAR sensing range. Figure 6 It is evident that when the spacecraft passes through the solar array connection area and the module interface area, the system will trigger a local refresh near the complex structure, learn a new local CBF, and correct the nominal control input given by the upper MPC through CBF-QP, thereby maintaining collision-free flight.

[0071] Figure 7 This diagram illustrates the online expansion process of local CBF coverage areas in complex structural scenes. Figure 6For the same complex local structure simulation process, each sub-graph corresponds to a representative stage in the same simulation process. The red curve represents the flight path completed by the spacecraft before the current stage, the black dots represent the locations where local safety constraint updates are triggered, and the blue spherical areas represent the local sensing range of the spacecraft in the current stage. As the spacecraft orbits the space station, the local sensing area continuously changes, and new local control obstacle functions are sequentially added to the global combination. A total of 123 local updates were triggered in this scenario, retaining 67 local CBFs. This result demonstrates that, near complex local structures, this invention can continuously generate, filter, and accumulate local safety constraints, allowing the coverage of the combined safety function to continuously expand with the spacecraft's motion.

[0072] Figure 8 This paper compares the closed-loop trajectories of two baseline methods in complex structural scenarios: the MPC-only method and the MPC+SDF baseline method. The MPC-only method directly uses the nominal control input of the upper-level model's predictive control output, lacking lower-level safety filtering corrections. While the MPC+SDF baseline method incorporates a safety correction based on the signed distance function, this constraint primarily reflects geometric distance relationships and cannot fully account for the impact of spacecraft velocity, control capabilities, and disturbances on safety. Figure 8 It is evident that both baseline methods are in When the spacecraft enters the safety buffer zone near the upper solar array and a safety distance violation occurs, its position is approximately [missing information]. The closest distance to the space station structure is approximately Because the preset safe distance is The corresponding safety margin is This comparison illustrates that, near complex non-convex structures, relying solely on nominal MPC or simple SDF geometric constraints may not be sufficient to maintain the preset safe distance.

[0073] Figure 9 The curves show the comparison between the closest distance and the safety residual for different control methods. Figure 9 Figure (a) shows the change in the closest distance between the spacecraft and the space station structure. Figure 9 (b) shows the changes in safety residuals for different methods. Figure 9 As can be seen in (a) above, the MPC-only method and the MPC+SDF baseline method... Nearby below the preset safe distance This indicates that both baseline methods enter the safety buffer; by Figure 9As can be seen in (b), the safety residual of the MPC+SDFbaseline method becomes negative when it approaches a complex structure, while the safety residual of the Learned-CBF method of the present invention remains non-negative, indicating that the incremental CBF safety constraint constructed by the present invention can continuously satisfy the CBF-QP safety filtering condition.

[0074] The simulation results above show that the method of the present invention can gradually update local safety constraints as the spacecraft moves without relying on a complete global environment model, and achieve safe obstacle avoidance near complex non-convex space structures. Compared with existing methods, the present invention has stronger safety and adaptability, and verifies the feasibility and superiority of the proposed hierarchical control, local learning, incremental combination and perturbation robust CBF design.

[0075] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints, characterized in that, include: Establish a dynamic model of the relative motion of the spacecraft and obtain the current state information of the spacecraft; A model predictive control method is used to generate nominal control inputs based on the current state information; Acquire environmental information within the current local sensing range of the spacecraft, and construct a local safe state region based on the environmental information; Within the local safe state area, an expert sample set is generated through safe reachability analysis, and the local control obstacle function is obtained through online learning using the expert sample set; The newly learned local control obstacle function is incrementally combined with the previously learned local control obstacle function to form a global safety function; A safety filter is constructed based on the global safety function, and the nominal control input is corrected using the safety filter to obtain the actual safety control input, thereby driving the spacecraft's motion.

2. The spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints according to claim 1, characterized in that, The secure reachability analysis employs the Hamilton-Jacobi-Isaacs reachability equation. By solving the equation, a local security value function is obtained, and based on this local security value function, secure samples, buffered samples, and insecure samples are divided. For the secure samples, expert control inputs are obtained according to the Hamiltonian maximization condition, forming the expert sample set.

3. The spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints according to claim 1, characterized in that, The local control barrier function is parameterized using compactly supported radial basis functions. By applying classification constraints to safe samples, buffer samples, and unsafe samples in the expert sample set, and combining this with expert control input, the learned local control barrier function can simultaneously satisfy boundary safety and control feasibility.

4. The spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints according to claim 3, characterized in that, For each of the local control barrier functions, if the centers of all tightly supported radial basis functions with positive coefficients are located within the effective region of the local control barrier function, and the distance from the region boundary is not less than the support radius of the radial basis function, then the local control barrier function takes a negative value outside its effective region.

5. The spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints according to claim 1, characterized in that, The incremental combination adopts a point-by-point maximization approach, defining the global safety function as the maximum value of all learned local control barrier functions: ,in K This is the set of indices for the learned local control barrier functions. For the first k A local control barrier function.

6. The spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints according to claim 5, characterized in that, The incremental combination also introduces an approximate activation set, which is defined as follows: ,in ϵ >0 is the preset threshold.

7. The spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints according to claim 6, characterized in that, The safety filter is a quadratic programming safety filter based on the control barrier function. Its optimization objective is to minimize the deviation between the actual safety control input and the nominal control input. The constraints include: satisfying the control barrier function safety constraint for local control barrier functions in all approximate activation sets, and control input amplitude constraints.

8. The spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints according to claim 7, characterized in that, The safety constraints of the control barrier function are as follows: ， in, d To disturb the upper bound, α (⋅) is a class K function.

9. The spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints according to claim 1, characterized in that, The relative motion dynamics model of the spacecraft is described by the HCW equation, and the environmental information within the local sensing range is acquired by a LiDAR sensor.

10. The spacecraft obstacle avoidance control method integrating model predictive control and incremental safety constraints according to claim 1, characterized in that, The model predictive control employs a predictive-correction iterative solution method. At each sampling time, a predicted trajectory is generated based on the current control sequence, and the nominal prediction model is linearized to the first order at the prediction point. The control correction is obtained by solving a local quadratic programming subproblem. The control sequence is iteratively updated until convergence, and the first term of the optimal control sequence is taken as the nominal control input.

Images