Abstract
An on-board autonomous task planning system is designed and implemented in this article, aiming at the problem that the current remote sensing satellite needs complex instruction support to perform tasks and depends on the ground system too much. The complex earth observation task description and injection decomposition modules are designed in the system. No more than 128 ordered points describe the curve area task or irregular polygon area task, and the complex task is decomposed into several strips according to the satellite imaging width. Then, the task’s orbit, attitude, mobility, energy, and time constraints are calculated through the modules of on-board orbit prediction, agile attitude calculation, track planning, energy prediction, and visible arc calculation. Finally, the on-board autonomous task planning and execution are completed through task solution space search and metatask command generation modules. The on-orbit flight verification is carried out on the high-resolution multimode (GFDM) satellite. The results show that this paper’s on-board autonomous task planning system can complete complex task injection and autonomous planning and finally execute.
1. Introduction
Space-based remote sensing systems play a critical role in scenarios of flood and forest disaster prevention, land survey, and other emerging fields. Ground operation and management are becoming more complex due to the increasing trends in the complexity of observation requirements and the number of satellites in constellation or swarm [1–4]. Hence, the traditional task planning mode is hard to apply [5, 6]. In addition, current satellite management and control procedure relies on the ground telemetry and telecontrol (TM & TC) station, which leads to a longer time duration from the input of task requirement to remote sensing product generation. In such a centralized mode, the task planning system has flaws in image information timeliness, emergency task response, multisatellite collaborative scheduling, etc. Therefore, it is urgent to develop on-board intelligence and autonomy, including on-board task planning, real-time evaluation, decision-making, multisatellite, multipayload cooperation, and other technical research.
Most current task planning systems are applied in the ground computing environment; they do not consider the computing and storage resources of on-board computers [7–14]. Correspondingly, most currently developed satellites have minimal computing ability and memory space, which is impossible to plan complex tasks on orbit. So, the algorithms have to be redesigned to fit the ability of on-board computers. To overcome shortage of computational resource, Zhang et al. designed a satellite and ground station hybrid method to plan remote sensing tasks [15]. An autonomous task plan framework that could run on-board was proposed to solve agile satellite imaging task planning problems [16, 17]. On JWST satellite, the event-driven system was established to deal on-board tasks [18]. Here, high-resolution multimode satellite (GFDM) is taken as an attempt to experiment with on-orbit task and attitude planning, its processing capacity is no more than 50 MHz, and SRAM is no more than 6 MByte. To suit the GFDM satellite, the on-board task planning system was modified to ensure computational efficiency in limited on-board resources and complete planning in small storage space.
This paper designed an on-board automatic task planning system and conducted a set of on-orbit flight verification, aiming at the new requirements of task planning for agile remote sensing satellites. The second section will give the overall design, planning algorithm design, and processing flow design of the on-board task planning system. The third section will detail the on-orbit flight verification based on the GFDM satellite. Finally, the conclusion and prospect of this paper are given in the Section 4.
2. Design of On-Board Task Planning System
The existing agile satellite takes a single strip or a spot area as a metatask to complete remote sensing [19]; here, the metatask refers to the task that can be imaged or executed by switching on and off once by the payload [20]. The on-board task planning system designed in this paper is compatible with these metatasks; the architecture is shown in Figure 1. Except for metatasks, for a better description, the complex task is defined in this paper as the task that can only be imaged entirely requiring several contiguous strips, such as observation tasks of curve strip or polygonal area. The on-board task planning system of agile remote sensing satellites considers the satellite orbit, imaging width, and agile ability. The metatask is the basic observation task executed by the on-board task planning system, and the complex task needs to be decomposed into multiple metatasks. The task planning system will decompose the curve strips or irregular polygons into a series of strips or spots according to the comprehensive constraints of satellite imaging width, subsatellite tack, memory size, power, and acceleration of angle velocity.
2.1. Overall Design
The ground operation and control center receives users’ observation requirements and injects multiple complex tasks or metatasks into satellite. The on-board task planning system autonomously completes task planning, instruction generation, and other operations. The system architecture is shown in Figure 1. After receiving and decomposing the complex tasks, the satellite implements task planning according to the real-time on-orbit state telemetry. A task planning process is divided into 8 subprocedures, that is, from requirement gathering to instruction generating, including the following:
Step 1. Gather requirements. The users send tasks to the ground station by means of polygons, curve strips, gaze, and tracking tasks and then forward to satellites through the satellite-ground link.
Step 2. Examine the validity of the task. Check task time, attributes, and types to determine whether the task is valid or not.
Step 3. Decompose complex tasks. The on-board task planning system decomposes and processes complex tasks according to the subsatellite track vector, generating the metatask.
Step 4. Planning execution window. Select imaging orbital arc.
Step 5. Calculate external constraints. Calculate the constraints of power [21], storage, attitude, and so on.
Step 6. Plan the attitude and attitude maneuver trajectory.
Step 7. Generate metatasks.
Step 8. Generate instructions.
2.2. Complex Task Description and Decomposition
2.2.1. Task Description
Complex tasks include irregular curve strips, irregular areas, point group target, etc. Tasks can be described by the following essential attributes: (1)Task ID: the unique ID of the task, which is used to retrieve, delete, or modify the task(2)Task type: it includes point target task, nontrack strip task, curve strip task, and region task(3)Task priority: it includes high, medium, and low priorities. Task priority is mainly used to ensure the execution of tasks with high priority after multiple tasks are mutually exclusive(4)Payload type: it can be divided into visible light, multispectral, infrared, and other types according to the type of remote sensor(5)Time range of observation task: the observation and data return of the task must be completed within the time range of the observation task. Some emergency observation tasks, such as flood and other disaster monitoring, will lead to the loss of due value of observation data beyond a specific time range(6)Solar elevation angle: solar elevation angles for visible light imaging tasks(7)Cloud cover: the prediction of cloudiness over imaging target and the probability of covering to target
2.2.2. Decompose Tasks
Wu et al. presented a novel splitting algorithm to region target for satellite swarm [8], which is the first comprehensive solution for satellite task planning. Taking the method in the paper as a reference, a region target is handled by strips paralleled to subsatellite track.
(1) Decomposition of Polygonal Area Task. It is necessary to decompose the polygonal areas according to the trend of the subsatellite track since a single strip cannot cover a complex task area, as shown in Figure 2.
The first thing is to examine whether all area points are within the visibility corridor boundaries related to the subsatellite track and satellite maneuver before decomposing. The area would be decomposed according to the boundaries’ width if it passed the examination. Otherwise, this area becomes unobservable this time, so there is no need to decompose.
The visibility corridor range can be calculated by where and are the satellite’s minimum and maximum observation angle, is the satellite maneuver range, and is the FOV (field of view).
The polygon area is decomposed based on the width after obtaining the maneuver range, and the dynamic decomposition process is as follows:
Step 1. The irregular polygon is initialized into a convex polygon by convex hull algorithm, and the subtask (metatask) set is initialized, .
Step 2. Find the left most point position and the right-most point position in the irregular area, set the left direction to negative and the right direction as positive, and let .
Step 3. Obtain the strip , and let . Strip is obtained by finding the intersected figure of polygon and straight lines and , where is obtained by drawing a line through paralleled to the subsatellite track, and is obtained by translating by width .
Step 4. Update the current location, ; the updated result is shown in Figure 3.
Step 5. Check whether is less than ; if true, turn to Step 2.
Step 6. Output , end decomposition.
(2) Decomposition of Curve Strip Task. The curve strip task is described by point set in which is the characteristic point of a curve and is checked according to agile ability of satellite and imaging width . Only the curve whose characteristic points set are all in observable corridor can be decomposed. The procedure is as follows:
Step 1. Selecting as a start point of vector and selecting point , one by one, as end point of vector , then, the vector is obtained.
Step 2. According to imaging width and vector , the rectangle whose length is variable can be obtained.
Step 3. Selecting in , to get m_max end point, then to form vector . Further, strip that can cover all characteristic points between and can be obtained.
Step 4. Using to update , then repeating Step 1 to Step 3, until .
The decomposition procedure of the curve strip task is shown in Figure 4.
2.3. Task Execution Time Window Calculation Algorithm
Visible window can be calculated based on the satellite’s maximum maneuver angle and FOV , as is shown in Figure 5.
A brief process of the visible window algorithm is as follows:
Step 1. Initialize the satellite observation range and the execution window set .
Step 2. The starting execution window and the ending execution window for metatask are calculated according to the satellite orbit prediction and , where is the starting point and is the ending point.
Step 3. Let .
Step 4. For , repeat Step 2 and Step 3.
Step 5. Final set will be the prediction of execution time window for every metatask decomposed from this complex task.
2.4. Task Planning
is recorded as the complex task after decomposing (briefly introduced in Section 2.2.2), where is the number of the meta tasks.
For meta-task , (i) is longitude and altitude at starting point (ii) is longitude and altitude at ending point (iii) is strip length
The execution time is finally determined by the attitude and track planning, satisfying multiple constraints including attitude angle, angle velocity, and execution time window.
2.4.1. Attitude Planning
The attitude representation method is based on 1-2-3 Euler angular rotation sequence. The roll angle is , pitch angle is , and yaw angle is . The attitude planning algorithm is as follows:
Step 1. Calculate the observation point position at the execution time in J2000 coordinate system.
; the starting point position at in the J2000 coordinate system can be calculated by its longitude and latitude, where the execution time window for metatask is obtained in Section 2.3.
Step 2. Calculate the position and velocity of the satellite at the execution time in the J2000 coordinate system.
Orbital data at can be obtained from the satellite GPS receiver, including : semimajor axis; : eccentricity; : orbit inclination angle; : right ascension of ascending node; : argument of perigee; and : mean anomaly.
According to spacecraft dynamics equation [22, 23], , the position and velocity in the J2000 coordinate system can be obtained by the following formulas, after calculating the true anomaly : where is the geocentric distance of the satellite; is the satellite velocity; is the argument of latitude of the satellite; is the track angle; is the geocentric gravitational constant, and .
Step 3. Calculate the satellite attitude parameters.
Roll angle and pitch angle are obtained by the following formulas: where is the vector difference between the imaging point and the satellite in the J2000 coordinate system and is the orbit coordinate system converted from .
And roll angular velocity and the pitch angular velocity can be calculated using quadratic interpolation method, to improve the calculating speed and suit the limited on-board computational resources.
Then, obtain the yaw angle according to , , , and . The yaw angular velocity is calculated in the same way as obtaining and .
Step 4. Calculate the attitude parameters at the ending point.
is obtained by the formula where is the push-broom speed of the satellite and is the strip length of a metatask.
If , which means the execution window cannot cover ; this metatask has no solution. Otherwise, can be obtained by repeating Step 1–Step 3 according to the orbit data at .
2.4.2. Track Planning
The satellite needs to switch attitude through track planning, from at to at between executing metatasks and , with constraints of (i)maximum switching time (ii)maximum angular velocity (iii)maximum angular acceleration
Other than that, the variation curves of angle , angular velocity , and angular acceleration in every axis (yaw, pitch, and roll) of the satellite need to be smooth. This is to ensure that the remote sensing satellite can perform tasks without jitter.
The specific algorithm is as follows: (i), where is the minimum track planning time (we set in this paper)(ii)Calculate the satellite attitude parameters at (iii)Obtain the acceleration using the quadratic interpolation method
Step 1. Initialize , and calculate the attitude parameters.
Step 2. Obtain the attitude parameters at every control period time (0.1 second in this paper) in , based on the 5th-order track planning method.
The formulas of this method are shown below: where is the track planning starting angle, angular velocity, and angular acceleration at and is the track planning ending angle, angular velocity, and angular acceleration at .
Step 3. Calculate the compound angular velocity and compound angular acceleration using 1-2-3 Euler angle rotation kinematical equation, which can obtain maximum compound angular velocity and maximum compound angular acceleration.
Step 4. Let and . If and , let , track planning end. Otherwise, let , update , and repeat Step 1–Step 3.
2.4.3. Analyzing Time Complexity
The proposed method calculates angle relations between each orbital points and target vertexes that compose strip, region, and curve. Orbital points are orbital position in ECI (Earth Central of Inertial) coordinate system, and the points are sorted by time. The count of orbital points is second; for example, there are seconds in a day. If the count of vertexes is , the time complexity of the proposed algorithm is .
3. On-Orbit Verification Results and Analysis
An on-orbit flight test on satellite (GFDM) with a single-core BM3803 processor (main frequency: 50 MHz) has been taken to verify the system and method proposed in this paper. Each satellite subsystem executes the generated instruction by receiving it from the on-board routing system using the 1553B avionics bus.
This verification was carried out on July 5th, 2020. This verification was carried out using 1/6 of the computing resources for automatic task planning. The result turns out that the proposed method can accomplish complex task, including decomposition and automatic planning.
A complex task consisting of a non-along-track strip task and a curve area task is taken as an example in this paper to introduce the on-orbit verification briefly. The specific process and results of this verification are as follows: (1)Orbit data in verification
GFDM satellite takes the trajectory between two descending nodes as one orbit. Data output interval is set as 1 second. In on-orbit verification, the orbit data is between 2020-7-5 11:40:00 (UTCG) and 2020-7-5 19:40:00 (BJT), semimajor axis 7020.634788 km, eccentricity 0.000381, inclination angle 97.905112°, right ascension of ascending node 260.231848°, argument of perigee 155.560876°, and mean anomaly 229.827296°. (2)Task injection in verification
Data of the two subtasks (a curve area task and a non-along-track strip task) are as follows: (1)Geographic position of curve area task: <-51.7°, 4.3°>; <-51.2°, 3.9°>; <-50.7°, 2.2°>; <-50.5°, 1.9°>; and <-50.0°, 1.7°>(2)Geographic position of non-along-track strip task: <-58.4°, -33°>; <-58.4°, -35°>
The generated instruction package taking the curve area task as an example is shown in Figure 6. (3)On-board task decomposition in verification
The command to start the on-board autonomous mission planning is sent at 19:50:11, July 5th, 2020, Beijing time. After about 117 seconds, the on-board mission decomposition and planning results are received at 19:52:08, July 5th, 2020, Beijing time, and transmitted to the ground center in the form of telemetry package. As shown in Figure 6, the curve task is decomposed into three imaging metatasks. Taking metatask 1 as an example, its planned start time is 173887001 (0x0a5d4e19) seconds of on-board time. The corresponding Beijing time is 2020-7-5 21:55:48. The time coding method is similar to most typical embedded system, like IoT (Internet of Things) [24, 25].
The command to boot the on-board autonomous task planning is at BJT 2020-7-5 19:50:11. And the results of on-board task decomposition and planning are at BJT 2020-7-5 19:52:08. The instruction package is shown in Figure 7. (4)Attitude data after planning
The results of attitude planning and track planning are transmitted down to the ground station in the form of instruction package, as shown in Figure 8. (5)Curve strip task verification
To verify the curve strip task, a curve task with 5 characteristic points is planned on board. Five points by longitude and latitude with degree are <-51.7, 4.3>; <-51.2, 3.9>; <-50.7, 2.2>; <-50.5, 1.9>; and <-50.0, 1.7>. The curve and its partial zoomed position are shown in Figure 9 in which Figure 9(a) is the global view. The curve strip task is planned and executed by GFDM satellite, and its partial image (corresponding to Figure 9(b)) is shown in Figure 10. (6)Non-along-track strip task verification
A non-along-track strip with start point at <-58.4, -33> and end point at <-58.4, -35> is injected to on-orbit satellite. The non-along-track strip is lined in Figure 11(a), and its partial is zoomed into Figure 11(b).
Non-along-track strip task was successfully planned and executed on-board on July 5th, 2020, Beijing time, and its result of image at Buenos Aires is shown in Figure 12.
4. Conclusion
The results from the experiment of the GFDM show that the automatic on-board task planning system can bring the following capabilities’ improvement: (1)The on-orbit task planning is feasible. The automatic level of satellite is improved; meanwhile, the dependence of satellites on the ground has been decreased(2)Facilitated customer manipulation. Customers can focus on task-related matters and do not need to consider the satellite’s complex parameters and knowledge(3)Fully utilized satellite capability. The task execution is optimized based on the real-time constraints on the satellite, including energy, storage, and orbit
Data Availability
The data used to support the findings of this study are included within this paper.
Conflicts of Interest
The authors declare that there is no conflict of interest regarding the publication of this paper.
Acknowledgments
This research was supported by the Fundamental Research Funds for the Central Universities of China (No. 3072022JC0202), the project D030307.