PRE2019 1 Group1
Autonomous systems for space traffic management
Group Members
| Name | Study | Student ID |
|---|---|---|
| Stijn Eeltink | Mechanical Engineering | 1004290 |
| Laura Kulter | Psychology & Technology | 0851512 |
| Annelies Severens | Biomedical Engineering | 1232787 |
Planning
Each week will consist of two meetings. Before each meeting, the team will work individually on the tasks they have been assigned for that meeting. During the meetings, the results of these tasks will be discussed and finalized.
L = Laura, S = Stijn, A = Annelies.
| Week | Monday (morning) | Wednesday (afternoon) |
| 1 | ALL : choose topic | ALL : literary research problem definition make the planning define structure of the report |
|---|---|---|
| 2 | L : introduction/problem statement L : wiki page A : state of the art |
ALL : PCR A: state of the art S : stakeholders L : introduction/problem statement L : edit planning in wiki |
| 3 | political aspects economical aspects technical aspects (state of the art) PCR |
A: state of the art on wiki S: edit stakeholders L: RPC’s and update wiki A: air traffic management ALL: 2 questions for stakeholders (more is allowed) |
| 4 | S: questionnaire with information S: stakeholders in googledocs L: autonomous mining, how does the system work? A: Structure of the system. Would a peer-to-peer system work? Which systems does NASA or ESA use?
ALL: comment each other’s work |
All: More specific RPC, how can results be measured? All: Specify and categorize specific user questions A: System architecture of third parties S: Stakeholders L: AI systems (rule-based vs learning system) |
| 5 | contact users
read whole report and comment: ALL |
Contact users Prepare for tutor meeting Monday |
| 6 | Contact users |
S: Roadmap/Requirements A+L: Ethical decision styles L: Write out user research L: Machine Learning ALL: How could the system reward good behavior? A: How would satellites be detected? |
| 7 | finalize report and wiki |
Final evaluation (peer review) finalize report and wiki |
| 8 | Presentation Deadline : report and wiki |
Introduction
Space debris often gets the most attention when one talks about threats that exist to active satellites and other spacecraft (previous 0LAUK0 groups have done extensive research on that topic before: PRE2016_3_Group19, PRE2018_3_Group1, PRE2018_4_Group9). However recent developments in the space industry present a ‘new’ threat to active satellites.
Where traditionally space travel was government-driven, the privatization and commercialization of space activities have gained momentum and have developed different interests like faster and cheaper access to space. Currently, several organizations plan to launch thousands of satellites up into Earth’s orbit in the next several years. These range from governments like the UK planning to launch 2000 satellites by 2030[10] to large companies like SpaceX planning to launch 12000 satellites for its Starlink constellation.[10] If we compare this to the currently 4987[10] satellites in orbit, of which only 1957[10] are still active and functional, one quickly sees how ‘full’ the currently quite ‘empty’ low orbit space around earth will become in the near future.
This could overwhelm current space flight safety processes. However, there are encouraging signs that the government, industry, and the space community are acting to address these issues. This project will look at a possible solution for managing these many thousands of satellites by using an autonomous system for space traffic management(STM).
Problem Statement
As stated in the introduction this project will focus on the need for an Autonomous Space Traffic Management System (henceforth called ASTM). The main reason for the need for such a system is the ever-increasing presence of active satellites in low earth orbit, which will make it no longer feasible in the near future to avoid collisions by depending on human input. There are however a lot of aspects that will need to be analyzed and discussed before a proper concept for an ASTM system can be proposed. First, the current state of the art will need to be analyzed to see how STM is handled right now and where there is room for improvement. Next, the individual stakeholders will be looked at to find the wants and needs of each stakeholder. These things will be summarized in the RPC (requirements, preferences, constraints) and work as a framework for designing a proper ASTM system.
The main reason the end goal is a concept instead of a fully-fledged system is the fact that right now STM is still quite in its infancy. Even large organizations like NASA have only started to fairly recently look into autonomous systems for STM.[8] This means that there are many unexplored factors in regards to creating and deploying an ASTM system and so it is beyond the scope of this project to analyze all these factors in only eight weeks.
The concept will mostly be a recommendation for what type of autonomous system would be most suited to handle STM. Therefore this project will mostly ignore the technical aspects(radio communication in space, the inner workings of satellites, deploying an ASTM system on earth vs in low orbit, etc.) and instead focus on analyzing what an ASTM system could(and should) offer compared to doing STM by hand. This will also include a recommendation which kind of Rational Agent Model (RAM) would be best suited to control an ASTM system.
State of the art
Legal
At the moment, there are no international or even national Space Traffic Management systems. However, because of the increasing amount of non-governmental organizations executing space activities, rules are needed to ensure safety in air space. Generally speaking, Space Traffic Management can be defined by the safety insurance of: 1. Safe access to outer space, 2. The conduction of operations in outer space, and 3. The return of space objects from outer space free from interference of any form.[1]
Currently, the Outer Space Treaty forms a basis of international space law.[2] The treaty was opened in 1967, when the United States, the United Kingdom, and the Soviet Union signed the treaty. More countries followed in the coming years. As of 2019, 109 countries are parties of the treaty. This treaty focuses on the limitation of the use of celestial bodies and restricts nations from claiming sovereignty of outer space. It does not include any legal regulation of a Space Traffic Management. At the time that the treaty was set up, the STM concept was not considered a priority. In 2015, the UNCOPUOS committee had received approval to add STM as an agenda item in 2016.[3]
The Cosmic Study from IAA created a definition for STM. It was the first step, but too premature to implement any regulations limiting freedom.[4]
The 2016 “Orbital Traffic Management Study – Final Report” does not contain a definition for space traffic management. Instead, it defines Space Traffic Safety. Management would imply centralized command and control, which was seen as problematic.[5]
The 2017 German Aerospace Center (DLB) White paper on the “Implementation of a European Space Traffic Management System” defines STM as:[6]
Over the years, there have been different definitions and approaches to STM from the United States, European Space Agency (ESA) and the International Academy of Astronautics (IAA). However, they have some similar key operations, one of which is collision avoidance. This focuses on point 2. of Space Traffic Management: the conduction of operations in outer space.[1] Recently, the European Space Agency (ESA) has performed a collision avoidance maneuver for the first time. The satellite was moved off a potential collision course with a SpaceX satellite in the Starlink constellation. At the moment, ground operators make decisions, which might not always be optimal. The avoidance process between two satellites is largely manual and will be no longer practical if the number of satellites and other space vehicles increases. According to Holger Krag, Head of Space Safety at ESA:[13]
The use of machine learning, artificial intelligence, is being explored to support ground operators when planning and implementing collision avoidance.[7] This is one application that artificial intelligence can be used for. In [8] an initial architecture for a Space Traffic Management system is proposed, based on open Application Programming Interfaces (APIs). The use of machine learning incomplete STM systems is being explored at the moment, a great step towards complete autonomous STMs. According to [1] a national system will be most probably implemented before an international regime. This mainly has to do with the fact that having a good, working STM system, data must be shared between governmental and non-governmental organizations, which remains a difficult topic.
Similar Systems
While there is a lack of proper STM in space and an autonomous system for it is still non-existent, there do however already exist similar systems in other real-world applications.
Air Traffic Control
Air traffic control includes the ground-based personnel and equipment concerned and monitoring air traffic within a particular area. Air controllers ensure safe operations of commercial and private aircraft. They, manually, keep them at safe distances from one another and direct them during takeoff and landing. An equivalent system is needed to direct satellites and other spacecraft from earth into space into their increasingly crowded flight paths. Whenever a rocket is launched into space, it must pass through the airspace where thousands of planes fly through every day, causing planes to reroute. However, an analogous system to that of Air Traffic Control is difficult for several reasons. The Federal Aviation Administration (FAA) runs the air traffic control system. Airspace is divided into zones, and each zone into sectors, segregated by national authority. When an aircraft moves through a particular sector, it is monitored by the air traffic controllers responsible for that area. Space is not separated into sections or divisions with a responsible entity monitoring and controlling that area. Nations are responsible for their spacecraft and would rather improve their system before internationalizing an STM. Also, both governmental and non-governmental organizations are unwilling to give full access to all their data regarding the position of their satellites and other spacecraft. Without widespread participation and positional information, it is difficult to create a command-and-control system that works well for STM. Data collection is, therefore, one of the main issues for any STM.
Autonomous Mining
A big implementation of autonomous systems can be found in the mining industry. Fortescue Metals Group in Australia uses autonomous systems for various operations, of which their autonomous haulage system[21] (AHS) comes closest to a real-world example of an autonomous traffic management system. AHS is used to control the massive driverless trucks at mining sites that are responsible for transporting ores etc. from point A to B. The system is responsible for making sure the driverless trucks don’t collide with each other or other obstacles and stick to their designated routes.
According to a public news report[20], there has been only one reported accident involving the driverless trucks so far since the active operation started in 2012. In February 2019 one of the trucks collided at low speed with another parked truck after a wifi outage caused the truck to lose its connection with the control center. While two driverless trucks colliding with each other might not be a big deal and would mostly have a low monetary impact, such a result would not be acceptable with for instance communication satellites.
While only a single accident in 7 years is a noteworthy feat there are a few things to note here. The trucks drive in a closed environment owned by the company. This means that the system seldom has to take into account non-company owned trucks or other vehicles. The system knows everything about all the members at the mining site. However, in space, there are thousands of satellites with hundreds of different owners and not everyone is willing to share their information. Another factor is that since the mining site is owned by the company, they decide (within the boundaries of the Australian law) what happens and who can or cannot enter the site. The international space treaties however basically state that nobody is the boss in space and that everyone should have the right to shoot satellites into space. This means that while the autonomous system does work and is certainly a technological feat, it owns much of its success to the extremely limited and controlled environment it operates in.
Architecture Model
Space agencies like ESA and NASA are currently using distributed systems for their programs.[a],[b] In most of the implementations, the system consists of a central computer, the server, and many user computers, the clients.
There are several architectures of distributed systems: peer-to-peer and client-server. Client-server is a system that is divided into servers, that carry out tasks, and clients, which require the service. A server computer can manage several clients simultaneously, and at the same time, a client can be connected to several servers, that provide different services, simultaneously. Peer-to-peer (P2P) is an architecture where all computers and devices, called peers, work together. There is no central administrator and each peer is equal to the other peers. Files can be shared directly between systems without a central server, each network can be seen as a server and a client, that work towards a common goal. These computers work together to function as a single application for the client. The decentralization makes the network efficient and more tolerant for faults.
Important programs from NASA, such as NASA’s Earth Observing System Data and Information System (EOSDIS) are designed as distributed systems.[a] For the last couple of years, NASA has been investigating Distributed System Missions (DSM). A DSM is a mission that involves spacecraft to achieve one or more common goals. However, these missions are currently focused on the use of client-server architectures. P2P architectures have yet to be investigated.[c]
ESA uses a distributed computing environment as well, in the form of client-server architectures.[b] The Advanced Concepts Team (ACT) is part of ESA that focuses on new technologies that could be of importance in the long term, especially on innovations in distributed computing. P2P architectures are not yet implemented in the system that ESA uses for its programs, however, the ACT department is looking in this relatively new technology.[d]
Summary: NASA and ESA use distributed architecture systems. There are several types. (1) Server-client, which NASA and ESA use, where there are server computers and client computers. (2) peer to peer systems (P2P), where all computers are both clients and servers. Not much can be found that indicates NASA is researching P2P, ESA has a team (ACT) that is occupied with innovative technologies for the long term, such as P2P.
Conclusion: There has been too little research on the effects of the implementation of p2p systems, to say with certainty that a p2p system would work as STM. As a first concept, it would be best to choose a system that is compatible with the ones used by space agencies at the moment, which is a distributed client-server architecture.
Stake Holders
The last years have seen rapid growth and change in the space industry. Where traditionally space travel was government-driven and solely focused on security, political or scientific activities. The privatization and commercialization of space activities have gained momentum and have developed different interests like faster and cheaper access to space. This easier access to space has opened participation to many more participants than was historically possible. Private companies have proposed, funded and begun deployment of very large constellations of satellites.
These new activities could overwhelm current space flight safety processes. However, there are encouraging signs that the government, industry, and the space community are acting to address these issues. But is their effort enough, and is there a need for an Autonomous Space Traffic Management System?
Social
Space debris is not some far off distant problem that we need to worry about. The probability of collisions occurring is not only continually rising, but there have been several collision events that have damaged the international space station and other satellites. Every time we launch into space, we generate some sort of unwanted waste. This is happening exponentially more since the privatization of space activities. Small particles like aluminum-oxide, explosive bolt fragments, and paint chips can cause serious issues. While this debris is small, it is also moving at 30.000 km/h, which means that it has enough kinetic energy to damage important satellites and even the international space station.
This could be devastating to society. Satellites with critical duties we rely on every day could be struck. Global communication, GPS and navigation, and weather data could suddenly disappear. If this debris collides with such a satellite, it will be destroyed instantly. This satellite then turns into thousands of little pieces, that are capable of destroying other satellites. This could trigger an unstoppable chain reaction, which is called the Kessler syndrome. If this were to happen, the loss of our space infrastructure would set lots of technology advancements back, and limit the technological advancements made from space travel in the future.
NASA and ESA are running multiple missions [22] dedicated to observing the Earth, to gather information on how the planet is changing. Missions like Aqua [23] collect information on ocean evaporation, atmospheric water vapor, clouds, precipitation, soil moisture, sea and land ice, and snow cover. With this information, NASA wants to show the evidence for climate change, the causes, effects, and try to find solutions. NASA also observes our planet’s atmosphere, where ozone, nitrogen dioxide, and particulate matter [24] can be measured to indicate the air quality. Because air pollution is responsible for about 1 in 9 deaths worldwide, this data is very important.
Other missions from NASA like SMAP are valuable to agriculture. The satellite uses a radiometer that can see through the clouds to measure the soil moisture levels on earth. By measuring the moisture levels in the soil, it allows you to predict droughts, monitors floods and even predict crop yields for a given year. The data from this program is widely available and is used by all countries for better agriculture. The technological advancements made in space travel have lots of different spinoff technologies. If space travel were to stop, the constant flow of new patents and technologies would also stop. Advancements in health, medicine, transportation, public safety, computer technology, and industrial productivity are key to the development of human society.
Political
While space activity has democratized with many new players, the U.S. government is still the single largest actor and stakeholder in the lower earth orbit (LEO) operations environment. The U.S. government re-established the National Space Council in the summer of 2017. One of its first actions was to establish a working group to recommend a way forward on space traffic management. “National Space Traffic Management Policy” was issued on June 18, 2018, and outlines several steps changing how space traffic is managed and regulated. This paper addresses the need for improved space situational awareness (SSA), data sharing with other organizations, and space traffic management (STM).
The Department of Commerce wants to simplify the regulatory structure for licensing for commercial companies, which the industry has needed for a long time. It will also take the function of STM and SSA for the U.S. Air Force. By creating an open-architecture space data repository they will actively share information with and between operators, and encourage new technologies for SSA.
The Federal Communications Commission has been regulating practical orbital debris for commercial companies that operate in the U.S. market. The new rules would explicitly address the issue of large constellations and post-mission disposal reliability. These new rules also contemplate active SSA data sharing, transponders, enhanced signatures, and shared maneuver plans, which would greatly decrease the amount of space debris.
Space is however fundamentally an international concern since no nation owns or controls the environment. The foundational document for international space law is the Outer Space Treaty. Though this treaty is not enough. The United Nations Committee is considering new rules for topics like space debris management and creating guidelines for the long-term sustainability of space.
While these combined actions have mitigated some of the risks in the transition, further action is recommended. The best source of innovation and solutions are the organizations that are building new systems. Industry-driven norms and standards of behavior are among the most effective methods for preventing the new activity from contributing to space debris. The government should encourage these industry-driven, voluntary approaches.
Economics
This change in space activities, especially the very large LEO constellations, represents major investments by commercial companies like SpaceX. Every U.S. operator proposing a large constellation has stated the intention of following best practices and being ‘’good citizens’’ of space. These operators have a significant vested interest in maintaining the space environment, and in protecting their investments that will run into the billions of dollars. Some of the new operators are among the strongest proponents advocating for increased regulation and scrutiny. They intend to build in high reliability for post-mission disposals, like their intent to deploy satellites at a low altitude, and then raising the orbit once checkout is complete. While there are some disadvantages to this approach, when a satellite fails, drag can bring it down much earlier.
They are building in the capability of high-precision orbit knowledge and are actively willing, even seeking, to share position and maneuver data. Also, a high level of automated collision avoidance and automated deorbit of failed systems are being developed, much like the ASTM proposed by our group. To make sure the post-mission disposal plans are successful, companies are planning to deorbit on a fixed schedule, rather than maximizing mission life as is commonly done. Also, operators are adding grappling fixtures, reflectors, and other retrieval aids, even if they have no intent for on-orbit servicing or retrieval.
Scientific
As is touched upon previously, space travel is key for scientific research, technological advancements, and spinoff technologies. Originally space activities were only used for scientific research and with private companies looking to make money in this area, we need to make sure we can sustain our research in this field. More missions are researching the earth than ever before. The atmosphere, the climate, the continental drift and geodynamics, the gravity, hurricanes, the ice, the land and vegetation, the oceans, ozone, the sun and its influence on Earth, the water cycle, the weather, and wildfires are all studied by multiple missions. [25] An STM system is mandatory to keep expanding this ever-growing research field, as well as to make sure private-owned constellations, satellites and space ships will not interfere with this research.
Space travel already requires a lot of data science, AI and cybersecurity development, and the STM system will only contribute to this. This is because a server network capable of tracking and monitoring more than 5 million objects is needed. This server system must be protected against cyber-attacks. It could form a huge threat to society if the server system falls into the wrong hands, and it is of uttermost importance to keep this from happening.
Requirements, Preferences, Constraints (RPC)
With the state of the art and stakeholders in mind, it is time to set down the framework by defining the requirements, preferences, and constraints. With these, it will be possible to analyze rational agent models and to begin constructing a proper concept for an ASTM system.
Requirements
The system should be able to do the following:
- Autonomous space flight and collision avoidance for all participating members in low earth orbit;
- Nowadays more than 500.000 objects are tracked by NASA. As the amount of space debris is expected to grow tenfold in the next decade, our system should be able to handle more than 5 million objects to provide safety for the participating members for the coming 10 years.
- Because there can be many non-participating members like space debris or active satellites not part of the ASTM control the system should be able to detect these non-participating members and keep participating members from colliding with them;
- Be able to work with incomplete, inaccurate or slightly false information. Especially military organizations will be unwilling to disclose full or any information regarding strategic satellites. There is also the chance of inaccurate sensor information. In each case the system should try to use the combined data of sensors and satellites in its group to make an as accurate guess as possible;
- Fully autonomous operation. Human interference should only be needed in situations the system cannot solve (like an approaching collision with no ‘no loss’ solution);
Preferences
The following items, while not absolute requirements, would still be desired for a good ASTM system:
- Easy compatibility. To make the system as accessible to as many organizations as possible the system should be able to easily connect with different kinds of satellites, including different messaging systems and/or operating systems. This could be achieved by centralizing the system, instead of it needing to be installed on satellites it would just require to be able to listen and talk to satellites in their ‘language’;
- Find least-cost solutions to avoid approaching collisions (making a group of 10 satellites move out of the way instead of the group of 1000 for example); (MAYBE REQUIREMENT?)
- Ability to assist in coordinating spacecraft back to earth when the end of life has been reached;
- The ability to not just react to collisions when they are about to happen but to also use 3D models and learning algorithms to predict possible collisions early on and take preventative measures if predicted collision risk reaches a certain threshold;
Constraints
The constraints should never be violated, this also has mostly to do with international space treaties. So a system that can not meet one of these constraints will automatically not be an option:
- Original owners/operators (o/o) should always be able to regain control of satellites. The system is a service, not an owner;
- System should in no way violate the international space treaties(for instance nobody is the boss in space, so the system will not influence satellites that aren’t participating);
- The system should be impartial in its judgment and only use a cost-benefit analysis to make decisions;
- Like any other form of robots or artificial intelligence the system has to follow Asimov’s three laws of robotics;[15]
Concept
AI systems: Rule-Based Systems vs Machine Learning Systems
AI systems can mainly be split into two groups, rule-based systems(RBS) and machine learning systems(MLS), with each their benefits and drawbacks. Choosing between these systems is important because they form the foundation when one wants to build an autonomous system and so in proposing an ASTM concept a choice for one or the other should also be presented. First follows a basic explanation of both systems with each their pros and cons. The ongoing discussions in the field of AI about whether an RBS system is actually intelligent or not will not be considered here.
Rule-Based Systems
In principle, RBS use coding in the form of if-then-else statements using the knowledge of a human expert. This means that the system uses a (large) set of predetermined rules and facts to make decisions and take actions. A key benefit being that these rule sets can be easy to write, especially in limited environments.[x2]
In most closed systems (like a robot arm in a car manufacturing plant) this is fine, but in more complex and changing environments this leads to major drawbacks in the long run. The biggest drawbacks of this system are that adding rules later on without accidentally creating contradictions with earlier rules is tough, basically, the entire system needs to be checked when rules are added making maintenance time consuming and expensive in the long run.
Pros:
- Easy to write (if all possible cases are already known);
- System can only do as told, so no need to worry about unexpected behavior;
- A few special cases can easily be added later on;
Con’s;
- System has limited knowledge and actions, when it encounters new situations it will get stuck until the program is updated by human interference;
- In more complex environments where a lot can change over time maintenance will become time consuming and expensive, making the system unwieldy;
- System can not make predictions since it can not update its own rules;
Machine Learning Systems
Machine learning systems, on the other hand, can adapt to situations more easily and often on their own without human interference. When a learning system encounters new information it can change or discard existing models in favor of (hopefully) better models. This means that a learning system is not limited to its initial knowledge or static rules set. However with learning systems, depending on the area of application, training can be very important to help the system build ‘good’ models initially instead of starting from nothing. Now in some areas that can be an issue, there might be no available prior knowledge and so a system builder would let several learning systems take a shot at it and hope that one of them comes with a useable model. Luckily with STM, there is already a lot of prior data to use which can help in creating ‘good’ models for the system to start from. However, the adaptability of this system is also its major drawback, if the system creates a wrong model it can lead to a lot of damage based on how late or early the faulty model is discovered.
Pros:
- Can easily adapt to new information and situations;
- In more complex environments the system can be implemented more easily since it can create its models where human expertise might be incomplete;
- Can use prediction models by looking at historic data;
Con’s:
- Models used by MLS are like a black box, users can see the input and output but do not know why the system makes certain decisions, especially in legal situations this can be a problem;
- MLS need to be trained, depending on the availability of good historic data this can make or break a system early on;
Rational Agent Models
Conclusion
References
[1] (https://iislweb.org/docs/Diederiks2017.pdf)
[2] http://www.unoosa.org/pdf/publications/STSPACE11E.pdf
[3] UN General Assembly resolution A/RES/70/82, “International cooperation in the peaceful uses of outer space”, 21 December 2015 Online: http://www.unoosa.org/oosa/oosadoc/data/resolutio ns/2015/general_assembly_70th_session/ares7082. html, (accessed 06.09.2017);
[4] K.U. Schrogl, “Space Traffic Management: The new comprehensive approach for regulating the use of outer space – Results from the 2006 IAA cosmic study”, Acta Astronautica 62, 2008, pp. 272-276
[5] O.Brown et al.: “Orbital Traffic Management Study – Final Report”, prepared for National Aeronautics and Space Administration (NASA) Headquarters, prepared by Science applications InternationalCorporation(SAIC), 21 November 2016.
[6] R. Tüllmann et al.: “On the Implementation of a European Space Traffic Management System – Volume I. A White Paper; Volume II. The Safety and Reliability Strategy; Volume III. Technical Requirements”, conducted on behalf of European Space Agency (ESA) by German Aerospace Center (DLR) and partner Institutes and Companies, 27 April 2017.
[7] file:///C:/Users/20166004/Downloads/artificial-intelligence-support1.pdf: more explanation about collision maneuvers
[8] https://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/20180007349.pdf
[9] http://www.esa.int/Our_Activities/Space_Safety/ESA_spacecraft_dodges_large_constellation
[10] https://www.pixalytics.com/satellites-orbiting-earth-2019/
[11] https://www.space.com/spacex-starlink-satellites-launch-just-beginning.html
[13]http://www.esa.int/Our_Activities/Space_Safety/ESA_spacecraft_dodges_large_constellation
[14]https://www.definitions.net/definition/rational+agent
[20] https://www.dmp.wa.gov.au/Documents/Safety/MSH_COP_SafeMobileAutonomousMiningWA.pdf
[21] https://www.mining.com/driverless-trucks-not-flawless-two-crash-fortescue-mine-australia/
[22] https://climate.nasa.gov/nasa_science/missions/?page=0&per_page=40&order=title+asc&search=
[24] https://airquality.gsfc.nasa.gov/
[25] https://www.nasa.gov/content/earth-missions-list
[x2] https://deparkes.co.uk/2017/11/24/machine-learning-vs-rules-systems/
[…] file:///C:/Users/20166004/Downloads/PREPRINT-DASC2011-AutomaticCollisionAvoidanceSystemDesignDevelopmentandFlighttests.pdf : automatic collision avoidance system