NAVISP asks ChatGPT about the PNT Trends

Last Updated: 23/02/2023 09:41     Created at: 22/02/2023 15:17

NAVISP asks ChatGPT about the PNT Trends

Artificial intelligence (AI) has been playing an important role for several years now, and the trend is accelerating. We seem to only slowly understand how far such technologies can take us. Particularly conversational tools based on AI have dominated the media in recent weeks. But can these tools actually be used to shape the navigation of tomorrow?

 

Global Navigation Satellite Systems (GNSS) offer tremendous value to society by providing worldwide, real-time services with precise timing and positioning information. In recent decades, GNSS applications have opened the doors to a wide range of applications, e.g., in the information and communications sector, the energy sector, agriculture, transportation, and many other areas. 

The information offered by GNSS is processed using precise mathematical models, with algorithms that have performed exceptionally well in supporting the provision of accurate Positioning, Navigation, and Timing (PNT) that can reach centimetre-level accuracy under optimal signal conditions. However, these models are based on assumptions and errors can occur with a detrimental effect on GNSS performance. The use of Artificial Intelligence (AI), and specifically Machine Learning (ML), can help to reduce such errors. In this context, the question occurs: Can Machine Learning and Artificial Intelligence help to tackle current PNT- challenges?

The term artificial intelligence (AI) comprises all techniques that enable computers to mimic intelligence, for example, computers that analyse data or the systems embedded in an autonomous vehicle. One mechanism that enables AI is machine learning (ML), and as the term implies, ML enables machines to teach themselves to learn. In ML, a relatively simple algorithm is trained to become increasingly complex, becoming better the more data it is fed. Since the early days of the Internet, this type of artificial intelligence has made great leaps forward.  In this context, ML can support, for example, the analysis of large amounts of data from different GNSS data sources. This can help to improve the overall GNSS performance, for example by modelling selected system segments using ML algorithms, thereby enabling GNSS system simulations (for more information on what AI can do in Space please follow this link).

AI has also made its way into society in various applications, starting with the use of various AI-based tools to create and generate art, codes or all kind of different applications. Recently, the chatbot ChatGPT gained a lot of attention, enabling the user to pose all kinds of questions from complex quantum computing to basic party ideas. 

ChatGPT is a natural language processing (NLP) Model chatbot, based on the language model GPT-3, that has been trained on an astonishing 175 billion parameters. The tool is considered a breakthrough in the application of AI due to its human-like responses to questions posed by the user, its contextual understanding, and large textual training data that consists of a variety of sources such as social media, newspapers, and more, allowing it to learn a wide range of language patterns and styles. A new conversational AI service called Bard, powered by its Language Model for Dialogue Applications (LaMDA) is expected to become a competitor of ChatGPT.

 

So, full of expectations we decided to see what ChatGPT knows about PNT and NAVISP and how does it compare with the knowledge of our experts. Overall, ChatGPT passed the test with flying colours, however it is important also to be aware of its limitations as potential biases could lead to systematic errors. Overall, our conclusion is that humans must continue to be involved in advancing the art of navigation, i.e., when it comes to the identification of innovative, out-of-the-box solutions which by definition could have not been part of the training set of the AI tool.

*NLP is a branch of artificial intelligence that enables computers to comprehend and generate natural language.
 

So, let's start with our test: 

ChatGPT, what is the difference between GNSS and PNT?

“GNSS stands for Global Navigation Satellite System and is a system that provides global positioning and timing services using a constellation of satellites. PNT, on the other hand, stands for Positioning, Navigation, and Timing and is a broader term that refers to the ability to determine one's position, navigate to a destination, and maintain accurate timing, often with the help of various technologies including GNSS. In other words, GNSS is a specific technology used for PNT purposes.”

And what are the future trends in PNT technologies?

The future trends in PNT (Positioning, Navigation, and Timing) technologies include:

1. Increased use of multi-constellation and multi-frequency systems for improved accuracy and reliability.
2. Development of new PNT technologies, such as quantum-based systems, to provide enhanced security and precision.
3. Integration of PNT with other emerging technologies, such as autonomous vehicles and the Internet of Things.
4. Increased use of software-based solutions, such as cloud-based PNT services and machine learning algorithms for improved accuracy and efficiency.
5. Continued efforts to mitigate threats to PNT systems, including jamming, spoofing, and cyber-attacks, through improved security and redundancy measures.

Since these initial general responses were satisfactory, it was decided to investigate what details were available on the various PNT trends. For this purpose, a free interview was conducted guided by the provided inputs from ChatGTP. 

 How can IOT be used to support PNT?

 IoT devices can support PNT through sensor data collection, improved communication and networking, data analysis, and   integration with other technologies. By providing accurate and real-time information on position, orientation, and velocity, IoT   devices can help improve the accuracy, reliability, and availability of PNT systems.

 

 






What are cloud based PNT services in comparison to common PNT services?

Cloud-based PNT services are a relatively new approach to providing PNT (Positioning, Navigation, and Timing) services, and they   differ from traditional PNT services in several ways.

Here are some key differences:

1. Data storage and processing: With cloud-based PNT services, data is stored and processed in the cloud, rather than on a user's device or a dedicated server. This allows for greater scalability and flexibility, as users can access the service from anywhere with an internet connection.
2. Cloud-based PNT services are often more accessible than traditional PNT services, as users can access them from any device with an internet connection. This can be particularly useful for applications that require real-time data and analysis, such as autonomous vehicles and drones.
3. Integration: Cloud-based PNT services can be easily integrated with other cloud-based services, such as data analytics and machine learning. This can enable advanced analysis and processing of PNT data and can improve the accuracy and reliability of the service.
4. Cost: Cloud-based PNT services can be more cost-effective than traditional PNT services, as users do not need to purchase and maintain their own hardware and software infrastructure. Instead, they pay for the service on a subscription basis, which can be more affordable for small businesses and startups.

Overall, cloud-based PNT services offer several advantages over traditional PNT services, including greater accessibility, scalability, and integration with other services. However, there may be some concerns about data security and privacy, as users must rely on the cloud service provider to manage and protect their data.

What can ML do to prevent jamming and spoofing of PNT services?

Machine learning (ML) can be used to prevent jamming and spoofing of PNT (Positioning, Navigation, and Timing) services by enabling real-time detection and mitigation of these attacks. Here are some specific examples:

1. Jamming detection: ML algorithms can be trained to detect patterns in PNT data that indicate the presence of jamming attacks. For example, a sudden decrease in GPS signal strength or an abnormal pattern of GPS signal noise can indicate the presence of a jamming attack. ML algorithms can be used to monitor these patterns and provide real-time alerts to users.
2. Spoofing detection: ML algorithms can also be trained to detect patterns in PNT data that indicate the presence of spoofing attacks. For example, a sudden change in GPS location or velocity that is not consistent with the user's intended path can indicate the presence of a spoofing attack. ML algorithms can be used to monitor these patterns and provide real-time alerts to users.
3. Mitigation: ML algorithms can also be used to develop mitigation strategies to counteract jamming and spoofing attacks. For example, ML algorithms can be used to predict the expected behaviour of PNT signals based on historical data and other contextual information. This information can be used to identify anomalies in real-time and adjust the PNT system's parameters to compensate for the attack.
4. Anomaly detection: ML algorithms can be trained to detect anomalous behaviour in PNT signals that may indicate the presence of a jamming or spoofing attack. For example, ML algorithms can analyse the statistical properties of the PNT signals and identify outliers that do not conform to the expected patterns. These outliers can be used as an indication of a possible attack.

Overall, ML can be a powerful tool for detecting and mitigating jamming and spoofing attacks on PNT services, which are critical for a wide range of applications, including transportation, logistics, and national security. By providing real-time detection and mitigation of these attacks, ML can help ensure the accuracy, reliability, and availability of PNT services.

Can you provide me with an algorithm for anomaly detection in PNT systems?

“Yes, there are several algorithms for anomaly detection in PNT (Positioning, Navigation, and Timing) systems. One of the commonly used algorithms is the Mahalanobis distance algorithm. Here is a summary of the algorithm […] This algorithm can be implemented using various programming languages and machine learning libraries, such as Python and scikit-learn. However, it is important to note that the choice of algorithm and its parameters may vary depending on the specific application and data characteristics. Here's an example code for implementing the Mahalanobis distance algorithm for anomaly detection in PNT systems using Python and scikit-learn library:

Can you also provide me an example code of this algorithm in python?

Figure 1: Code provided by ChatGPT on Mahalanobis distance

ChatGPT seemingly delivers very interesting results, while presenting more or less detailed insights into specific technological solutions. In this way, the user can not only gain additional knowledge, but also have the possibility to explore potential ways forward, be it in the technical design of a product or in the evaluation of technological trends. The next step of the interview was to find out if the tool is also able to provide specific insights into the European Space Agency's PNT activities. So, let's see what the tool has to say:

What is ESA specifically doing to develop and commercialize PNT technologies? 

“The European Space Agency (ESA) is involved in a variety of activities to develop and commercialize PNT technologies. Here are a few examples:

1. Developing the Galileo satellite navigation system: ESA is one of the key partners in the development of Galileo, which is designed to provide highly accurate and reliable PNT services to users worldwide.
2. Supporting PNT research and development: ESA provides funding and support for research projects related to PNT technologies through various programs, such as the Navigation Innovation and Support Programme (NAVISP).
3. Promoting the use of PNT technologies in commercial applications: ESA works with industry partners to promote the use of PNT technologies in a variety of commercial applications, such as autonomous vehicles and precision agriculture.
4. Supporting the development of PNT-related standards: ESA is involved in the development of international standards for PNT technologies, such as the recently adopted Galileo Open Service Navigation Message Standard.
5. Overall, ESA's activities related to PNT technologies aim to support the development of innovative and reliable solutions for a wide range of commercial and societal applications.”

 

I'm currently working on the development of an innovative method to prevent Spoofing of PNT services. Is there a programme of the European Space Agency which supports the development of innovative PNT technologies?

Yes, the European Space Agency (ESA) has several programmes that support the development of innovative PNT technologies, including methods to prevent spoofing attacks. Here are some examples:

1. Navigation Innovation and Support Programme (NAVISP): NAVISP is an ESA programme that supports the development of innovative PNT technologies. It provides funding and technical support to researchers and companies working on PNT-related projects.
2. Galileo: Galileo is the European Union's global navigation satellite system (GNSS), which provides accurate and reliable PNT services. Galileo has advanced features designed to improve security, including encrypted signals that are resistant to spoofing and jamming attacks.
3. GNSS Evolution for Innovative Road Surface Applications (GENIORS): GENIORS is an ESA-funded project that is developing new methods to improve the accuracy and robustness of PNT services, particularly for road surface applications. The project is also exploring ways to prevent spoofing attacks.
4. Precise Time and Time Interval (PTTI) Technology Demonstration (TDE): The PTTI TDE is an ESA programme that supports the development of advanced PNT technologies, including those focused on time and frequency synchronization. The programme provides funding and technical support to innovative projects, including those focused on preventing spoofing attacks.

By leveraging these programmes, you can access funding, technical support, and expertise to help develop your innovative method to prevent spoofing of PNT services. The ESA is committed to supporting the development of innovative PNT technologies, and there are many opportunities for researchers and companies to get involved in this exciting field.

Careful reflection on this response reveals that answers must be treated with caution. The tool mixes information on funding programs with technical development activities. In addition, the GENIORS initiative mentioned in the third bullet point is not a European Space Agency initiative, nor does it even exist. Similarly, when asked about the Navigation Innovation and Support Program (NAVISP), the tool got confused with the abbreviations and returned different results to the question of what NAVISP is. Only after several iterations and manual corrections in our chat with the tool a satisfactory definition was reached.

 However, we noted that this definition was not permanently incorporated in the database of the tool yet as a separate user was still getting the wrong answers.

This showed us, that ChatGPT like any other AI algorithm, can be sometimes biased, resulting in unreliable answers. The AI model may provide inaccurate or incorrect answers due to a knowledge cut-off, where it was only trained on data up to the year 2021. As a result, it may not be able to provide information on events that occurred after this date. This limitation can, be counteracted by feeding the algorithm with new data by giving it feedback during the individual chat sessions (fine-tuning), as happened in our questions on NAVISP. Fine-tuning is a supervised machine learning technique, which trains the model based on labelled data (the user feedback in this case). This process adjusts the weights and biases in the neural network through 'backpropagation' to improve the accuracy and performance by reducing the error rate of the output. In this way, ChatGPT gives more accurate answers based on the provided feedback to the algorithm without the need of retraining the original training dataset. Or how ChatGPT worded it, “This is like how a person might learn something new based on feedback or information received, without needing to be completely retrained on a task.” 

In conclusion, AI applications like ChatGPT can be an asset in research of current PNT innovation efforts, however, it is important to be aware of its limitations, such as potential biases that could lead to systematic errors, thus compromising the reliability and accuracy of PNT systems. In addition, there is a risk to rely too heavily on patterns in the training data, resulting in incorrect, unexpected scenarios that can be adjusted by fine-tuning the algorithm to match the user's feedback. To put this latest technology hype into perspective: the usefulness of ChatGPT will not replace expert knowledge and with its ambiguity and bias, there is still a long way to go before it can effectively deliver qualitative outcomes for PNT innovation. Humans must continue to be involved in determining technological trends in PNT, identifying technological gaps for advancing the art of navigation, and helping European industry maintain competitiveness in the PNT market. ESA, its staff, and in particular ESA's Navigation Innovation & Support programs are therefore dedicated to this area by leveraging these AI technologies in the most appropriate way.

To date, NAVISP has supported the development of several innovative ideas for applying machine learning and artificial intelligence in the context of PNT. An exemplary list of current ongoing and completed NAVISP activities can be found below:

ID	Title	Aim
EL1 020	AI/ML Sensor Fusion for Autonomous Vessel Navigation	Study AI & ML techniques for the combined use of multiple sensors in maritime PNT receivers for autonomous vessel navigation and implement a proof-of-concept prototype autonomous navigation PNT receiver using the techniques identified
EL1 034	AI-enabled baseband Algorithms for High-fidelity Measurements	Establish a new paradigm in the design of GNSS algorithms, leveraging on artificial intelligence and fed by collected data in field trials; Design algorithm to provide high-fidelity raw measurements, along with quality indicators related to local environments.
EL1 035	Machine Learning Techniques Model GNSS	Assess usability of machine learning techniques in support of GNSS system simulations by modelling selected system segments using ML algorithms.
EL1 038	Application of Machine Learning Technology for GNSS IOT data fusion	Validate the application of Machine Learning, Big Data and Data Mining techniques to vast amounts of data resulting from diverse GNSS data sources for a set of use cases (e.g., improved tropospheric / ionospheric characterisation and earth's magnetic field mapping).
EL1 053	Real-Time Big data Processing for GNSS Integrity	Demonstration that crowd-sensed GNSS signals can be effectively used as the sole (or main) data source for an integrity service.
EL1 056	Advanced Algorithms and Techniques for resilient time provision	Combine different time sources with advanced algorithms, in particular Kalman filtering (in phase and frequency) and Machine Learning (ML) techniques, in order to provide the required level of time accuracy, stability, availability and integrity to timing users.
EL1 064	Block-Box for an optimised GNSS Spectrum Monitoring Network Using AI	Design of an external RF2RF GNSS receiver enhancement device providing local GNSS threats and system anomalies detection and classification based on an on-bard implemented Artificial Intelligence engine and GNSS threats mitigation, signal cleaning and retransmission at Radio Frequency as well as Cloud data processing and sharing, enabling fine tuning of the AI/ML algorithms.
EL2 128	Monitoring and Estimation of Drone’s Passage for augmented Security in Critical areas	Application of innovative algorithms to the carrier phase measurements of the GNSS signals to detect phase anomalies due to the passage of the drone. These core algorithms are applied on all the deployed GNSS sensors’ mesh and, the detection results are then sent to the Control Centre where innovative techniques (e.g., Machine Learning) are applied to the detection information to estimate the drone trajectory.
EL3 005	Trondheimsfjorden Test Area	Development of infrastructure and methods for testing and verification of technologies and concepts including positioning technologies, advanced sensors, machine-learning, artificial intelligence and improved connectivity (Internet of Things)