GenAI in ITS

Introduction: GenAI pillars

The origins of GenAI can be established by the birth of the concept of Artificial Intelligence (AI), which can be defined as programs that sense, reason, adapt and act. It is an overall framework which encompasses everything in theory and development of computer systems to be able to perform tasks normally require human cognition and intelligence.

We may consider that the origins of artificial intelligence itself go back to the mid-20^th century, thanks to the British logician and computer pioneer Alan Mathison Turing. The computer conceptualized by Turing consisted of a stored-program concept, where the machine had the possibility of modifying or improving, its own program. Turing often discussed how computers could learn from experience as well as solve new problems through the use of guiding principles-a process now known as heuristic problem-solving.

Another basic ingredient for reaching today's GenAI is Machine Learning (ML), which focuses on the "adapt" part of AI, using algorithms that improve the performance of the program where it stands by feeding and learning from increasing amounts of data over time, without the need of being explicitly programmed.

The third concept that has been highly developed as part of the Machine Learning algorithms is Deep Learning, in which Multilayered Deep Neural Networks learn automatically from vast amounts of data, adapting from experience in each iteration. By itself, Deep Learning has evolved from being initially a discriminative model to a Generative Model where they learn from the probability of the distribution of data from the data they receive as an input, allowing them to generate a new data instance, based on the logic of the data they have learned.

Finally, we arrive to the Generative AI (GenAI) which uses a subset of Deep Learning techniques to generate new content, after learning through a massive amounts of data and long training periods. The most profound difference between GenAI and traditional AI is the former's ability to operate without the need for labeled examples. Instead, GenAI learns patterns and structures from data and uses this understanding to generate entirely new content. This includes everything from textual responses and artworks to new musical compositions and realistic human voices.

Applying GenAI in motorway management

The possibilities for GenAI are endless, and we have just started exploring. A new era is already here, which offers great opportunities for improving our lives completely for all sorts of different use cases both in professional and personal life.

There are many sectors and topics that can be discussed to explore GenAI's applications, but on this occasion, I would like to focus on its application in the sector I have been working on for more than 18 years: Interurban roads. In the following section I have come up with a short list of possible use cases where GenAI could be applied to, at interurban roads, highlighting the transformative benefits it will provide.

1. Video analysis: incident detection

In the Intelligent Transportation Systems (ITS), which are the systems that are commonly implemented at motorways (SOS phones, CCTV cameras, variable message signs, vehicle detection systems, etc.), it is also pretty common to implement Automatic Incident Detection (AID) systems based on cameras, which are typically installed at critical infrastructures and certain road sections as in tunnels for example. These systems have been being applied for more than 20 years, incorporating new technology improvements throughout time, not only by using better cameras but also by introducing AI and ML components into the incident detection algorithms, trying to reduce false incident alerts reported to the operators usually caused by light reflections, birds, or shadows for example. Nevertheless, even with the introduction of these improvements, many of these incident detections are still limited, still have a non-minor amount of false positive alerts, and therefore still require verification by an operator before taking the corresponding action.

One of the features GenAI introduces is the Large Vision Models (LVM) which refer to advanced artificial intelligence models designed to process and interpret visual data, typically images or videos. These models are "large" in the sense that they have a significant number of parameters, often in the order of millions or even billions, allowing them to learn complex patterns in visual data.

One of the characteristics of AID systems is the fact that the use cases have to be programmed one by one, covering many conventional, frequent, and critical use cases. Each one is implemented in the system in a specific manner, with a specific set of parameters and limitations, leaving many other use cases uncovered, meaning that there are many possible use cases that may be critical or important but are totally missed by the AID system as they do not fit the specifics of any of the limited number of programmed use cases.

In case of applying GenAI algorithms to the same video sets, the system will be able to understand the video as a whole, identifying all types of abnormalities, not trying to make them fit into one of the use cases within a discrete list, but from a broader perspective, as an operator would do, being able to identify anything which does not seem right in a specific video, due to a whole set of possible reasons, e.g., Being a motorway there are other things apart from vehicles on the road (pedestrians, animals, lost cargo); detection of hazardous driving: sudden lane change, or abrupt change of speed. It will also easily be able to identify and overrule birds, shadows, or reflections, for example, quite easily, eliminating a greater range of false positives.

Once an abnormality is identified, the GenAI system will be able to classify it by the severity level and take the required actions automatically or create a detailed report in the best format, including all necessary information for an easier inspection and understanding on behalf of an operator. This information could include, for example, the number of vehicles implicated, vehicle brands, models, and colors, the number of people in the scenario, if there are any children or old people, if any of the people are the police or the fire brigade, road physical conditions as if there are any water or oil spills for example, and it all can be delivered in writing or even backed up with a stetch, image, or video creation of its own to ease the comprehensiveness of the operator.

It is important to remark on the word "understand": While an AID system may detect several independent events: Smoke, slow vehicle, lane change, vehicle driving in the shoulder, or stopped vehicle; the GenAI system is actually able to understand the video scene, matching all the incidents together as an entire action, start to end, probably reporting the sequence of events mentioned before as "vehicle breakdown" with an overall comprehensive approach. This is possible for GenAI after learning through massive amounts of data about what is and what is not "normal," as well as taking into consideration complementary information collected from many different sources, providing GenAI with a wider view and understanding of the world we live in, accomplishing higher levels of understanding when analyzing a scene and after being able to provide a highly detailed description in the required format that almost guarantees the comprehension of the report.

1. Video analysis: Prevention

Using the same GenAI algorithms across vast amounts of historical and real-time videos, GenAI will be a massive game-changer for preventing or minimizing congestion, ensuring smoother traffic flow or even accidents, and increasing road safety. The GenAI algorithms will be able to identify patterns that lead to congestion or accidents before they even happen, being able to take actions in real-time such as adjusting speed limits or writing specific information messages by acting through variable message signs or adjusting the traffic lights timings in ramp metering systems controlling the input and output flow of traffic into the motorway.

GenAI will have a broad view and understanding of the entire motorway and its surroundings, being able to make smarter decisions as a whole system that will work end to end, not only in a specific location, taking into care of the video feed data (historical or real-time) but massive amounts of other data which will condition its decisions and optimize the output as the time of the year, the time of the day, weather conditions, types of vehicles in the area (cars, trucks, emergency, etc.), traffic actual conditions and expected evolution in the current or surrounding roads or towns and many more.

Other types of systems that GenAI could be able to act upon, if available on the motorway, would be sending information messaging to users directly through radio, SMS, mobile APPs, or even directly to the vehicle for connected and autonomous vehicles through specific roadside equipment, providing them with traffic information and suggestions about speed, lane where to drive, best route options, etc. It could also send alerts to users before they initiate a trip about the current situation and the best available options: Routes, public transport, best times to travel, etc. before they get on track, all with a higher level of understanding and information transmitted to what similar systems may provide nowadays.

1. Video analysis: Simulation

Another powerful tool that GenAI will improve hugely will be simulation. Through a written or verbal chatbot, the operator will be able to describe a specific traffic scenario and ask about the best way to manage the situation, either to prevent it, optimize it, minimize the impact, increment the security, or all together. The GenAI algorithm will be able to provide a list of options with their pros and cons, considerations, and even a recommended option. Again, and as mentioned with other cases above, the fact that the GenAI will be taking into care of massive amounts of historical information and a great number of extra data and affecting parameters will provide the operators with the best possible plan execution for a given situation.

As an example of what type of situations may be interesting to be requested for analysis, here is a short list:

• Origin - destination route for transportation of dangerous goods or large cargo
  System output: Required documentation, best date and time, speed, specific route, transportation and road signalling, measures in case of an accident depending on the goods being transported, etc.
• Planning for specific days with high traffic expectancy: Festivities, holidays, concerts, etc.
  System output: Road messaging and signing at different dates and times, recommended speed limits, required traffic management: where and how, etc.
• Planning maintenance works: lane closures, deviations, speed limits, etc.
  System output: Road messaging and signing at different dates and times, working area best signalling, workers' safety measures, work efficiencies to work smarter and quicker, etc.

In all cases above it will be required but quite easy to add specific possible extras as existence of work zones, occurrence of accidents, different levels of heavy vehicles, etc. to be considered by the GenAI as part of each scenario.

1. Maintenace and technical support

In this case, the application is not specific only to roads and can be applied to all sorts of maintenance and technical support. Thanks to GenAI, maintenance workers or technicians will not need to rely on their organization skills, remember what document they need for a specific job, etc. Thanks to GenAI, it is very straightforward to implement a simple chatbot, which can be used either in writing or verbally, where the workers can have a conversation in any language, with the interface where the GenAI engine will be able to explore vast amounts of unstructured data delivering the required information, from the correct document, in the correct page in a fraction of a second!

The same type of chatbot can also be applied as a user interface for technical support, where the GenAI algorithms can provide a comprehensive conversation, helping solve the problem using internal documentation and even complementing Internet information. Moreover, it can conform a highly detailed ticket for the technicians to be able to generate custom-made schematics, drawings, representations, images, etc., pre-contrasted with the user and attached together with a thorough description, table with all the user and device or devices information for an easier understanding of the problem, even being able to list the required list of expected tools to be used to solve the problem.

In summary, the introduction of GenAI algorithms into our lives will introduce a great change for the better, improving and optimizing our lives in all professional and personal aspects thanks to its power to analyze and process millions of data from thousands of different sources. Specifically in the interurban roads sector, there are a whole set of use cases that will not only help the operators and maintenance people improve their work but will introduce automatic real-time analytics and responses to many real-time situations to a level no human could do, with more precise and studied outputs that will help to make our roads safer, improving the traffic flow, optimizing our journeys, preventing accidents and who knows what many other wonders which may come in a short future.

GenAI has just arrived and is here to stay, turning our world around, but it is a coin with two sides; on one side, it will help us improve our lives, but it will also introduce new scams, identity theft, and many other threats which we never thought existed nor had to worry about, so it is good to take advantage of the new GenAI algorithms in our benefit but even more important to learn and be aware of the new dangers we will have to tackle which will make us a lot more vulnerable.

Public link

Please use the above public link if you want to share this noodl on another website.