The share of AI in transportation is increasing annually by 15.8% and by 2025 in financial terms it will reach $ 3.8 billion. Humanity needs the most comfortable and safe transportation, and only the latest technologies can guarantee them. Sirin Software decided to share their vision of how transportation software development can transform our lives and your business.
Artificial intelligence in transportation: promising but difficult
Self-driving cars, smart parking lots, traffic lights, lane change assistants and smart energy systems are under development. A new class of smart transport systems (ITS) is emerging to connect vehicles and infrastructure.
ITS help automate processes, thereby achieving:
- improving the quality of service;
- increased security;
- optimization of routes;
- minimization of downtime;
- simplified flow control and maintenance of parks.
Many governments are funding AI in urban mobility projects. In Canada, it is the ACATS program offering large grants to companies for communications. And in Singapore, there are five projects at once, including “intelligent planning of cargo transportation”, which will be available for us by 2030.
Pilot tests of ITS have been successfully carried out in many countries, but have not yet reached commercial implementation. This is because scaling AI in the transportation industry still requires a really delicate balance between benefits and challenges.
Roadmap for the implementation of AI-based transportation
There is no one-size-fits-all algorithm, but there are steps you can take that will help determine the direction of your product.
Develop a data management strategy
The introduction of artificial intelligence in transportation is creating a huge number of big data sources. You need to prepare to manage them. Create an algorithm for processing, protecting, standardizing and using incoming information.
For scalable AI-based transportation algorithms to quickly and efficiently cope with their tasks, you need:
- A representative and safe data collection process.
- Enough datasets to train and validate the model.
- Low latency data processing between connected systems.
- Methods for consolidating and implementing information in different ways and formats.
Almost 95% of AI in urban mobility projects are unsuccessful due to the inability to collect, store, analyze and use data for decision-making.
Models still have a reputation for being non-interpretable black boxes, which means users can simply ignore predictions. To avoid such an issue, it is necessary to highlight the most important factors in prediction and validate the models. Also, be prepared for their instability and constant readjustment and customization.
Companies are adopting artificial intelligence in transportation based on feasibility, financial and technical capabilities. Accordingly, the tasks are implemented in different ways. The IEEE has published a study highlighting the capabilities of popular data sources.
Legislators are also working on providing better source access, standardizing public data formatting, and opening data API integrations.
For example, the US Department of Transportation implements three separate programs for this purpose:
- Safety Data Initiative.
- Waze Pilot.
- Computer vision tools.
Explore the possibilities of AI-based transportation methods
AI is not a copy of the brain, but just an attempt to use its individual abilities, expressed in mathematical form. The solutions will be different for different applications. In general, it is a collection of supervised, unsupervised, and reinforcement learning methods for systems:
- Artificial neural networks (ANN);
- Genetic Algorithms (GA);
- Simulated annealing (SA);
- Fuzzy Logic Model (FLM);
- Ant Colony Mimic Optimization (ACO).
Each AI in urban mobility method is limited. ANNs are better suited for predicting traffic demand, infrastructure maintenance, and monitoring driver behavior. ACOs are effective in optimizing transport routes and passenger flows. But scaling “intelligence” is difficult, because the speed of algorithms drops sharply with increasing repeatability.
Thus, no method of transportation software development gives a universal result. This requires compromises with insufficient quality and functionality, or increases development time.
Focus on the explainability of the model
Artificial intelligence in transportation makes vital decisions that are difficult for humans to trust. For example, speed limiting reduces the risk of accidents, but lengthens the path to parking and increases traffic in tight spaces. You can let the ambulance pass and save someone, or block the route for the fire truck to pass to the blazing building. It is difficult for people to decide, but how can a small iron box do it?
The ethical dimension of AI in the transportation industry is an acute issue. Explainable Artificial Intelligence (XAI) in the future will be able to explain decision-making mechanisms. Developers can control the accuracy and fairness of the model.
The explicability of the model can be considered at all stages of transportation software development. Both for interpreted (linear and logistic regression, decision trees) and for black box models (perceptron, CNN, RNN, LSTM).
For models that are difficult to interpret, the most popular a posteriori explanations (post-modeling explainability) are LIME, SHAP, and LRP. Since AI in urban mobility systems are not self-learning, a trade-off in black box thinking levels may be required.
Collaborate with partners
Competition in the transport industry leads to problems:
- Low compatibility of disparate systems;
- Lack of uniform standards for data, quality, integration, etc.;
- Difficult access to basic infrastructure.
To realize the benefits of AI-based transportation, specifications need to be standardized:
- Data structure, content, exchange protocols;
- Definitions and classification of types;
- Methods for creating, sharing and using data.
Transportation software development needs to be driven with future regulatory requirements in mind to keep applications competitive in the long run.
- ISO / TC 204 ─ regulation of systemic and infrastructural aspects of ITS;
- TC 241 – Road Safety Rules and Management Standards;
- INSPIRE-MMTIS – access to static and dynamic data for multimodal tourism information services;
- Auto-CITS – unified autonomous driving standards and the introduction of V2X technology in the EU.
When evaluating technology, it is important to consider the benefits of partnerships. Determine if the investment will only benefit your business or open up opportunities for cross-integration. In other words, without the joint efforts of all interested communities, large-scale implementation of technologies is futile.
Towards AI in the transportation industry without a doubt
Technology is the future, but it is not so easy to implement. For deployments to be able to scale, it is important to lay the foundation. Contact Sirin Software for AI in Transportation Consulting, and make your project a profitable and attractive investment now and for years to come!