Artificial Intelligence (AI) in transportation has never been a new topic. The evolution with the integration of the concept started decades ago, but in the recent 10-15 years, the exponential growth has precipitated a transformative shift in the industry.
Once a daily battleground for commuters and pedestrians alike, several solutions have been proposed by placing deep learning into the network and feeding it with existing factors to make predictive analyses to overcome the challenges faced.
Transportation, despite its impressive growth, has long struggled with enduring issues such as pedestrian safety concerns, traffic congestion, and prolonged travel times. The 2023 Global Status Report on Road Safety by WHO reveals that the annual traffic fatality count remains approximately 1.19 million.
Bloomberg reported drivers lose 156 hours on average stuck in traffic in London in 2022, listed as one of the most congested cities in the world. Residents of Bogota and Rio de Janeiro, a few other extremely congested cities, as per the Inrix Report 2019, lose around 190 hours in traffic, prolonging their travel time, according to the World Economic Forum.
Add Bangladesh to the discussion; Dhaka city easily finds its place on the leaderboard of traffic jams. National daily Dhaka Tribune reported approximately 25,000 road traffic fatalities annually in Bangladesh, resulting in a 5.3 percent loss in GDP. The nation’s road safety is highly concerning, with 3,502 deaths and 3,479 injuries reported from 3,701 road accidents in the first eight months of 2021 alone, making the country 103rd out of 183 countries for having the most road accidents.
Now, back to the point we want to discuss: how can AI help us mitigate them? For many, the discussion is hard even to start. So, another question pops up: is it actually possible to implement AI in a densely populated country like Bangladesh, or more specifically, in Dhaka?
Singapore, a country characterized by dense population and land constraints like Bangladesh, focuses on using autonomous self-driving vehicles as first and last-mile traveling complementary means augmented by ride-sharing and car-sharing services, keeping public transport the main for mass commuting. Autonomous bus is one of the initiatives.
China also holds the rank for owning some of the most congested cities in the world. Suiteng OS is the rail transit industry’s first new-generation intelligent rail transit operating system based on the Industrial Internet and the Internet of Things (IoT) launched in the country. It utilizes intelligent perception, data interaction, intelligent analysis, and linkage operation of all equipment systems in the station. This has allowed a lot of features such as ‘smart security,’ enabling users to pass the device security check-in system in less than 2 seconds, for instance.
Huawei in China provides the intelligence transportation solution ‘TrafficGo’ as a leading innovation to improve traffic efficiency in certain areas such as Bantian and Shenzhen. The average delay in the main direction of Shangdi’s third street is said to have decreased by 25.2 percent, which led to the average delay of surrounding roads decrease by 10-20 percent, as stated on the TrafficGo official website.
Many companies affiliated with transportation facilities have since merged machine learning into their platform to achieve better performance. To present a well-known name, Uber, often referred to as an AI-first company, established its de-facto machine learning platform called Michelangelo in 2016, training their models to predict ETA (Estimated Time of Arrival), provide support to customers with COTA (Customer Obsession Ticket Assistant), and detect frauds with RADAR. This automatically closes the gap between demand and supply to a great extent, exchanging routes to the shortest possible ones and thereby reducing travel time. Altering routes in such a way also helps tame the congestion during peak hours.
Today, a few AI-based approaches have already been taken in Bangladesh. Dhaka North City Corporation (DNCC) introduced an AI-based traffic management system at several points, including the Gulshan-2 intersection, in June 2023. The system was intended to monitor all vehicular movements and file cases due to traffic violations automatically upon detection. It was expected to reduce traffic violation incidents by 99 percent. However, no case was ever filed based on the AI camera footage during the eight-month running time of the pilot project, effectively showing the effort failed.
Only DNCC officers had access to the AI-camera generated videos, not the traffic division of police responsible for managing the traffic system, leading to utter confusion and ultimately no relevant use of the resource at the location. Drivers were unfamiliar with the setup, and traffic police were too accustomed to log cases and control traffic manually; all of these contributed to the haphazard case, leaving the system as ailing as before.
Section 3.5 of the National Strategy for Artificial Intelligence Bangladesh (2019-2024) and Section 4.5 of National Artificial Intelligence Policy 2024 underscore the potential of AI to optimize traffic systems, ensure road safety, real-time monitoring, predictive maintenance in public transportation, and adoption of electric and autonomous vehicles in the country.
Nevertheless, referring to previously failed projects and the current policy, Sumon Ahmed Sabir, Chief Technology Officer of Fiber@Home Ltd., argued that a relevant and effective national data strategy consisting of collection, preservation, and synchronization should precede the implementation of such an AI strategy. He opined that without clean and accurate data to train, validate the parameters, and evaluate the AI model’s performance, the AI initiative’s likelihood of success will fall close to a low percentage.
Besides a robust data strategy, improvements should be made to the internal issues of the network itself. “Dhaka city is not yet fit for AI integration into the transport system,” remarked Dr. Md. Shamsul Hoque, a road communication expert and professor at BUET. Heterogeneity in vehicle types and traffic flow causes the complexity in AI prediction.
Some of the vehicles on the street do not even have standard PCUs. Auto-rickshaws, three-wheelers, and human haulers (leguna) are analyzed through speed area, headway, and simulation methods to derive a PCU for the respective vehicle. Nevertheless, a single value is yet to be recognized officially by the Roads and Highways Department of Bangladesh.
Drivers with impulsive behavior, home-grown traffic rules favoring VIP movements, and political rallies add more to these challenges. Moreover, substantial investment is required for sensors and cameras for data collection and processing, which can be quite cumbersome for a developing country like Bangladesh. Skilled professionals, preferably local experts, are also required, who would be more familiar with the nature of traffic and driver’s unpredictable behavior than foreign consultants.
Stating such problems, Dr. Md. Musleh Uddin Hasan from the Department of Urban and Regional Planning at BUET remarked, “Without solving these internal problems within the network first, AI implementation would not be feasible.”
Dr. Md. Shamsul Hoque also discusses the Dhaka Cantonment system for its efficiency in utilizing traffic signals and lane-oriented roads. A properly maintained network system like that would simplify much of the complexity that arrives in AI prediction.
There is no denying that Bangladesh will eventually have to adopt AI. However, the pre-feasibility assessment remains the most crucial element. Where to implement, the scale and forms of implementation, how long it will take, and the development of infrastructure —all these need to be well assessed before taking any drastic measures.
