Optibus co-founder and CEO Amos Haggiag explains how the AI methodology of deep learning can lead to more efficient mobility.
Citywide transportation is complex, especially with regard to the orchestration of millions of people travelling in a city, using multiple modes of transportation. An outside observer may conclude that some supercomputer is at work, determining where different vehicles and drivers go. This isn’t yet the case; but it may eventually happen when artificial intelligence (AI) transforms mass transportation.
Mass transportation is a huge market, estimated at around $1 trillion (€0.9 trillion) globally. It is at the core of every city in the world and forms an integral part of any functioning economy. Cities are investing billions in mass transport because they realise that better public transportation has many benefits, including increased economic opportunities and jobs, decreased air pollution and less congestion. Indeed, every £1 (€1.14) invested in public transportation generates £4 (€4.54) in economic returns.
Yet even though mass transportation is a huge market with a significant impact on the economy, it currently uses little to no advanced technology. At the same time, the mobility industry as a whole is undergoing a major transformation, one that includes autonomous vehicles, ride-hailing and ridesharing services; and micro-mobility vehicles such as scooters. The intersection of these two trends – a large market with limited or legacy technology and the changing face of mobility as a whole – lays the groundwork for a major shift in the technology used to plan and run mass transportation.
Driving the transition to more efficient mobility
Optibus is a software as a service (SaaS) company taking mass transportation networks – the core of mobility in cities – to the next level by providing the technology to plan and run mass transportation networks.
Optibus is driving the transition to effective, sustainable and efficient mobility and helping cities deliver what mass transport ought to look like: a system driven by data and by the ability to efficiently learn from and plan around that data – powered by artificial intelligence and advanced optimisation algorithms.
Headquartered in Tel Aviv, Optibus has offices in London, Dusseldorf and San Francisco, Optibus was founded in 2014 by CEO Amos Haggiag and CTO Eitan Yanovsky and has raised $54.5 million from investors including Insight Venture Partners, Alibaba, Verizon and Pitango Venture Capital.
Using AI in mass transportation to determine performance levels
To some extent, the future has already arrived. The mass transportation industry can already use deep learning, a form of artificial intelligence, to build data-based models and figure out the expected performance of a given transport network. This enables us to answer questions such as: given a certain bus route in a certain city at a certain time of day and day of the week, how long will the route take? How long will the bus stop at each station? How many people will be on the bus?
Deep learning is a method of learning similar to the way humans learn; like neural networks, but on a much larger scale. If humans were thinking about how long a certain route would take, they would need to take into account the time of day and the day of the week. We all know that the roads look different on Mondays at 8AM than on Saturday afternoons. Nourished by a deep trove of data about the movement of vehicles in a city, deep learning takes into account how long each route will take at different times; and each time period has its own model of probability.
We train the deep learning models on many months of data at a time and each city has its own models. The AI methodology helps us learn from the data, picking up patterns and predicting the likelihood of the on-time performance of each trip.
Building an optimal transportation network
Once we have AI models we embed them into the scheduling optimisation engine, so that we have automatic suggestions for new, more efficient schedules.
For instance, our deep learning models can show schedulers the likelihood that a given trip will start on time and end on time. If they can see that a large share of their buses are unlikely to stick to the timetable, they can optimise their schedules to improve on-time performance.
We use optimisation algorithms developed in house to take AI in mass transportation to the next stage. Our optimisation models enable us to provide the most optimal routes, timetables and rosters under the huge constraints imposed by the transport agencies and operators in over 300 cities around the world that use our technology. These constraints include budget, fleet size, vehicle type, number of drivers, union rules regarding driver breaks and working conditions; and countless other factors.
This helps us solve large scale optimisation problems very quickly, often in under a minute. We also use serverless computing and deploy our cloud native serverless technology on Amazon Web Services. This dramatically increases the speed of our optimisation, reduces the need for clients to have dedicated scheduling servers or support staff; and makes it easy for customers to do their planning and scheduling from any computer.
This mix of technology solutions – artificial intelligence combined with optimisation algorithms and cloud-native serverless computing – allows us to optimise transit easily even for a large city; we can add machines in the cloud on demand, so for a large city we may use 100 machines for just a few minutes to carry out the job.
Feeding data into AI models
In order to get results from deep learning models, we first need to supply them with the following types of aggregated data:
- Static data: This includes data about the network (routes, trips, stops, et cetera), as well as about operational rules and preferences, such as fuel costs and driver break frequency.
- Real time data: Automatic vehicle location (AVL) data can be used to understand how long a trip takes, how long a bus stops at each bus stop, the location and speed of a bus and other factors. We use that to improve the network over time.
- Demand data: If 5,000 people are going from one neighbourhood to another at 8AM, they will need a good transit network to get them there. We integrate demand data with location-based data that comes from cellular and location-based providers. Cellular signals, aggregated over time, show the movement patterns of people in a city. This helps us automatically optimise transit planning and scheduling, as opposed to the current manual practices pervasive in the industry.
Looking to the future of AI in mass transportation
We’ve already tackled how transportation systems can be planned using historical data. The next step is deploying AI to manage route planning and schedules in real time, to predict transportation demand and ultimately to orchestrate all the different types of transportation in a given city.
Today, planning and real time dispatching are two entirely different worlds. They are even carried out by different teams – planners on the one side and dispatchers on the other. Planners focus on creating a plan that will match demand, that will be the most efficient in terms of cost and feasibility. Dispatchers are focused on solving issues in real time. They don’t have time to find the most optimal solution, because they need to solve problems fast so that there will be a minimal impact on passengers.
Yet AI can easily check billions of options based on real time data and make smart decisions in the moment. Expect to see applications that use real time data about vehicle movement and demand in order to predict issues in real time, such as delays, demand changes or on-time performance issues. Then all the options will be evaluated, the constraints of the system and its rules of operation will be taken into account; and the best solution will be discovered and implemented.
Going beyond managing public transport schedules, AI could also be used to choreograph the entire transportation network – all the different modes of transport in a city – by making decisions relating to the movement of each vehicle in the network.
Imagine an AI brain that controls the city’s entire transportation network, with some vehicles running on a fixed route at fixed times and others whose operation is flexible and demand-driven. This system can then change timetables or routes, or decide to run different types of vehicles; all based on real time demand.
Co-founder and CEO