Perfecting mapping with AI and machine learning
Across the world, mapping technology with Artificial Intelligence (AI) and machine learning allow users to have a variety of choices on their travels. Be it driving, flying, or walking, GPS systems are now a lifesaver in keeping users on track. Before this, most of us often used old maps or would buy travel maps whenever we wanted to move around.
Today, map applications are not only available on GPS devices, but also on our mobile phones and are even built into our vehicles to provide better route directions. Despite this, there are still some challenges when it comes to mapping and location tagging.
Some of the general map applications we have today include Waze and Google Maps. Companies like Garmin are also known for their wearable GPS devices which are often used not only by fitness enthusiasts but also pilots for private air travel.
In the logistics and supply chain industry, mapping technology is used to position the movement of autonomous machinery and vehicles while the e-hailing and food delivery industry also rely purely on maps and geospatial data.
According to Philip Kandal, Head of Engineering, Geo at Grab, there are three main challenges in mapping today that affect almost anyone – coverage, accuracy, and freshness. With a myriad of applications relying heavily on mapping services, Machine learning and AI are enabling fresher and better maps, which is also a unique opportunity for innovation.
“Compared to other parts of the world, Southeast Asia has a very diverse ecosystem with transport from scooters, bikes, motorcycles, cars, vans. Mapping all of this is highly complex and would be impossible without the help of machine learning and AI. Maps in more developed markets are often focussed on a car-centric society.”
“In this region, we need to rethink this from the ground up given the massive modes of transportation available,” said Philip during his presentation at the AI Accelerator Summit in Singapore.
Philip believes that machine learning can make operation teams 10 times more productive with the use of various techniques such as robotic process automation, computer vision as well as natural language processing (NLP).
Working with Grab, Philip explained that the company’s geo services serve over 800 billion requests every month. Developing a service model at that scale without machine learning and AI capabilities is not possible, especially with the uniqueness of the transport methods in the region.
“Be it picking passengers or food delivery, guiding drivers and riders is very complex in the region. With a variety of vehicles to deal with, we need to look at an optimal way to get things done. Different sets of rules apply to different vehicles. For example, in food delivery, it’s about finding the best and fastest route to get the delivery from the kitchen to the person’s home” said Philip.
The challenges of mapping
When Google introduced its Google Car several years ago to take pictures of streets for its Google Maps application, the world was wowed by this. Unfortunately, some of the areas covered by the vehicle have since changed physically but have not been updated in their maps.
The three challenges of mapping – coverage, accuracy, and freshness, need to be perfected.
In coverage, the mapping needs to be right, especially in identifying roads, points of interest, and such. Computer vision technology such as optical character recognition can be used to enhance surroundings and routes for greater coverage.
For accuracy, GPS in urban areas are often off by about 50 meters. Businesses need to get this down to just five meters as in some areas, a location off by 15 meters, can make a huge difference. Machine learning algorithms can be used to cover areas with weak GPS signals. In some larger areas like malls, for example, entrances can be used as georeferencing for pickup points.
To ensure the freshness of maps, they need to be updated within days. Analyzing traffic data for changes in traffic movement, route changes, etc generates more data that can be analyzed and updated to route systems to ensure maps are updated as soon as possible.
“At Grab, we use AI to analyze several billion trips to get the best insights. We have petabytes worth of data and in its unstructured form, helps up learn and improve maps a lot more,” added Philip.
At the end of the day, the use of AI and machine learning in mapping is primarily to create a seamless multi-modal navigation experience for users. Be it for e-hailing, delivery, traveling, or any other use case, new technologies enable maps to be constantly updated.