New AI-based navigation to enhance GPS accuracy

Chennai: Your navigation app may soon be able to tell not just which road your vehicle is on, but whether it is moving closer to the left edge, centre or right edge of the road. Researchers from IIT Madras and VIT Vellore have developed a new AI-based map-matching framework that could improve navigation accuracy on Indian roads by reducing GPS drift, lane ambiguity and positioning errors in dense traffic conditions. The project, showcased at the IIT-M Wadhwani School of Data Science and AI (WSAI) Annual Research Showcase 2026 on Monday, uses a ‘Lane Edge Integrated Hidden Markov Model (LEI-HMM)' that combines raw global navigation satellite system (GNSS) satellite data with lane-level road geometry to more accurately determine a vehicle's position on the road. According to the researchers, the system reduced positioning errors from decimetre-level deviations in raw GNSS data to centimetre-level accuracy in some test scenarios. "More number of satellites means greater amount of data, which reduces the error. But the second important thing is that we are implementing left-side and right-side kerb or lane information in the map matching, which is not usually available in Google Maps," said Subhojit Mandal from WSAI. "If this work is implemented, you will be seeing whether the vehicle is on the left side or right side of the road," he said. Raw GPS or GNSS locations often contain inaccuracies caused by satellite clock errors, orbit inaccuracies, atmospheric disturbances and signal reflections from buildings. Dense urban roads, tree cover and severe weather further worsen positioning errors. Mandal said existing digital maps themselves can contain inaccuracies of 2m to 5m, which could become critical in advanced driver-assistance systems. To reduce such errors, the researchers first built an accurate road network using Differential GPS (DGPS) measurements collected from multiple satellite signals. The Hidden Markov Model was then implemented using left-edge, right-edge and centreline information of roads. The team collected GNSS measurements from 84 trips across Chennai and Vellore using cars, autorickshaws and two-wheelers under different traffic conditions. The framework also incorporated historical movement patterns to reduce erratic jumps in location during traffic halts or signal drift. "What we are seeing in the data is that during static conditions the GPS data becomes haphazard because of satellite drifting errors," Mandal said. "The unique thing about this algorithm is that it tracks the historical information for the last five to 10 steps to predict the most probable road segment." Mandal said the framework could support real-time navigation, intelligent transport systems and low-cost lane-level positioning solutions for Indian roads.Prof Balaraman Ravindran, head of WSAI at IIT-M, said the annual showcase featured around 82 projects completed over the last year and aimed to help students present their work, interact with industry and receive feedback.