Advancing Mobile Navigation Accuracy: Lessons from the IPIN2020 Competition and VDR Technology

With the growing reliance on smartphones for navigation, achieving faster and more accurate positioning requires not only improved hardware but also sophisticated technical solutions. Positioning is the fundamental technology, and it has become a key research focus for the Amap (Gaode) technology team.

In the recent IPIN2020 competition (Track 6), Wuhan University and the Amap Intelligent Technology Center formed a joint team to explore vehicle navigation solutions that leverage built‑in smartphone sensors (GNSS, gyroscope, accelerometer, magnetometer). The competition tested outdoor open‑area, partially obstructed, and indoor satellite‑blocked scenarios, and the joint team won the championship.

IPIN (Indoor Positioning and Indoor Navigation) is the world’s largest academic conference on indoor positioning, established in 2010. Its accompanying competition is one of the three major global indoor‑positioning contests, alongside events organized by NIST and Microsoft, and is ranked among the top venues in the field.

The competition emphasized challenging real‑world conditions to test the algorithms' inertial navigation capabilities. Smartphone navigation faces significant challenges: GPS signals are vulnerable to non‑line‑of‑sight and multipath effects, leading to signal loss and drift, especially indoors (e.g., tunnels, parking lots). While vehicle navigation systems typically fuse sensors with GPS, smartphones are constrained by cost and sensor quality, making it difficult to meet vehicle‑grade accuracy.

IPIN2020 introduced a novel smartphone‑in‑vehicle positioning track. Participants received off‑site data (GPS, accelerometer, gyroscope, magnetometer, barometer) collected from a phone mounted on a car and were required to reconstruct the vehicle’s trajectory.

The test route spanned about 19 km, with the first 4.5 km used for algorithm initialization and the remaining 14.35 km for validation. During validation, GPS signals were intentionally blocked (simulating GPS outage) and indoor driving segments were included. In total, there were 19 GPS outages, the longest lasting over 400 seconds, forcing algorithms to rely solely on inertial navigation.

To overcome these challenges, the Wuhan‑Amap team leveraged big‑data mining and artificial neural networks, employing a relatively new VDR (Vehicle‑Dead Reckoning) core technology. By modeling vehicle motion patterns and calibrating sensor biases, they mitigated hardware limitations and achieved high‑precision post‑processing results, ultimately winning the competition.

Dead Reckoning (DR) combines IMU data, vehicle speed, and other cues to provide continuous positioning when GNSS is unavailable. Amap has long been a leader in vehicle‑DR, deploying it in hundreds of car models to ensure reliable navigation in urban canyons, tunnels, and parking structures.

Recent efforts focus on integrating smartphone sensors with big‑data and machine‑learning techniques to address persistent navigation pain points such as over‑pass detection and main‑auxiliary road recognition. By extracting phone attitude changes via VDR and fusing them with GPS and satellite SNR features, Amap has built learning models that improve localization in complex scenarios, many of which are now live on Amap Maps.

Compared with vehicle DR, smartphone DR faces additional difficulties: variable phone mounting positions, lack of odometry, and heterogeneous sensor quality across devices. Ensuring consistent accuracy across diverse smartphones remains an open research problem.

The collaboration aims to further enhance smartphone positioning in challenging environments, promising continued breakthroughs in high‑precision navigation.