The Multi-cam Multi-map Visual Inertial Localization (VILO) system is a high-performance, real-time localization solution designed for robotics. Unlike traditional VINS and SLAM systems that accumulate drift or rely on delayed corrections, VILO provides drift-free, causal localization directly integrated into the control loop, ensuring precise, instant feedback for autonomous operations in large, dynamic environments.
-
Real-Time Localization and Mapping
VILO provides accurate, drift-free pose estimates with support for both online and offline mapping. Real-time online mapping optimizes data collection on the fly, while offline processing enables the generation of high-precision, dense 3D maps through two-stage optimization. -
Robust Multi-Cam Visual-Inertial Odometry (VIO)
Utilizing multi-camera and IMU data, VILO’s VIO module includes resilient initialization and feature matching to maintain accurate localization even in challenging environments with high outlier rates. -
Causal Evaluation Metrics
VILO introduces tailored, causal evaluation metrics that assess localization accuracy in real time, eliminating the need for post-processing and enabling immediate feedback for live navigation and control.
The VILO system is a multi-camera, multi-map visual-inertial localization solution designed for real-time, drift-free localization in dynamic environments.
Captures multi-sensor data in real-time, providing immediate feedback for optimized map creation. Offline processing then generates high-precision, dense 3D reconstructions.
Uses pre-built maps for accurate, consistent localization by combining multi-camera VIO with robust feature matching and outlier rejection, ensuring drift-free performance across diverse environments.
We offer a robust, nine-month dataset from Zhejiang University’s Zijingang Campus, spanning over 55 km and diverse real-world conditions. The dataset includes synchronized surround-view cameras, IMU, LiDAR, GPS, and INS data, making it an ideal resource for long-term testing of localization accuracy and robustness.
Multi-sensor setup with synchronized cameras, IMU, LiDAR, GPS, and INS for robust data collection in dynamic environments.
Custom synchronization module ensures unified timestamps across sensors, critical for precise mapping and localization.
The dataset includes nine months of multi-sensor data from Zhejiang University’s Zijingang Campus, covering 55 km under diverse environmental conditions. This dataset provides a robust basis for testing VILO’s performance in real-world, long-term scenarios.
Two front-facing stereo cameras and two fisheye cameras (left and right front) capture a wide field of view.
A synchronized LiDAR-camera setup provides detailed spatial data.
Captured across multiple sessions, the dataset includes seasonal and environmental changes, showcasing differences in lighting, weather, and structural modifications across campus scenes.
Access the VIO part's code from Multi-cam-Multi-map-VIO.
Access the full relocalization code from Baidu Netdisk.
Password: 85ym
Download the complete dataset from Baidu Netdisk.
Password: uu8t