VILAM: Infrastructure-assisted 3D Visual Localization and Mapping for Autonomous Driving
Talkat 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24) · 2024-04-16..04-18·
summary
VILAM claims decimeter-level localization and mapping accuracy using only consumer cameras by fusing roadside LiDAR measurements as global references. It implements elastic point cloud registration to correct irregular deformation in vehicle SLAM maps and a lightweight factor graph optimization to propagate corrections across coverage gaps. Real-world testbed on smart lampposts runs over 17 km with 40 infrastructure nodes, achieves map/trajectory correction in under ~300 ms, maintains >10 Hz updates, and reduces localization error far below baseline.
AI-written summary · about
chapters (14)
- 0:00VILAM introduction
- 0:16Localization and mapping importance
- 1:41SLAM challenges and drift
- 3:09Loop closing and GPS limits
- 4:55Smart roadside infrastructure
- 6:00VILAM system overview
- 7:25Elastic registration method
- 9:35Globally consistent mapping
- 11:00Evaluation setup and tests
- 12:46Real-time performance
- 13:28Accuracy results
- 13:48Effectiveness demo
- 14:54Conclusion and future
- 15:14Q&A
related talks
12mMAVL: Multiresolution Analysis of Voice Localization
52mFlying services with a drone
21mThe race to replace human drivers
26mSelf Driving cars with Opencv
24mKEYNOTE Google's Self Driving Cars
22mA Fully Immersive Driving Simulator for Autonomous Vehicles
21mPmomo: Projection Mapping on Movable 3D Object
48mFlying services with a drone