RoadWeaver: AI Generates Large-Scale HD Maps for Autonomous Driving Simulation
A new framework named RoadWeaver has been developed by researchers to create extensive, lane-level HD maps from the ground up for autonomous driving simulations. This innovative system generates a comprehensive road layout, transforms it into an interconnected road network, and builds lane-level geometry with consistent lane connectivity. Experimental findings indicate a reachability of 99.8%, a dead-end ratio of 10.7%, and an endpoint alignment error of just 0.24 meters. RoadWeaver achieves a 94.4% reduction in endpoint alignment error compared to leading generation methods, producing full HD maps in 1.39 to 3.50 seconds. This technology addresses the demand for varied and scalable HD maps for extensive evaluations in intricate road networks. The framework is discussed in a paper available on arXiv (arXiv:2608.11580v1).
Key facts
- RoadWeaver is a coarse-to-fine framework for from-scratch generation of diverse, large-scale HD maps.
- It synthesizes a global road layout, expands it into a connected road network, and constructs lane-level geometry.
- Achieves 99.8% reachability, 10.7% dead-end ratio, and endpoint alignment error of 0.24 m.
- Reduces endpoint alignment error by 94.4% compared to SOTA generation methods.
- Generates complete HD maps in 1.39–3.50 seconds.
- Generated maps can be directly deployed in driving simulators.
- Paper available on arXiv with identifier arXiv:2608.11580v1.
- Addresses scalability limitations of handcrafted or reconstructed real-world maps.
Entities
Institutions
- arXiv