SMART: Advancing Scalable Map Priors for Driving Topology Reasoning

TL;DR

SMART is proposed, a scalable solution that leverages easily available standard-definition and satellite maps to learn a map prior model, supervised by large-scale geo-referenced high-definition maps independent of sensor settings.

Abstract

Topology reasoning is crucial for autonomous driving as it enables comprehensive understanding of connectivity and relationships between lanes and traffic elements. While recent approaches have shown success in perceiving driving topology using vehicle-mounted sensors, their scalability is hindered by the reliance on training data captured by consistent sensor configurations. We identify that the key factor in scalable lane perception and topology reasoning is the elimination of this sensor-dependent feature. To address this, we propose SMART, a scalable solution that leverages easily available standard-definition (SD) and satellite maps to learn a map prior model, supervised by large-scale geo-referenced high-definition (HD) maps independent of sensor settings. Attributed to scaled training, SMART alone achieves superior offline lane topology understanding using only SD and satellite inputs. Extensive experiments further demonstrate that SMART can be seamlessly integrated into any online topology reasoning methods, yielding significant improvements of up to 28% on the OpenLane-V2 benchmark.

Figures (4)

• Figure 1: Comparison between baseline and SMART-OL. Existing topology reasoning methods suffer from limited sensor data. SMART augments online topology reasoning with robust map priors learned from scalable SD and satellite maps, substantially improving lane perception and topology reasoning.
• Figure 2: Outline of the proposed approach. In the first stage (bottom row), SMART is trained at scale using SD and satellite maps for lane graph prediction, supervised by large-scale geo-referenced HD maps. In the second stage (top row), the robust map priors learned by SMART are seamlessly integrated into any online driving topology reasoning models, significantly enhancing lane perception and topology reasoning.
• Figure 3: Qualitative comparison of SMART-OL to baselines. The top-left shows the SD map plotted on top of the satellite image. Our method improves baselines consistently, producing more complete lane graphs.
• Figure 4: Impact of varying sensor data availability. With only 40% of sensor data, SMART-OL achieves performance comparable to using the full sensor data, demonstrating its robustness with reduced sensor data.