Online Temporal Fusion for Vectorized Map Construction in Mapless Autonomous Driving

2024-09-01

IEEE Robotics and Automation Letters

AI Summary

Single-frame vectorized mapping in HD-map-free autonomous driving suffers from poor robustness and incomplete geometric-topological consistency. Method: We propose the first long-term online vectorized mapping framework, featuring (i) a hash-based sparse semantic voxel map that fuses multi-frame historical landmark detections; (ii) a domain-knowledge-driven road structure reasoning mechanism enabling instance-level clustering and joint geometric-topological modeling; and (iii) end-to-end integration into a planning-and-control (PnC) closed-loop system. Contribution/Results: Our approach breaks the single-frame limitation, achieving the first online, real-time, semantically enhanced vector map generation. Experiments in complex urban environments demonstrate significantly improved map consistency and accuracy: localization and structural errors are reduced by over 50% compared to single-frame baselines. The framework has been successfully deployed in a production autonomous driving system.

Technology Category

• Intelligent Robots: Localization, Mapping, and Navigation
• Computer Vision: Visual Reasoning & Symbolic Representations
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

To reduce the reliance on high-definition (HD) maps, a growing trend in autonomous driving is leveraging onboard sensors to generate vectorized maps online. However, current methods are mostly constrained by processing only single-frame inputs, which hampers their robustness and effectiveness in complex scenarios. To overcome this problem, we propose an online map construction system that exploits the long-term temporal information to build a consistent vectorized map. First, the system efficiently fuses all historical road marking detections from an off-the-shelf network into a semantic voxel map, which is implemented using a hashing-based strategy to exploit the sparsity of road elements. Then reliable voxels are found by examining the fused information and incrementally clustered into an instance-level representation of road markings. Finally, the system incorporates domain knowledge to estimate the geometric and topological structures of roads, which can be directly consumed by the planning and control (PnC) module. Through experiments conducted in complicated urban environments, we have demonstrated that the output of our system is more consistent and accurate than the network output by a large margin and can be effectively used in a closed-loop autonomous driving system.

Problem

• Reduces reliance on HD maps for autonomous driving
• Overcomes single-frame input limitations in map construction
• Enhances robustness and accuracy in complex urban environments

Innovation

• Fuses historical detections into semantic voxel map
• Uses hashing-based strategy for sparse road elements
• Estimates road structures for planning and control