MaskFuser: Masked Fusion of Joint Multi-Modal Tokenization for End-to-End Autonomous Driving

TL;DR

MaskFuser introduces a unified multi-modal tokenization and masked auto-encoder training for improved end-to-end autonomous driving, enhancing fusion and robustness under sensory damage.

Contribution

This work presents MaskFuser, the first to unify multi-modal features into a shared semantic space and apply masked auto-encoder training for autonomous driving.

Findings

Achieves 49.05 driving score on CARLA benchmark

Improves route completion to 92.85%

Enhances stability under sensory damage

Abstract

Current multi-modality driving frameworks normally fuse representation by utilizing attention between single-modality branches. However, the existing networks still suppress the driving performance as the Image and LiDAR branches are independent and lack a unified observation representation. Thus, this paper proposes MaskFuser, which tokenizes various modalities into a unified semantic feature space and provides a joint representation for further behavior cloning in driving contexts. Given the unified token representation, MaskFuser is the first work to introduce cross-modality masked auto-encoder training. The masked training enhances the fusion representation by reconstruction on masked tokens. Architecturally, a hybrid-fusion network is proposed to combine advantages from both early and late fusion: For the early fusion stage, modalities are fused by performing monotonic-to-BEV translation attention between branches; Late fusion is performed by tokenizing various modalities into a unified token space with shared encoding on it. MaskFuser respectively reaches a driving score of 49.05 and route completion of 92.85% on the CARLA LongSet6 benchmark evaluation, which improves the best of previous baselines by 1.74 and 3.21%. The introduced masked fusion increases driving stability under damaged sensory inputs. MaskFuser outperforms the best of previous baselines on driving score by 6.55 (27.8%), 1.53 (13.8%), 1.57 (30.9%), respectively given sensory masking ratios 25%, 50%, and 75%.