StandardE2E: A Unified Framework for End-to-End Autonomous Driving Datasets

StandardE2E introduces a unified framework that standardizes interfaces across six major autonomous driving datasets, eliminating the need for researchers to rebuild preprocessing pipelines for each dataset. By providing a single PyTorch DataLoader and canonical data schema, the framework accelerates end-to-end autonomous driving research and cross-dataset experimentation.

StandardE2E addresses a critical infrastructure gap in autonomous driving research. The autonomous driving field has accumulated numerous high-quality sensor datasets—Waymo, Argoverse, NAVSIM, and WayveScenes—but each maintains proprietary formats, coordinate systems, and APIs. This fragmentation forces research teams to invest substantial engineering effort replicating preprocessing logic rather than focusing on model innovation. The framework's unified interface eliminates this friction by establishing a canonical schema that maps raw sensor data into standardized formats.

The shift toward end-to-end models in autonomous driving represents a fundamental change in how systems approach vehicle control. Rather than modular pipelines that separately handle perception, prediction, and planning, E2E models directly map sensor inputs to control outputs, often incorporating auxiliary supervision tasks like 3D detection and motion forecasting. This approach demands more diverse training data to achieve robust generalization. StandardE2E enables researchers to combine multiple datasets within a single DataLoader, facilitating cross-dataset pretraining and reducing data scarcity challenges that limit model performance.

For the broader autonomous driving industry, this standardization accelerates research velocity and democratizes access to multi-dataset experimentation. By reducing implementation overhead, StandardE2E allows researchers at organizations with limited infrastructure to conduct sophisticated cross-dataset studies previously accessible only to well-resourced teams. The open-source release as a Python package establishes an important precedent for collaborative infrastructure development. Supporting six datasets immediately creates network effects—adding future datasets requires only per-dataset mapping logic rather than reimplementing entire pipelines.

• →StandardE2E unifies preprocessing across six major autonomous driving datasets under a single canonical schema and PyTorch interface.
• →Cross-dataset pretraining becomes feasible through a unified DataLoader, enabling researchers to leverage multiple data sources simultaneously.
• →Framework design reduces implementation overhead for new datasets to single per-dataset mappings, preserving downstream pipeline compatibility.
• →Open-source release democratizes access to multi-dataset research previously limited to well-resourced organizations.
• →Standardization accelerates end-to-end autonomous driving research by eliminating recurring infrastructure development.