Research map / Label-efficient and robust learning / Recognition and long-tail
Hidden Biases of End-to-End Driving Models
Bernhard Jaeger, Kashyap Chitta, Andreas Geiger
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use target point conditioning instead of discrete navigation commands (ablation) | TransFuser reproduction · CARLA validation routes | Route Completion: 56 -> 84 | Improved |
| add shift and rotation augmentation to NC-conditioned model (ablation) | TransFuser with NC conditioning · CARLA validation routes | Driving Score: 32 -> 35 | No clear effect |
| replace global average pooling with transformer decoder pooling (ablation) | TransFuser reproduction · CARLA validation routes | Route Completion: 84 -> 93 | Improved |
| replace global average pooling with transformer decoder pooling (ablation) | TransFuser reproduction · CARLA validation routes | static collisions per km: 1.04 -> 0.55 | Improved |
| add shift and rotation augmentation to transformer decoder model (ablation) | TransFuser reproduction · CARLA validation routes | Driving Score: 43 -> 49 | Improved |
| predict path plus confidence-weighted speed classification instead of waypoints (ablation) | TransFuser reproduction · CARLA validation routes | Driving Score: 49 -> 50 | No clear effect |
| use argmax speed class instead of confidence-weighted speed (ablation) | TransFuser reproduction with path output · CARLA validation routes | Driving Score: 50 -> 40 | Hurt |
| add two-stage training with perception pre-training (ablation) | TransFuser reproduction · CARLA validation routes | Driving Score: 50 -> 54 | Improved |
| freeze pre-trained backbone in second training stage (ablation) | TransFuser reproduction · CARLA validation routes | Driving Score: 54 -> 44 | Hurt |
| triple dataset size from 185k to 555k frames (ablation) | TransFuser reproduction · CARLA validation routes | Driving Score: 54 -> 60 | Improved |
| double training epochs in a single stage (ablation) | TransFuser reproduction · CARLA validation routes | Driving Score | No clear effect |
| use TF++ instead of TransFuser | TF++ · CARLA LAV validation towns | Driving Score: 39 -> 70 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TransFuser — TransFuser: Imitation With Transformer-Based Sensor Fusion for Autonomous Driving (TPAMI 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Bench2Drive: Towards Multi-Ability Benchmarking of Closed-Loop End-To-End Autonomous Driving (NeurIPS 2024 · similarity 0.89)
- DriveHierarchy: A Benchmark for Diagnosing VLM Driving Capabilities from Open-Loop Understanding to Closed-Loop Execution (NeurIPS 2026 · similarity 0.88)
- DriveGPT: Scaling Autoregressive Behavior Models for Driving (ICML 2025 · similarity 0.86)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.86)
- Grounding Driving VLA via Inverse Kinematics (NeurIPS 2026 · similarity 0.86)
- Future-Aware End-to-End Driving: Bidirectional Modeling of Trajectory Planning and Scene Evolution (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
