Research map / Vision-language and multimodal / Human and embodied modeling
Interpretable Learning for Self-Driving Cars by Visualizing Causal Attention
Jinkyu Kim, John Canny
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 |
|---|---|---|---|
| add visual attention to CNN | CNN+FCN · Comma.ai (testing) | MAE in degrees: 2.54 -> 2.44 | Improved |
| add visual attention to CNN | CNN+FCN · HCE (testing) | MAE in degrees: 1.27 -> 1.18 | Improved |
| add visual attention to CNN | CNN+FCN · Udacity (testing) | MAE in degrees: 4.12 -> 4.15 | No clear effect |
| add LSTM without attention to CNN | CNN+FCN · HCE (testing) | MAE in degrees: 1.27 -> 1.57 | Hurt |
| add attention on top of CNN+LSTM | CNN+LSTM · Comma.ai (testing) | MAE in degrees: 2.58 -> 2.44 | Improved |
| increase attention penalty coefficient from 0 to 20 (ablation) | Attention · Comma.ai (testing) | MAE in degrees: 2.52 -> 2.44 | Improved |
| increase attention penalty coefficient from 0 to 20 (ablation) | Attention · HCE (testing) | MAE in degrees: 1.18 -> 1.20 | No clear effect |
| increase attention penalty coefficient from 0 to 20 (ablation) | Attention · Udacity (testing) | MAE in degrees: 4.15 -> 4.19 | No clear effect |
| use more LSTM layers (1 to 5) (ablation) | soft attention model · steering prediction | MAE | No clear effect |
| set exponential smoothing factor to 0.05 (ablation) | Attention · Comma.ai | MAE | Improved |
| add causal filtering of attention blobs | Attention · Comma.ai | spurious attention blobs: 62% | Improved |
| add causal filtering of attention blobs | Attention · HCE | spurious attention blobs: 58% | Improved |
Closest papers by meaning
- Textual Explanations for Self-Driving Vehicles (ECCV 2018 · similarity 0.94)
- Advisable Learning for Self-Driving Vehicles by Internalizing Observation-to-Action Rules (CVPR 2020 · similarity 0.90)
- Where, What, Why: Towards Explainable Driver Attention Prediction (ICCV 2025 · similarity 0.88)
- Towards Safer and Understandable Driver Intention Prediction (ICCV 2025 · similarity 0.88)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.87)
- DualAD: Disentangling the Dynamic and Static World for End-to-End Driving (CVPR 2024 · similarity 0.85)
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.
