Research map / Label-efficient and robust learning / Recognition and long-tail
Textual Explanations for Self-Driving Vehicles
Jinkyu Kim, Anna Rohrbach, Trevor Darrell, John Canny, Zeynep Akata
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 the CNN+FC controller (ablation) | vehicle controller · BDD-X vehicle control | MAE acceleration (m/s^2): 6.92 -> 6.87 | Improved |
| add prior inputs (vehicle motion, navigation) to CNN+FC (ablation) | CNN+FC controller · BDD-X vehicle control | MAE course (degree): 12.1 -> 6.74 | Improved |
| add prior inputs to CNN+LSTM+Attention (ablation) | vehicle controller · BDD-X vehicle control | MAE acceleration (m/s^2): 6.87 -> 2.29 | Improved |
| increase entropy regularization lambda_c from 0 to 1000 (sparser attention) (ablation) | vehicle controller · BDD-X vehicle control | MAE acceleration (m/s^2): 2.29 -> 5.02 | Hurt |
| increase entropy regularization lambda_c from 0 to 100 (ablation) | vehicle controller · BDD-X vehicle control | MAE acceleration (m/s^2): 2.29 -> 2.68 | Hurt |
| add controller-explanation attention alignment (WAA) vs rationalization (ablation) | explanation generator · BDD-X explanations | CIDEr-D: 61.99 -> 69.52 | Improved |
| add strongly aligned attention (SAA, lambda_c=100) vs rationalization (ablation) | explanation generator · BDD-X explanations | CIDEr-D: 61.99 -> 66.09 | Improved |
| add entropy regularization lambda_c=100 to WAA (ablation) | explanation generator · BDD-X explanations | CIDEr-D: 64.19 -> 69.52 | Improved |
| add weak alignment (WAA) instead of strong alignment (SAA), lambda_c=100 (ablation) | explanation generator · BDD-X explanations | BLEU-4: 7.074 -> 7.281 | Improved |
| use controller inputs and attention in the justifier vs S2VT (ablation) | explanation generator · BDD-X explanations | CIDEr-D: 53.35 -> 61.99 | Improved |
| add spatial attention to S2VT (ablation) | S2VT · BDD-X explanations | CIDEr-D: 53.35 -> 51.37 | Hurt |
| add spatial and temporal attention to S2VT (ablation) | S2VT · BDD-X explanations | CIDEr-D: 53.35 -> 52.74 | No clear effect |
Closest papers by meaning
- Interpretable Learning for Self-Driving Cars by Visualizing Causal Attention (ICCV 2017 · similarity 0.94)
- Advisable Learning for Self-Driving Vehicles by Internalizing Observation-to-Action Rules (CVPR 2020 · similarity 0.89)
- Grounding Human-To-Vehicle Advice for Self-Driving Vehicles (CVPR 2019 · similarity 0.89)
- Explainable Object-Induced Action Decision for Autonomous Vehicles (CVPR 2020 · similarity 0.88)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.86)
- Less Language, More Latents: Annotation-Efficient VLAs for Driving (NeurIPS 2026 · similarity 0.84)
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.
