Research map / Video / Action recognition
Learning to Drive by Watching YouTube Videos: Action-Conditioned Contrastive Policy Pretraining
Qihang Zhang, Zhenghao Peng, Bolei Zhou
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 |
|---|---|---|---|
| pretrain ACO on YouTube videos instead of NuScenes (ablation) | ACO · CARLA IL, 10% demos | success rate (%): 13.3 -> 30.7 | Improved |
| pretrain ACO on YouTube videos instead of NuScenes (ablation) | ACO · CARLA IL, 100% demos | success rate (%): 82.7 -> 96.0 | Improved |
| add cropping and flipping augmentation (ablation) | ACO · CARLA IL, 20% demos | success rate (%): 66.0 -> 36.0 | Hurt |
| add cropping and flipping augmentation (ablation) | ACO · CARLA IL, 100% demos | success rate (%): 96.0 -> 89.3 | Hurt |
| add cropping and flipping augmentation (ablation) | MoCo · CARLA IL, 20% demos | success rate (%): 47.3 -> 60.3 | Improved |
| add cropping and flipping augmentation (ablation) | MoCo · CARLA IL, 100% demos | success rate (%): 84.0 -> 80.7 | Hurt |
| add batch normalization after backbone (ablation) | ACO · CARLA IL, 20% demos | success rate (%): 34.0 -> 66.0 | Improved |
| add batch normalization after backbone (ablation) | ACO · CARLA IL, 100% demos | success rate (%): 78.0 -> 96.0 | Improved |
| add batch normalization after backbone (ablation) | MoCo · CARLA IL, 100% demos | success rate (%): 80.0 -> 80.7 | No clear effect |
| add batch normalization after backbone | ACO · CULane (UFLD) | F1: 69.5 -> 70.4 | Improved |
| replace ImageNet pretraining with ACO | ResNet backbone · CARLA IL, 10% demos | success rate (%): 21.3 -> 30.7 | Improved |
| replace random init with ACO pretraining | ResNet backbone · CARLA IL, 20% demos | success rate (%): 0.0 -> 66.0 | Improved |
Closest papers by meaning
- DynaMo: In-Domain Dynamics Pretraining for Visuo-Motor Control (NeurIPS 2024 · similarity 0.89)
- Latent Action Pretraining from Videos (ICLR 2025 · similarity 0.87)
- Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representations (ICML 2025 · similarity 0.87)
- Videos are Sample-Efficient Supervisions: Behavior Cloning from Videos via Latent Representations (NeurIPS 2025 · similarity 0.87)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · similarity 0.86)
- DMC-VB: A Benchmark for Representation Learning for Control with Visual Distractors (NeurIPS 2024 · 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.
