Research map / Reinforcement learning / Imitation learning
Keyframe-Focused Visual Imitation Learning
Chuan Wen, Jierui Lin, Jianing Qian, Yang Gao, Dinesh Jayaraman
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 |
|---|---|---|---|
| Upweight action-changepoint frames using APE-based step weights | History-based behavioral cloning · CARLA NoCrash-Dense | %success: 33.000 ± 4.190 -> 43.444 ± 0.786 | Improved |
| Upweight action-changepoint frames using APE-based step weights | History-based behavioral cloning · CARLA NoCrash-Dense without speed | %success: 25.667 ± 0.981 -> 36.778 ± 5.808 | Improved |
| Use history rather than a single observation | Behavioral cloning · CARLA NoCrash-Dense without speed | %success: 9.222 ± 2.380 -> 25.667 ± 0.981 | Improved |
| Use history rather than a single observation | Behavioral cloning · CARLA NoCrash-Dense | %success: 42.667 ± 8.668 -> 33.000 ± 4.190 | Hurt |
| Apply APE-based softmax weighting rather than APE-based step weighting (ablation) | Keyframe-focused behavioral cloning · Hopper | reward: 905±135 -> 951±117 | Improved |
| Apply APE-based softmax weighting rather than APE-based step weighting (ablation) | Keyframe-focused behavioral cloning · HalfCheetah | reward: 470 ± 205 -> 819 ± 96 | Improved |
| Apply APE-based softmax weighting rather than APE-based step weighting (ablation) | Keyframe-focused behavioral cloning · Walker2D | reward: 654±53 -> 769±97 | Improved |
| Replace APE-based weighting with Bayesian changepoint detection weighting (ablation) | Keyframe-focused behavioral cloning · CARLA NoCrash-Dense | %success: 43.444 ± 0.786 -> 28.667 ± 2.494 | Hurt |
| Replace APE-based weighting with action-category frequency weighting (ablation) | Keyframe-focused behavioral cloning · CARLA NoCrash-Dense | %success: 43.444 ± 0.786 -> 20.333 ± 5.825 | Hurt |
| Replace APE-based weighting with AdaBoost-style reweighting (ablation) | Keyframe-focused behavioral cloning · CARLA NoCrash-Dense | %success: 43.444 ± 0.786 -> 3.000 ± 1.414 | Hurt |
| Omit ego velocity from the input observations | Single-observation behavioral cloning · CARLA NoCrash-Dense | %success: 42.667 ± 8.668 -> 9.222 ± 2.380 | Hurt |
| Use 1k expert queries with DAGGER rather than offline APE-weighted imitation | Behavioral cloning · HalfCheetah | reward: 819 ± 96 -> 822 ± 186 | Improved |
Closest papers by meaning
- Robust Visual Imitation Learning with Inverse Dynamics Representations (AAAI 2024 · similarity 0.89)
- OSVI-WM: One-Shot Visual Imitation for Unseen Tasks using World-Model-Guided Trajectory Generation (NeurIPS 2025 · similarity 0.89)
- Videos are Sample-Efficient Supervisions: Behavior Cloning from Videos via Latent Representations (NeurIPS 2025 · similarity 0.88)
- What's the Move? Hybrid Imitation Learning via Salient Points (ICLR 2025 · similarity 0.88)
- Imitation Learning from a Single Temporally Misaligned Video (ICML 2025 · similarity 0.87)
- DynaMo: In-Domain Dynamics Pretraining for Visuo-Motor Control (NeurIPS 2024 · similarity 0.87)
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.
