Research map / Reinforcement learning / Visual and representation RL
Learning Visual Servoing with Deep Features and Fitted Q-Iteration
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| learn feature and control weights with FQI rather than TRPO | VGG conv4_3 feature dynamics · Car following, training cars | trajectory cost: 1.69 ± 0.41 -> 1.11 ± 0.29 | Helped |
| learn feature and control weights with FQI rather than TRPO | VGG conv4_3 feature dynamics · Car following, novel cars | trajectory cost: 1.48 ± 0.36 -> 0.90 ± 0.26 | Helped |
| use VGG conv4_3 features instead of locally connected pixel features | FQI visual servo · Car following, training cars | trajectory cost: 7.98 ± 0.77 -> 1.11 ± 0.29 | Helped |
| use VGG conv4_3 features instead of locally connected pixel features | FQI visual servo · Car following, novel cars | trajectory cost: 8.36 ± 0.79 -> 0.90 ± 0.26 | Helped |
| replace per-feature learned weights with a single feature weight (ablation) | VGG conv4_3 feature dynamics · Car following, training cars | trajectory cost: 1.11 ± 0.29 -> 2.70 ± 0.52 | Hurt |
| replace per-feature learned weights with a single feature weight (ablation) | VGG conv4_3 feature dynamics · Car following, novel cars | trajectory cost: 0.90 ± 0.26 -> 2.40 ± 0.50 | Hurt |
| use conv1_2 rather than conv4_3 features (ablation) | FQI visual servo · Car following, training cars | trajectory cost: 1.11 ± 0.29 -> 1.89 ± 0.32 | Hurt |
| use conv2_2 rather than conv4_3 features (ablation) | FQI visual servo · Car following, novel cars | trajectory cost: 0.90 ± 0.26 -> 1.28 ± 0.30 | Hurt |
| use conv3_3 rather than conv4_3 features (ablation) | FQI visual servo · Car following, training cars | trajectory cost: 1.11 ± 0.29 -> 1.56 ± 0.40 | Hurt |
| use conv3_3 rather than conv4_3 features (ablation) | FQI visual servo · Car following, novel cars | trajectory cost: 0.90 ± 0.26 -> 1.04 ± 0.31 | Mixed |
| use the deepest VGG conv5_3 features rather than conv4_3 features (ablation) | FQI visual servo · Car following, training and novel cars | trajectory cost: 1.11 ± 0.29 -> 2.49 ± 0.35; 0.90 ± 0.26 -> 2.56 ± 0.41 | Hurt |
| replace the learned feature-dynamics servo with an end-to-end raw-pixel TRPO policy (ablation) | Car following servo · Car following, training and novel cars | trajectory cost: 1.11 ± 0.29 -> 3.23 ± 0.22; 0.90 ± 0.26 -> 5.20 ± 0.40 | Hurt |
Closest papers by meaning
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.87)
- OSVI-WM: One-Shot Visual Imitation for Unseen Tasks using World-Model-Guided Trajectory Generation (NeurIPS 2025 · similarity 0.87)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.86)
- VICtoR: Learning Hierarchical Vision-Instruction Correlation Rewards for Long-horizon Manipulation (ICLR 2025 · similarity 0.86)
- Robust Visual Imitation Learning with Inverse Dynamics Representations (AAAI 2024 · similarity 0.86)
- Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representations (ICML 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.