Research map / Reinforcement learning / Imitation learning
Zero-Shot Visual Imitation
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 |
|---|---|---|---|
| add forward consistency loss (ablation) | GSP · Rope knot tying | success rate: 36% -> 60% | Helped |
| add previous-action history and forward consistency (ablation) | GSP · Office navigation, single-image goal in an unseen environment | Num Success: 4 -> 6 | Helped |
| add forward consistency loss (ablation) | GSP-NoFwdConst · Office navigation, maze demonstration, Run-1 | landmarks reached: 65% -> 100% | Helped |
| add forward consistency loss (ablation) | GSP-NoFwdConst · Office navigation, loop demonstration, Run-2 | landmarks reached: 0% -> 100% | Helped |
| collect exploration data with curiosity-driven rather than random exploration (ablation) | GSP (ours features) · VizDoom, different map and texture | Median %: 39.1 ± 2.0 -> 47.1 ± 3.0 | Helped |
| add forward consistency loss (ablation) | GSP-NoFwdConst · VizDoom, different map and texture, curiosity exploration | Median %: 40.2 ± 4.0 -> 47.1 ± 3.0 | Helped |
| replace forward consistency with a forward-feature regularizer (ablation) | GSP · VizDoom, different map and texture, curiosity exploration | Median %: 37.9 ± 1.1 -> 47.1 ± 3.0 | Helped |
| predict forward consistency in feature space rather than pixels (ablation) | GSP · VizDoom, different map and texture, curiosity exploration | Median %: 44.8 ± 4.0 -> 47.1 ± 3.0 | No clear effect |
| predict forward consistency in feature space rather than pixels (ablation) | GSP · VizDoom, different map and texture, curiosity exploration | Efficiency %: 29.5 ± 1.9 -> 24.1 ± 1.7 | Hurt |
| remove previous-action history and forward consistency (ablation) | GSP · Office navigation, single-image goal in an unseen environment | Num Success: 6 -> 2 | Hurt |
| remove previous-action history and forward consistency (ablation) | GSP · Office navigation, maze demonstration, Run-1 | landmarks reached: 100% -> 60% | Mixed |
Closest papers by meaning
- Zero-Shot Offline Imitation Learning via Optimal Transport (ICML 2025 · similarity 0.90)
- OSVI-WM: One-Shot Visual Imitation for Unseen Tasks using World-Model-Guided Trajectory Generation (NeurIPS 2025 · similarity 0.89)
- Zero-Shot Robotic Manipulation with Pre-Trained Image-Editing Diffusion Models (ICLR 2024 · similarity 0.88)
- Deep Demonstration Tracing: Learning Generalizable Imitator Policy for Runtime Imitation from a Single Demonstration (ICML 2024 · similarity 0.88)
- RLIF: Interactive Imitation Learning as Reinforcement Learning (ICLR 2024 · similarity 0.87)
- What's the Move? Hybrid Imitation Learning via Salient Points (ICLR 2025 · 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.