Research map / Video / Video segmentation and tracking
Hierarchical Foresight: Self-Supervised Learning of Long-Horizon Tasks via Visual Subgoal Generation
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 1 subgoal (ablation) | Visual foresight on Maze Navigation Hard · Maze Navigation Hard | success rate: 33 -> 47 | Helped |
| Add a second subgoal (ablation) | HVF on Maze Navigation Hard · Maze Navigation Hard | success rate: 47 -> 54 | Helped |
| Increase the number of subgoals beyond 2 (ablation) | HVF on Maze Navigation Hard · Maze Navigation Hard | success rate: 54 -> 39; 54 -> 2; 54 -> 0 | Hurt |
| Optimize max rather than mean planning cost across segments (ablation) | HVF on Maze Navigation Hard · Maze Navigation Hard, 1 and 2 subgoals | success rate: 0.45 -> 0.47; 0.53 -> 0.54 | Helped |
| Increase visual MPC samples from 200 to 1000 (ablation) | Visual foresight and HVF on Maze Navigation Hard · Maze Navigation Hard, 0, 1, and 2 subgoals | success rate: 0.33 -> 0.35; 0.47 -> 0.54; 0.54 -> 0.55 | Helped |
| Increase MPC planning horizon to 10 steps (ablation) | HVF on Maze Navigation Hard · Maze Navigation Hard, 2 subgoals | success rate: 0.54 -> 0.37 | Hurt |
| Increase MPC planning horizon to 15 steps (ablation) | HVF on Maze Navigation Hard · Maze Navigation Hard, 2 subgoals | success rate: 0.54 -> 0.24 | Hurt |
| Use a longer MPC planning horizon (ablation) | HVF on Maze Navigation Hard · Maze Navigation Hard, 1 subgoal | success rate: 0.47 -> 0.55 -> 0.39 | Mixed |
| Use VAE latent-space ℓ2 cost rather than pixel ℓ2 cost (ablation) | Visual foresight on Maze Navigation · Maze Navigation Easy, Medium, and Hard | success rate: 0.93 -> 0.62; 0.68 -> 0.49; 0.33 -> 0.23 | Hurt |
Closest papers by meaning
- FLIP: Flow-Centric Generative Planning as General-Purpose Manipulation World Model (ICLR 2025 · similarity 0.89)
- Zero-Shot Robotic Manipulation with Pre-Trained Image-Editing Diffusion Models (ICLR 2024 · similarity 0.88)
- Efficient Robotic Policy Learning via Latent Space Backward Planning (ICML 2025 · similarity 0.88)
- Generate Subgoal Images before Act: Unlocking the Chain-of-Thought Reasoning in Diffusion Model for Robot Manipulation with Multimodal Prompts (CVPR 2024 · similarity 0.88)
- Video Language Planning (ICLR 2024 · similarity 0.88)
- VideoVLA: Video Generators Can Be Generalizable Robot Manipulators (NeurIPS 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.