Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Visual Semantic Planning Using Deep Successor Representations
Yuke Zhu, Daniel Gordon, Eric Kolve, Dieter Fox, Li Fei-Fei, Abhinav Gupta, Roozbeh Mottaghi, Ali Farhadi
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 |
|---|---|---|---|
| add imitation-learning bootstrapping instead of RL from scratch | SR model / A3C · THOR Medium tasks | success rate: 0.00 -> 0.80 | Improved |
| add LSTM memory to the classifier baseline (ablation) | CLS-MLP · THOR Medium tasks | success rate: 0.65 -> 0.80 | Improved |
| add LSTM memory to the classifier baseline (ablation) | CLS-MLP · THOR Hard tasks | mean successful episode length: 475.86 -> 136.94 | Improved |
| add RL fine-tuning after imitation learning (ablation) | SR IL · THOR Medium tasks | mean successful episode length: 32.32 -> 26.56 | Improved |
| add RL fine-tuning after imitation learning (ablation) | SR IL · THOR Easy tasks | mean successful episode length: 2.70 -> 2.57 | Improved |
| add RL fine-tuning after imitation learning (ablation) | SR IL · THOR Medium tasks | success rate: 0.80 -> 0.80 | No clear effect |
| replace classifier baseline CLS-LSTM with SR IL | CLS-LSTM · THOR Hard tasks | mean successful episode length: 136.94 -> 34.25 | Improved |
| replace classifier baseline CLS-LSTM with SR IL | CLS-LSTM · THOR Medium tasks | mean successful episode length: 314.05 -> 32.32 | Improved |
| replace classifier baseline CLS-LSTM with SR IL | CLS-LSTM · THOR Medium tasks | success rate: 0.80 -> 0.80 | No clear effect |
| fine-tune only the reward vector w for new tasks instead of training from scratch (ablation) | SR model · THOR hard tasks, 4 new items | convergence speed | Improved |
| fine-tune the entire model instead of only w (ablation) | SR model · THOR hard tasks, 4 new items | convergence speed | Hurt |
| train with imitation learning instead of RL | A3C-style RL · THOR | failed actions | Improved |
Closest papers by meaning
- Visual Planning: Let's Think Only with Images (ICLR 2026 · similarity 0.89)
- Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 2025 · similarity 0.87)
- ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning (NeurIPS 2025 · similarity 0.87)
- World-aware Planning Narratives Enhance Large Vision-Language Model Planner (NeurIPS 2025 · similarity 0.86)
- OSVI-WM: One-Shot Visual Imitation for Unseen Tasks using World-Model-Guided Trajectory Generation (NeurIPS 2025 · similarity 0.86)
- VLMs-Guided Representation Distillation for Efficient Vision-Based Reinforcement Learning (CVPR 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.
