Research map / Image generation / Text-to-image diffusion
Can Pre-Trained Text-to-Image Models Generate Visual Goals for Reinforcement Learning?
Jialu Gao, Kaizhe Hu, Guowei Xu, Huazhe Xu
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 |
|---|---|---|---|
| use LfVoid-generated goals instead of InstructPix2Pix edited goals | example-based RL · Wipe | cleaned stain patches: 1.7±1.9 -> 21.3±5.8 | Improved |
| use LfVoid-generated goals instead of InstructPix2Pix edited goals | example-based RL · LED | success rate (%): 0.0±0.0 -> 75.0±50.0 | Improved |
| use LfVoid-generated goals instead of InstructPix2Pix edited goals | example-based RL · Push | success rate (%): 3.3±3.5 -> 27.9±12.7 | Improved |
| use generated goal images instead of CLIP language reward | example-based RL · Wipe | cleaned stain patches: 0.0±0.0 -> 21.3±5.8 | Improved |
| use generated goal images instead of CLIP language reward | example-based RL · LED | success rate (%): 10.0±20.0 -> 75.0±50.0 | Improved |
| use generated goal images instead of CLIP language reward | example-based RL · Push | success rate (%): 12.5±19.1 -> 27.9±12.7 | Improved |
| generated goals versus manually created real goal images | example-based RL · LED | success rate (%): 93.3±11.5 -> 75.0±50.0 | Hurt |
| generated goals versus manually created real goal images | example-based RL · Wipe | cleaned stain patches: 22.0±5.8 -> 21.3±5.8 | No clear effect |
| remove feature extracting module (Null-text P2P-DD) (ablation) | LfVoid · Structure-based goal generation | user study Elo: 91.6 -> 63.8 | Hurt |
| remove feature extracting module and P2P editing (Null-text DD) (ablation) | LfVoid · Structure-based goal generation | user study Elo: 91.6 -> 58 | Hurt |
| add P2P attention injection to Null-text DD (ablation) | Null-text DD · Structure-based goal generation | user study Elo: 58 -> 63.8 | Improved |
| LfVoid versus Imagic | goal editing · Appearance-based goal generation | user study Elo: 47.7 -> 87 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- InstructPix2Pix — InstructPix2Pix: Learning To Follow Image Editing Instructions (CVPR 2023) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GoalLadder: Incremental Goal Discovery with Vision-Language Models (NeurIPS 2025 · similarity 0.88)
- Pre-trained Text-to-Image Diffusion Models Are Versatile Representation Learners for Control (NeurIPS 2024 · similarity 0.87)
- RoboDreamer: Learning Compositional World Models for Robot Imagination (ICML 2024 · similarity 0.87)
- GoT-R1: Unleashing Reasoning Capability of Autoregressive Visual Generation with Reinforcement Learning (ICLR 2026 · similarity 0.87)
- Visual Planning: Let's Think Only with Images (ICLR 2026 · similarity 0.87)
- Text-Aware Diffusion for Policy Learning (NeurIPS 2024 · 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.
