Research map / Image generation / Text-to-image diffusion
Exploring Conditions for Diffusion Models in Robotic Control
Heeseong Shin, Byeongho Heo, Dongyoon Han, Seungryong Kim, Taekyung Kim
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 |
|---|---|---|---|
| replace null condition with generated text caption | SCR / SD 1.5 · Cheetah-run | score: 43.4 -> 37.5 | Hurt |
| replace null condition with generated text caption | SCR / SD 1.5 · DeepMind Control mean | normalized score: 68.3 -> 68.9 | No clear effect |
| add task prompts (ablation) | SD 1.5 null condition · DeepMind Control mean | normalized score: 68.3 -> 69.8 | Improved |
| add visual prompts (ablation) | SD 1.5 null condition · DeepMind Control mean | normalized score: 68.3 -> 70.5 | Improved |
| combine task and visual prompts (ablation) | single-prompt variants · DeepMind Control mean | normalized score: 70.5 -> 74.3 | Improved |
| task prompt only on Cheetah-run (ablation) | SD 1.5 · Cheetah-run | score: 43.4 -> 38.9 | Hurt |
| use upsampling-block features instead of down+mid features (ablation) | ORCA · DeepMind Control mean | normalized score: 74.3 -> 49.7 (up 2) | Hurt |
| concatenate down1-3 and mid layers (ablation) | single-layer features · DeepMind Control mean | normalized score: 68.4 -> 74.3 | Improved |
| fine-tune diffusion model downstream (ablation) | SCR · Adroit mean | success rate (%): 58.0 -> 9.3 | Hurt |
| fine-tune encoder downstream (ablation) | VC-1 · Adroit mean | success rate (%): 47.3 -> 31.3 | Hurt |
| learn task+visual prompts (ORCA) vs TADP | SD 1.5 · Adroit mean | success rate (%): 57.3 -> 65.3 | Improved |
Closest papers by meaning
- Pre-trained Text-to-Image Diffusion Models Are Versatile Representation Learners for Control (NeurIPS 2024 · similarity 0.91)
- Prediction with Action: Visual Policy Learning via Joint Denoising Process (NeurIPS 2024 · similarity 0.90)
- Solving New Tasks by Adapting Internet Video Knowledge (ICLR 2025 · similarity 0.89)
- Zero-Shot Robotic Manipulation with Pre-Trained Image-Editing Diffusion Models (ICLR 2024 · similarity 0.89)
- Conditioning Matters: Training Diffusion Policies is Faster Than You Think (NeurIPS 2025 · similarity 0.89)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.88)
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.
