Research map / 3D vision / Gaussian splatting
Efficient Camera Pose Augmentation for View Generalization in Robotic Policy Learning
Sen Wang, Huaiyi Dong, Jingyi Tian, lijiayi lijiayi, Zhuo Yang, Tongtong Cao, Anlin Chen, Shuang Wu, Sanping Zhou, Le Wang
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 GenSplat-rendered novel views to training | Diffusion Policy · Real-world 6 tasks, Large perturbation | success rate: 27.78% -> 43.33% | Improved |
| add GenSplat-rendered novel views to training | pi0 · Real-world 6 tasks, Large perturbation | success rate: +13.3% relative | Improved |
| add GenSplat augmentation, unperturbed source views | pi0 and Diffusion Policy · Real-world, source (no perturbation) | success rate | No clear effect |
| augment with diffusion-based VISTA views | pi0 · Real-world 6 tasks, Large perturbation | avg success: 48.9 -> 37.8 | Hurt |
| augment with NoPoSplat views | pi0 · Real-world 6 tasks, Large perturbation | avg success: 48.9 -> 36.7 | Hurt |
| use GenSplat instead of AnySplat views | pi0 · Real-world 6 tasks, Large perturbation | avg success: 56.1 -> 62.2 | Improved |
| replace VGGT with permutation-equivariant architecture (ablation) | GenSplat · Large (30deg, 6cm) | policy success: 66.0 -> 70.0 | Improved |
| add 3D-prior distillation (ablation) | GenSplat · Extreme (60deg, 12cm) | policy success: 40.0 -> 46.0 | Improved |
| add 3D-prior distillation (ablation) | GenSplat · novel view synthesis | LPIPS: 5.03 -> 3.72 | Improved |
| increase generated views from 0 to 1 per demo (ablation) | Diffusion Policy · Stack Block, Large perturbation | success rate: 30.0 -> 48.0 | Improved |
| increase generated views from 3 to 4 per demo (ablation) | Diffusion Policy · Stack Block, Large perturbation | success rate: 64.0 -> 62.0 | No clear effect |
| increase generated views from 3 to 4 per demo (ablation) | Diffusion Policy · Stack Cups, Large perturbation | success rate: 56.0 -> 54.0 | No clear effect |
Closest papers by meaning
- Exo2EgoPolicy: Pose-Aligned Cross-View Policy Learning (ECCV 2026 · similarity 0.88)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.86)
- Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulation (ICLR 2024 · similarity 0.86)
- GenNBV: Generalizable Next-Best-View Policy for Active 3D Reconstruction (CVPR 2024 · similarity 0.86)
- Scaling Proprioceptive-Visual Learning with Heterogeneous Pre-trained Transformers (NeurIPS 2024 · similarity 0.86)
- Focus On What Matters: Separated Models For Visual-Based RL Generalization (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.
