Research map / Video / Action recognition
RobAVA: A Large-scale Dataset and Baseline Towards Video based Robotic Arm Action Understanding
Baoli Sun, Ning Wang, Xinzhu Ma, Anqi Zou, Yihang Lu, Chuixuan Fan, Zhihui Wang, Kun Lu, Zhiyong 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 category prompt to backbone (ablation) | AGPT-Net backbone · RobAVA-R | mean accuracy: 50.2 -> 51.1 | Improved |
| add attribute prompts to backbone (ablation) | AGPT-Net backbone · RobAVA-R | mean accuracy: 50.2 -> 51.9 | Improved |
| replace category prompts with atomic attributes (ablation) | AGPT-Net · RobAVA-R | mean accuracy: 0.8% improvement | Improved |
| combine category and attribute prompts (ablation) | AGPT-Net · RobAVA-R | mean accuracy: 51.9 -> 52.7 | Improved |
| add joint semantic space constraint (JSC) (ablation) | AGPT-Net · RobAVA-S | mean accuracy: 93.9 -> 94.8 | Improved |
| add joint semantic space constraint (JSC) (ablation) | AGPT-Net · RobAVA-R | mean accuracy: 52.7 -> 53.8 | Improved |
| use spatial clustering ratio 1/4 versus no merging (ablation) | AGPT-Net · RobAVA-R | mean accuracy: 50.5 -> 51.0 | Improved |
| use larger clustering ratio 1/8 instead of 1/4 (ablation) | AGPT-Net · RobAVA-R | mean accuracy: 51.0 -> 50.4 | Hurt |
| use temporal downsampling {1,2} instead of {1} (ablation) | AGPT-Net · RobAVA-R | mean accuracy: 51.0 -> 52.7 | Improved |
| use temporal downsampling {2,3} instead of {1,2} (ablation) | AGPT-Net · RobAVA-R | mean accuracy: 52.7 -> 50.8 | Hurt |
| add third temporal scale {1,2,3} versus {1,2} (ablation) | AGPT-Net · RobAVA-R | mean accuracy: 52.7 -> 51.7 | Hurt |
| AGPT-Net versus UniFormerV2 on FineGym288 | AGPT-Net · FineGym288 | mean accuracy: 52.4 -> 54.7 | Improved |
Closest papers by meaning
- RoboInter: A Holistic Intermediate Representation Suite Towards Robotic Manipulation (ICLR 2026 · similarity 0.87)
- Recognizing Actions from Robotic View for Natural Human-Robot Interaction (ICCV 2025 · similarity 0.87)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.86)
- RoboMamba: Efficient Vision-Language-Action Model for Robotic Reasoning and Manipulation (NeurIPS 2024 · similarity 0.86)
- Latent Action Pretraining from Videos (ICLR 2025 · similarity 0.85)
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · similarity 0.85)
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.
