Research map / Video / Action recognition
MultiModal Action Conditioned Video Simulation
Yichen Li, Antonio Torralba
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 multisensory actions instead of UniSim sentence text | I2VGen simulator · ActionSense | FVD: 251.7 -> 203.5 | Improved |
| combine all four modalities instead of hand-pose only (ablation) | multisensory simulator · ActionSense | LPIPS: 0.349 -> 0.276 | Improved |
| combine all modalities instead of muscle EMG only (ablation) | multisensory simulator · ActionSense | FVD: 348.2 -> 203.5 | Improved |
| add text phrase to multisensory conditioning | multisensory simulator · ActionSense | FVD: 203.5 -> 200.4 | No clear effect |
| use our feature instead of ImageBind | multimodal feature · ActionSense | FVD: 315.6 -> 203.5 | Improved |
| use our feature instead of SignalAgnostic | multimodal feature · ActionSense | LPIPS: 0.361 -> 0.276 | Improved |
| use softmax fusion instead of max pooling (ablation) | multimodal fusion · ActionSense | FVD: 284.8 -> 203.5 | Improved |
| use softmax fusion instead of concatenation (ablation) | multimodal fusion · ActionSense | FVD: 279.9 -> 203.5 | Improved |
| add interaction feature y' regularization (ablation) | multisensory simulator · ActionSense | FVD: 339.0 -> 203.5 | Improved |
| use relaxed hyperplane instead of hard projection (ablation) | interaction feature · ActionSense | FVD: 265.5 -> 203.5 | Improved |
| train on all modalities, test with one modality removed (ablation) | multisensory simulator · ActionSense | FVD: 203.5 -> 205.6 | No clear effect |
| add simulator loss Lsim to policy training | diffusion policy · goal-conditioned policy | policy MSE: 1.01 -> 0.90 | Improved |
Closest papers by meaning
- Mastering Robot Manipulation with Multimodal Prompts through Pretraining and Multi-task Fine-tuning (ICML 2024 · similarity 0.85)
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · similarity 0.85)
- Learning an Actionable Discrete Diffusion Policy via Large-Scale Actionless Video Pre-Training (NeurIPS 2024 · similarity 0.85)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.85)
- WorldSimBench: Towards Video Generation Models as World Simulators (ICML 2025 · similarity 0.85)
- VidMan: Exploiting Implicit Dynamics from Video Diffusion Model for Effective Robot Manipulation (NeurIPS 2024 · 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.
