Research map / Vision-language and multimodal / Robot action and navigation
FedVLA: Federated Vision-Language-Action Learning with Dual Gating Mixture-of-Experts for Robotic Manipulation
Cui Miao, Tao Chang, Meihan Wu, Hongbin Xu, Chun Li, Ming Li, Xiaodong 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 Instruction-Oriented Scene-Parsing (IOSP) (ablation) | FedVLA · Real-world UR3 (average) | success rate (%): 41.1 -> 63.3 | Improved |
| add IOSP (ablation) | FedVLA · Real-world UR3 (Trash Collection) | success rate (%): 13.3 -> 46.7 | Improved |
| add Dual Gating MoE instead of standard FFN (ablation) | FedVLA · Real-world UR3 (average) | success rate (%): 31.7 -> 63.3 | Improved |
| add DGMoE instead of standard FFN (ablation) | FedVLA · Real-world UR3 (Clean Up) | success rate (%): 20.0 -> 53.3 | Improved |
| add Expert-Driven Aggregation instead of FedAvg aggregation (ablation) | FedVLA · Real-world UR3 (average) | success rate (%): 26.7 -> 63.3 | Improved |
| use FedVLA instead of FedAvg | FedAvg · Meta-World simulation (average) | success rate (%): 51.7 -> 63.3 | Improved |
| use FedVLA instead of FedAvg | FedAvg · Real-world UR3 (average) | success rate (%): 53.3 -> 63.3 | Improved |
| federated FedVLA instead of centralized training | Centralized · Meta-World simulation (average) | success rate (%): 65.0 -> 63.3 | No clear effect |
| federated FedVLA instead of centralized training | Centralized · Real-world UR3 (average) | success rate (%): 63.4 -> 63.3 | No clear effect |
| federated FedVLA instead of centralized training | Centralized · Meta-World simulation (Close Drawer) | success rate (%): 73.3 -> 80.0 | Improved |
| federated FedVLA instead of centralized training | Centralized · Meta-World simulation (Door Lock) | success rate (%): 86.7 -> 80.0 | Hurt |
| use dual-gating DGMoE instead of fixed top-k MoE | FedVLA · Real-world UR3 | avg activated experts per token: 1.219, 1.229, 1.225, 1.227 | Improved |
Closest papers by meaning
- FineVLA: Fine-Grained Instruction Alignment for Steerable Vision-Language-Action Policies (NeurIPS 2026 · similarity 0.89)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.89)
- ForceVLA: Enhancing VLA Models with a Force-aware MoE for Contact-rich Manipulation (NeurIPS 2025 · similarity 0.89)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
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.
