// AI news · 2026-10-28 · 量子位
FluxVLA Engine V2 Technical Report Released
FluxVLA Engine V2 introduces online reinforcement learning and expands evaluation across multiple simulators and real robots.
Published 2026-10-28 18:32 (Beijing time)Source 量子位Auto-summarised, fact-checked against source
01Summary
Model success rates improve by up to 15.4 percentage points on RoboTwin tasks, and inference frequency increases 2.31–9.64 times. The system supports 14 major strategies across simulation and hardware platforms with standardized data pipelines and deployment chains.
Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.
02Key points
- Online RL boosts success rates by +10.0% on LIBERO-10 and +15.4% on RoboTwin Clean
- Inference frequency improves 2.31–9.64x, e.g., π0.5 from 1.40Hz to 4.40Hz on AGX Orin
- Supports 14 mainstream strategies across simulation and real robots including LimX Oli
- RTC and trajectory post-processing ensure smooth, physically valid motion execution
03Source
工程能力一键复用，专心卷新能力！轮子造一次就够了｜详解FluxVLA Engine技术报告
量子位 · published 2026-10-28 18:32 Beijing time
Also reported by 2 other outlets
05Related news
- Exowatt beats frontier models with SageMaker serverless customization2026-10-28 · AWS Machine Learning Blog
- Cohere Reports Qwen Model Attempted Upload to Beijing Server2026-10-27 · TechCrunch AI
- GitHub Copilot Fusion supports local model routing2026-10-28 · github-changelog-copilot
- AWS Enables Ray Train DLC Distributed Training on EKS2026-10-29 · AWS Machine Learning Blog
From AIMap AI news: collected from 1,100+ official and media sources. Summaries only restate the original article; follow the link for the full story. All AI news · Today