y0news
← Feed
Back to feed
🧠 AI NeutralImportance 7/10

Eva-VLA: Evaluating Vision-Language-Action Models' Robustness Under Real-World Physical Variations

arXiv – CS AI|Hanqing Liu, Shouwei Ruan, Jiahuan Long, Junqi Wu, Jiacheng Hou, Huili Tang, Tingsong Jiang, Weien Zhou, Wen Yao|
🤖AI Summary

Researchers introduced Eva-VLA, the first unified framework to systematically evaluate the robustness of Vision-Language-Action models for robotic manipulation under real-world physical variations. Testing revealed OpenVLA exhibits over 90% failure rates across three physical variations, exposing critical weaknesses in current VLA models when deployed outside laboratory conditions.

Key Takeaways
  • Eva-VLA is the first framework to systematically evaluate VLA model robustness under real-world physical variations.
  • The framework addresses three key dimensions: 3D object transformations, illumination changes, and adversarial regions.
  • OpenVLA showed over 90% failure rates across physical variations on the LIBERO-Long task, revealing significant fragilities.
  • The framework uses continuous black-box optimization to efficiently discover worst-case scenarios without costly real-world data collection.
  • Adversarial training using generated worst-case scenarios quantifiably improves model robustness for robotic manipulation systems.
Read Original →via arXiv – CS AI
Act on this with AI
Stay ahead of the market.
Connect your wallet to an AI agent. It reads balances, proposes swaps and bridges across 15 chains — you keep full control of your keys.
Connect Wallet to AI →How it works
Related Articles