日本フィジカルAI新聞

世界のフィジカルAIを、日本語で。

VLAarXiv:2609.37334

ロボット実行誤差下でのVLAの制御:自己補償とストレステスト

Taming VLAs under Robot Execution Errors: Self-Compensation and Stress Testing

シェア:XThreadsFacebookLINEはてブBluesky

VLAポリシーがロボットの実行誤差をオンラインで自己補償する手法を提案し、シミュレーションベンチマークRoboStressと実機実験で有効性を示した。

詳しい要約

1. どんなもの?

- VLA policies often fail when executed motion deviates from commanded action due to execution errors from robot mechanics and operating conditions. - Proposes self-compensating VLA, a deployment-time adaptation method that pre-compensates for execution errors without task rewards or labels. - Introduces RoboStress, a controlled simulation benchmark for stress-testing VLA robustness across execution conditions. - RoboStress combines joint-level models of friction, backlash, compliance, and gravi…

2. 先行研究と比べてどこがすごい?

- Unlike methods that build in robustness during training, self-compensating VLA adapts online at deployment time using the residual between commanded and executed actions. - Achieves higher average task success on RoboStress than both base policies and training-time robustness methods. - On physical robots, it improves success rate by more than 30 percentage points, with gains extending to unseen objects. - Prior approaches may not address execution errors that depend on robot state and motion…

3. 技術・手法の肝は?

- Self-compensating VLA updates the policy online using the residual between the action commanded by a VLA and the motion executed by the robot. - It requires no task rewards or labels, enabling deployment-time adaptation. - RoboStress combines established joint-level models of friction, backlash, compliance, and gravity-compensation error into seven deployment scenarios. - These scenarios generate execution errors that depend on the robot's state and motion history.

4. どうやって有効だと検証した?

- Validated on RoboStress, a controlled simulation benchmark, where self-compensating VLA achieves higher average task success than base policies and training-time robustness methods. - On two physical robot arms with different usage histories, it raises the average task success rate by more than 30 percentage points on each arm. - Gains extend to objects not seen in the task demonstrations.

5. 議論はある?

- The abstract does not discuss limitations or potential negative results. - It highlights that execution errors arise from robot mechanics and operating conditions such as wear and payload changes. - The method's reliance on residual feedback and its applicability to other error types are not discussed in the abstract.

6. 次に読むべき論文は?

- The abstract references base VLA policies and methods that build in robustness during training, but does not cite specific papers. - Related works include VLA policies (e.g., RT-1, RT-2) and training-time robustness methods (e.g., domain randomization). - RoboStress is introduced as a benchmark, so reading its details would be beneficial.

※ AIが要旨から生成した要約です。正確性は原文をご確認ください。

著者: Sohyun Lee, Yoonjae Baek, Jaesang Won, Jinnyeong Kim, Kang Hyunwoo, Seung-Hwan Baek, Ivan Laptev, Suha Kwak

分類: cs.RO

原文アブストラクト

Vision-language-action (VLA) policies often fail when a robot's executed motion deviates from their commanded action. Such execution errors arise from the robot's mechanics and operating conditions, such as wear and payload changes. We propose self-compensating VLA, a deployment-time adaptation method that enables a VLA policy to pre-compensate for the robot's execution errors when generating commands. Without task rewards or labels, it updates the policy online using the residual between the action commanded by a VLA and the motion executed by the robot. To stress-test VLA robustness across execution conditions that are impractical to cover with physical robots alone, we introduce RoboStress, a controlled simulation benchmark. It combines established joint-level models of friction, backlash, compliance, and gravity-compensation error into seven deployment scenarios whose execution errors depend on the robot's state and motion history. On RoboStress, self-compensating VLA achieves higher average task success than both the base policies and methods that build in robustness during training. On two physical robot arms with different usage histories, it raises the average task success rate by more than 30 percentage points on each arm, and the gains extend to objects not seen in the task demonstrations.

関連論文

  1. 1Boston Dynamics、車輪付きAtlasの研究試作機を紹介研究・技術
  2. 2GXOとApptronik、Apolloの物流現場試験を拡大 最大1,000台の展開を構想量産・展開
  3. 3newmoとKDDIが資本業務提携 2028年に堺市で自動運転タクシー商用化目指す提携・買収
  4. 4NVIDIA、CoRLでGR00T 2とIsaac Lab 3.0を発表新製品・発表
  5. 5UC BerkeleyとNVIDIAの研究者、AIエージェントでロボット制御コードを仮想環境で改善研究・技術

PR・アフィリエイト広告

FRONTIER(GPU搭載BTOパソコン)

FRONTIER(GPU搭載BTOパソコン)の公式広告バナーFRONTIERでGPU搭載PCを見る ↗
PR本紙発行元 EmplifAI