Rafiqul Islam

When Faster VLA Deployment Changes Closed-Loop Behavior: Task Success-Latency Analysis of SmolVLA Across PyTorch and ONNX Variants

Vision-language-action (VLA) deployment can reduce inference latency while changing closed-loop task behavior while changing closed-loop task behavior through deployment evaluation, and deployment evaluation should jointly report latency, artifact inspection, interface constraints, and closed-loop success.