Yuntao Dai

Ruiying Wen, Yuntao Dai, Hongyong Wang

Monitoring the in-service aerodynamic performance of airliners is critical for operational efficiency and safety, but using operational Quick Access Recorder (QAR) data for this purpose presents significant challenges. This paper first establishes that the absence of key parameters, particularly aircraft moments of inertia, makes conventional state-propagation filters fundamentally unsuitable for this application. This limitation necessitates a decoupled, Equation-Error Method (EEM). However, we then demonstrate through a comparative analysis that standard recursive estimators with time-varying gains, such as Recursive Least Squares (RLS), also fail within an EEM framework, exhibiting premature convergence or instability when applied to low-excitation cruise data. To overcome these dual challenges, we propose and validate the Constant-Gain Equation-Error Method (CG-EEM). This framework employs a custom estimator with a constant, Kalman-like gain, which is perfectly suited to the stationary, low-signal-to-noise characteristics of cruise flight. The CG-EEM is extensively validated on a large, multi-fleet dataset of over 200 flights, where it produces highly consistent, physically plausible aerodynamic parameters and correctly identifies known performance differences between aircraft types. The result is a robust, scalable, and computationally efficient tool for fleet-wide performance monitoring and the early detection of performance degradation.

Yuntao Dai, Hang Gu, Teng Wang et al.

Vision-Language-Action (VLA) models have emerged as a unified paradigm for robotic perception and control, enabling emergent generalization and long-horizon task execution. However, their deployment in dynamic, real-world environments is severely hin dered by high inference latency. While smooth robotic interaction requires control frequencies of 20 to 30 Hz, current VLA models typi cally operate at only 3-5 Hz on edge devices due to the memory bound nature of autoregressive decoding. Existing optimizations often require extensive retraining or compromise model accuracy. To bridge this gap, we introduce ActionFlow, a system-level inference framework tailored for resource-constrained edge plat forms. At the core of ActionFlow is a Cross-Request Pipelin ing strategy, a novel scheduler that redefines VLA inference as a macro-pipeline of micro-requests. The strategy intelligently batches memory-bound Decode phases with compute-bound Prefill phases across continuous time steps to maximize hardware utilization. Furthermore, to support this scheduling, we propose a Cross Request State Packed Forward operator and a Unified KV Ring Buffer, which fuse fragmented memory operations into efficient dense computations. Experimental results demonstrate that ActionFlow achieves a 2.55x improvement in FPS on the OpenVLA-7B model without retraining, enabling real-time dy namic manipulation on edge hardware. Our work is available at https://anonymous.4open.science/r/ActionFlow-1D47.