Domain Adaptation of Drag Reduction Policy to Partial Measurements

2025-07-06

AI Summary

In fluid-system drag-reduction control, full-state measurement is often infeasible due to practical sensor limitations. Method: This paper proposes Domain-Specific Feature Transfer (DSFT), a deep learning framework that reconstructs the complete flow state and adapts control policies using only sparse, local sensor time-series data. DSFT jointly learns the nonlinear mapping from partial observations to the full wake state, automatically determines the optimal temporal observation window, and disentangles the intrinsic structure of the control policy. Integrating reinforcement learning with domain adaptation, it trains policies on high-fidelity simulation data with full-state access and bridges the sim-to-real gap via DSFT under realistic sensing constraints. Contribution/Results: Evaluated on a simplified road-vehicle aerodynamic drag-reduction task, DSFT achieves near-full-state-feedback control performance using only sparse surface-mounted sensors—demonstrating substantial improvements in policy robustness, practicality, and deployability under low-sensing conditions.

Technology Category

• Intelligent Robots: State Estimation
• Search and Optimization: Learning to Search
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Feedback control of fluid-based systems poses significant challenges due to their high-dimensional, nonlinear, and multiscale dynamics, which demand real-time, three-dimensional, multi-component measurements for sensing. While such measurements are feasible in digital simulations, they are often only partially accessible in the real world. In this paper, we propose a method to adapt feedback control policies obtained from full-state measurements to setups with only partial measurements. Our approach is demonstrated in a simulated environment by minimising the aerodynamic drag of a simplified road vehicle. Reinforcement learning algorithms can optimally solve this control task when trained on full-state measurements by placing sensors in the wake. However, in real-world applications, sensors are limited and typically only on the vehicle, providing only partial measurements. To address this, we propose to train a Domain Specific Feature Transfer (DSFT) map reconstructing the full measurements from the history of the partial measurements. By applying this map, we derive optimal policies based solely on partial data. Additionally, our method enables determination of the optimal history length and offers insights into the architecture of optimal control policies, facilitating their implementation in real-world environments with limited sensor information.

Problem

• Adapt feedback control from full to partial measurements
• Minimize aerodynamic drag with limited sensor data
• Reconstruct full measurements from partial data history

Innovation

• Domain Specific Feature Transfer map for reconstruction
• Optimal policies from partial measurement history
• Determines optimal history length for control