Searcharxiv⌕ Search

arXiv subjects

Chengyu Cao

Publications and source records attributed to Chengyu Cao.

6 recordsLinked to original sources
arXiv2025

Sparse Activation Editing for Reliable Instruction Following in Narratives

Complex narrative contexts often challenge language models' ability to follow instructions, and existing benchmarks fail to capture these difficulties. To address this, we propose Concise-SAE, a training-free framework that improves instruction following by identifying and editing instruction-relevant neurons using only natural language instructions, without requiring labelled data. To thoroughly evaluate our method, we introduce FreeInstruct, a diverse and realistic benchmark of 1,212 examples that highlights the challenges of instruction following in narrative-rich settings. While initially motivated by complex narratives, Concise-SAE demonstrates state-of-the-art instruction adherence across varied tasks without compromising generation quality.

cs.CL↗
arXiv2020

Robust Adaptive Control of Linear Parameter-Varying Systems with Unmatched Uncertainties

In controlling systems with large operating envelopes, it is often necessary to adjust the desired dynamics according to operating conditions. This paper presents a robust adaptive control architecture for linear parameter-varying (LPV) systems that allows for the desired dynamics to be systematically scheduled, while being able to handle a broad class of uncertainties, both matched and unmatched, which can depend on both time and states. The proposed controller adopts an L1 adaptive control architecture for designing the adaptive control law and peak-to-peak gain (PPG) minimization for designing the robust control law to mitigate the effect of unmatched uncertainties. Leveraging the PPG bound of an LPV system, we derive transient and steady-state performance bounds in terms of the input and output signals of the actual closed-loop system as compared to the same signals of a nominal system. The efficacy of the proposed method is validated by extensive simulations using the short-period dynamics of an F-16 aircraft operating in a large envelope.

eess.SY↗
arXiv2020

L1-Adaptive MPPI Architecture for Robust and Agile Control of Multirotors

This paper presents a multirotor control architecture, where Model Predictive Path Integral Control (MPPI) and L1 adaptive control are combined to achieve both fast model predictive trajectory planning and robust trajectory tracking. MPPI provides a framework to solve nonlinear MPC with complex cost functions in real-time. However, it often lacks robustness, especially when the simulated dynamics are different from the true dynamics. We show that the L1 adaptive controller robustifies the architecture, allowing the overall system to behave similar to the nominal system simulated with MPPI. The architecture is validated in a simulated multirotor racing environment.

eess.SY↗
arXiv2006

Guaranteed Transient Performance with $\mathcal{L}_1$ Adaptive Controller for Systems with Unknown Time-varying Parameters: Part I

This paper presents a novel adaptive control methodology for uncertain systems with time-varying unknown parameters and time-varying bounded disturbance. The adaptive controller ensures uniformly bounded transient and asymptotic tracking for system's both signals, input and output, simultaneously. The performance bounds can be systematically improved by increasing the adaptation gain. Simulations of a robotic arm with time-varying friction verify the theoretical findings.

math.OC↗
arXiv2006

Stability Margins of $\mathcal{L}_1$ Adaptive Controller: Part II

In Part I of this paper, we have developed a novel $\mathcal{L}_1$ adaptive control architecture that enables fast adaptation and leads to uniformly bounded transient and asymptotic tracking for system's both signals, input and output, simultaneously. In this paper, we derive the stability margins of $\mathcal{L}_1$ adaptive control architecture, including time-delay and gain margins in the presence of time-varying bounded disturbance. Simulations verify the theoretical findings.

math.OC↗