His research focuses on multi-modal learning, specifically advancing large vision-language models, embodied intelligence, and out-of-distribution detection. Xiang has published over 40 papers in top-tier venues, including CVPR, NeurIPS, ICML, AAAI, and ACM MM. He is the recipient of multiple awards, including the NTU Research Excellence Award and Best Student Paper at MIPR 2024, and serves as a reviewer for major AI conferences."

Q1: F or what purpose was the dataset created? As stated in Section 1, most previous datasets on DU focus on single-page DU. Our benchmark is constructed to bridge such a gap. Q2: Who created the dataset (e.g., which team, research group) and on behalf of which entity (e.g., Q3: What support was needed to make this dataset? Q1: What do the instances that comprise the dataset represent (e.g., documents, photos, people, countries)?

The ACIL gives identical results either in growing-exemplar or fixed memory settings.

Q1: F or what purpose was the dataset created? As stated in Section 1, most previous datasets on DU focus on single-page DU. Our benchmark is constructed to bridge such a gap. Q2: Who created the dataset (e.g., which team, research group) and on behalf of which entity (e.g., Q3: What support was needed to make this dataset? Q1: What do the instances that comprise the dataset represent (e.g., documents, photos, people, countries)?

Large Language Models (LLMs) have become key components of modern software, with prompts acting as their de-facto programming interface. However, prompt design remains largely empirical and small mistakes can cascade into unreliable, insecure, or inefficient behavior. This paper presents the first systematic survey and taxonomy of prompt defects, recurring ways that prompts fail to elicit their intended behavior from LLMs. We organize defects along six dimensions: (1) Specification and Intent, (2) Input and Content, (3) Structure and Formatting, (4) Context and Memory, (5) Performance and Efficiency, and (6) Maintainability and Engineering. Each dimension is refined into fine-grained subtypes, illustrated with concrete examples and root cause analysis. Grounded in software engineering principles, we show how these defects surface in real development workflows and examine their downstream effects. For every subtype, we distill mitigation strategies that span emerging prompt engineering patterns, automated guardrails, testing harnesses, and evaluation frameworks. We then summarize these strategies in a master taxonomy that links defect, impact, and remedy. We conclude with open research challenges and a call for rigorous engineering-oriented methodologies to ensure that LLM-driven systems are dependable by design.

The ACIL gives identical results either in growing-exemplar or fixed memory settings.

Multi-axle [1] Swerve-drive Autonomous Guided Vehicle (MS-AGV) is a type of heavy-duty vehicle equipped with multiple independently controlled steering wheels. This design provides MS-AGVs with a unique combination of high load capacity [2, 3] and exceptional maneuverability [4, 5], making them highly suitable for complex industrial environments [6, 7], such as automated warehouses and port logistics [8-12]. However, effectively controlling MS-AGVs presents several challenges. These include achieving accurate kino-dynamic modeling [13], ensuring precise trajectory tracking [14], and optimizing speed for operational efficiency [15, 16]. Recent works have explored prescribed performance control under uncertainties and faults, such as [17, 18], but they do not consider tire wear, which is critical in MS-AGV applications. Furthermore, practical concerns such as minimizing tire wear, which directly impacts maintenance costs, add complexity to the problem [19]. Despite significant advancements, no existing solution [20] comprehensively addresses these issues in an integrated manner, leaving a critical gap in MS-AGV planning and control strategies. Over the past several years, researchers have dedicated substantial effort to developing advanced control strategies to address the trajectory tracking problem in MS-AGV systems [21]. The core technical difficulty lies in managing the steering wheels, as the increased number of state variables [22] and the dynamic complexity [23] of the system make it challenging to predict and control [24] its behavior effectively.

For instance, both ChatGPT [42] and Bing [40] have made the power of transformer-based models widely accessible, democratizing advanced AI capabilities. These models leverage attention mechanisms [55] adeptly to capture long-range dependencies in sequences of input tokens, allowing them to accurately model contextual information. Besides, unlike traditional task-specific learning approaches, large transformer models (e.g., GPT [46] and BERT [10]) are trained on huge quantities of unlabeled textual data and are directly useful for a wide variety of applications such as sentiment analysis, language translation, content generation, and question answering. However, the application of large transformers still presents certain risks, particularly regarding privacy issues [35, 52]. Most popular transformer models operate in a pattern called Machine Learning as a Service (MLaaS), where a server provides the model and inference services to users who own the data. For instance, OpenAI provides ChatGPT as an online platform and offers remote APIs for developers, allowing users to access services by submitting prompts or messages. Nevertheless, this pattern raises privacy concerns: users need to transmit their private data to a company's server and have no direct control over how their data is handled. They must trust that the server processes the data honestly and follows the agreed terms of service. There exists a risk that the server could misuse the data, including unauthorized processing, storing the data indefinitely, or even selling it to third parties.

A robotic arm that can automatically turn cockroaches into controllable cyborgs could be used to create swarms of biological robots for search missions. Hirotaka Sato at Nanyang Technological University in Singapore and his colleagues have previously shown that groups of up to 20 Madagascar hissing cockroaches (Gromphadorhina portentosa) equipped with electronic backpacks can be steered across desert-like terrain. However, to be used in a real-world search-and-rescue mission, the team calculates that hundreds or thousands of cyborg insects would be needed.

This paper proposes a comprehensive strategy for complex multi-target-multi-drone encirclement in an obstacle-rich and GPS-denied environment, motivated by practical scenarios such as pursuing vehicles or humans in urban canyons. The drones have omnidirectional range sensors that can robustly detect ground targets and obtain noisy relative distances. After each drone task is assigned, a novel distance-based target state estimator (DTSE) is proposed by estimating the measurement output noise variance and utilizing the Kalman filter. By integrating anti-synchronization techniques and pseudo-force functions, an acceleration controller enables two tasking drones to cooperatively encircle a target from opposing positions while navigating obstacles. The algorithms effectiveness for the discrete-time double-integrator system is established theoretically, particularly regarding observability. Moreover, the versatility of the algorithm is showcased in aerial-to-ground scenarios, supported by compelling simulation results. Experimental validation demonstrates the effectiveness of the proposed approach.