Yunhuai Liu

Yunhuai Liu

Auto-generated profile, based on publicly available information

Yunhuai Liu

Peking University

Research Focus & Metrics

Academic Career Length

21Years (2006-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
21 November 2025 in ACM Computing Surveys
Verified
Journal Article
Open access
JJJiaming Ji
Jiaming Ji
TQTianyi Qiu
Tianyi Qiu
BCBoyuan Chen
Boyuan Chen
JZJiayi Zhou
Jiayi Zhou
BZBorong Zhang
Borong Zhang
WGWen Gao
Wen Gao
AI alignment aims to make AI systems behave in line with human intentions and values. As AI systems grow more capable, so do risks from misalignment. To provide a comprehensive and up-to-date overview of the alignment field, in this survey, we delve into the core concepts, methodology, and practice of alignment. First, we identify four principles as the key objectives of AI alignment: Robustness, Interpretability, Controllability, and Ethicality ( RICE ). Guided by these four principles, we outline the landscape of current alignment research and decompose them into two key components: forward alignment and backward alignment . The former aims to make AI systems aligned via alignment training, while the latter aims to gain evidence about the systems’ alignment and govern them appropriately to avoid exacerbating misalignment risks. On forward alignment, we discuss techniques for learning from feedback and learning under the distribution shift. Specifically, we survey traditional preference modeling methods and reinforcement learning from human feedback and further discuss potential frameworks to reach scalable oversight for tasks where effective human oversight is hard to obtain. Within learning under distribution shift, we also cover data distribution interventions such as adversarial training that helps expand the distribution of training data and algorithmic interventions to combat goal misgeneralization. On backward alignment, we discuss assurance techniques and governance practices. Specifically, we survey assurance methods of AI systems throughout their lifecycle, covering safety evaluation, interpretability, and human value compliance. We discuss current and prospective governance practices adopted by governments, industry actors, and other third parties, aimed at managing existing and future AI risks. This survey aims to provide a comprehensive yet beginner-friendly review of alignment research topics. Based on this, we also release and continually update the website www.alignmentsurvey.com which features tutorials, collections of papers, blog posts, and other resources.

4 August 2023
Verified
Conference Paper
Open access
BGBaoshen Guo
Baoshen Guo
SWShuai Wang
Shuai Wang
HWHaotian Wang
Haotian Wang
YLYunhuai Liu
Yunhuai Liu
FKFanshuo Kong
Fanshuo Kong
THTian He
Tian He
The popularity of online e-commerce has promoted the rapid development of last-mile logistics in recent years. In last-mile services, to ensure delivery efficiency and enhance user experience, the delivery zone is proposed to perform delivery task assignment, which is a fundamental part of last-mile delivery. Each courier is responsible for one delivery zone. Couriers will collect orders belonging to their delivery zones from the delivery station and deliver orders to customers. Existing delivery zone partition practices in last-mile logistics consist of manual experience-based and static optimization-based methods, which perform order amount balancing among different zone but suffer from dissatisfaction and inefficiency because of two limitations: (i) using order amount is not always a good balancing metric considering deliveries' various difficulties (e.g., residence or industrial park, with or without elevators); (ii) less considering couriers' familiarity and preference behaviors. To generate delivery zone partition with equitable workload assignment, in this paper, we propose E-partition, a data-driven delivery zone partition framework to achieve equitable workload assignment in last-mile logistics. We first design a learning-based workload prediction model to estimate service time given a partition plan that consists of unseen courier-zone matching scenarios. Then, a delivery zone partition algorithm is proposed to iterative optimize couriers' core-AOI (i.e., area of interest) generation and AOI assignment process. Extensive offline experimental results show that our model outperforms baselines in working time prediction and workload balancing performances. Real-world deployment results at JD Logistics also verify the effectiveness of equitable-assignment aware delivery zone partition, with a 2.2% increase in service on-time rate compared to state-of-practice partition solutions.

17 September 2024 in IEEE Transactions on Mobile Computing
Verified
Journal Article
Closed access
GDGao Demin
Gao Demin
LOLiyuan Ou
Liyuan Ou
YCYongrui Chen
Yongrui Chen
XGXiuzhen Guo
Xiuzhen Guo
RLRuofeng Liu
Ruofeng Liu
THTian He
Tian He
Wi-Fi is the de facto standard for providing wireless access to the Internet in the 2.4 GHz ISM band. Tens of billions of Wi-Fi devices (e.g., smartphones) have been shipped worldwide with limited types of wireless radios operating only when Wi-Fi connectivity is available, making it challenging to access data in heterogeneous IoT devices. However, the direct connection between Wireless Personal Area Network (WPAN) technologies, such as Bluetooth, and Wi-Fi presents challenges due to the inherent distinct physical layer. In our work, a novel communication method called BlueWi has been introduced, which serves as a cross technology communication method that enables BLE devices to establish connections and engage in communication with Wi-Fi based WPAN networks. We let BLE signals hitchhike on ongoing Wi-Fi signals, enabling Wi-Fi to recognize specific BLE signal waveforms in the frequency domain. By analyzing the decoded Wi-Fi payload, BlueWi can retrieve the BLE data, ensuring this method remains fully compatible with existing commodity Wi-Fi hardware. The direct sequence spread spectrum scheme is appended to handle general BLE frames and can be considered as “ COPY ” operation, which allows for better correlation and detection of the signal at the receiver. Evaluations conducted using both USRP and commodity devices have demonstrated that BlueWi can achieve concurrent wireless communication from BLE commercial chips to Wi-Fi networks with a frame reception rate exceeding 96%.

28 June 2024 in IEEE Transactions on Network Science and Engineering
Verified
Journal Article
Closed access
GYGuisong Yang
Guisong Yang
PWPanpan Wen
Panpan Wen
YLYutong Liu
Yutong Liu
LKLinghe Kong
Linghe Kong
YLYunhuai Liu
Yunhuai Liu
Sparse mobile crowdsensing, driven by the increasing ubiquity of smartphones, has emerged as a popular method for data collection. This approach can reconstruct the whole sensing map by inferring missing temporal data from their spatial-temporal correlations to sparse samples. Traditional methods face challenges in capturing non-linear correlations in sensors and error accumulation. To address these issues, we propose a dynamic signal map reconstruction model, i.e., Spatial-Temporal Graph Attention (STGA). Specifically, STGA first applies graph embedding to convert sensing points into graph structures, followed by using a message-passing mechanism based on Graph Attention Networks (GAT) to capture complex spatial-temporal dependent features in sensing data. Then, we propose hierarchical attention mechanism, where different nodes are selected and weighted, and the temporal and spatial sequences of the nodes are stratified separately. By collecting a small amount of observation data, the spatial-temporal characteristics of nodes in the current sparse sensing data graph are effectively learned and the signal graph is updated. Attention mechanism quantitatively evaluates and weights the measurement points to further reduce the measurement error. Evaluations conducted on three typical urban sensing datasets demonstrate the effectiveness of the proposed methods in enhancing the inference accuracy of sparse sensing data.

14 February 2025 in IEEE Transactions on Vehicular Technology
Verified
Journal Article
Closed access
XHXingyu He
Xingyu He
DLDongXu Luo
DongXu Luo
ZCZekun Chen
Zekun Chen
GYGuisong Yang
Guisong Yang
YLYunhuai Liu
Yunhuai Liu
Obstacle avoidance is the crux of formation control for a UAV swarm, no matter in urban or wild application environments. But there is a dilemma to promote both of obstacle avoidance flexibility and consistency in UAV formation control, which has not been solved effectively by existing researches. In view of this, this paper proposes a consensus based reinforcement learning method for differentiated formation control (DFC) of UAVs, to balance the obstacle avoidance flexibility and consistency. In this method, to promote obstacle avoidance flexibility of a UAV swarm, a consensus mechanism with differentiated formation control strategies is designed, it allows each UAV in a swarm changes its formation control strategy among aggregation, formation keeping and obstacle avoidance according to its local environment, and calculates its own current subgoal based on the selected strategy. Further, to improve the flight efficiency of the UAV swarm, a reinforcement learning model is provided to generate the optimal offset vector for each UAV according to its current subgoal. Moreover, to enhance the obstacle avoidance consistency of the UAV swarm, a collaborative obstacle avoidance algorithm is designed in the obstacle avoidance strategy, it requires UAVs to share their obstacle information and obstacle avoidance actions, and provides obstacle avoidance consensus rules to help UAVs to choose consistent obstacle avoidance directions. The experiment results show that the proposed method can combine obstacle avoidance flexibility and consistency of UAVs, thereby achieving higher flight efficiency and maintaining stable network connectivity.

13 March 2025 in Science China Information Sciences
Verified
Journal Article
Open access
PDF file
YLYunhao Liu
Yunhao Liu
LLLi Liu
Li Liu
YZYawen Zheng
Yawen Zheng
YLYunhuai Liu
Yunhuai Liu
FDFan Dang
Fan Dang
KMKe Ma
Ke Ma
Navigation is a fundamental component of modern information application systems, ranging from military, transportations, and logistic, to explorations. Traditional navigations are based on an absolute coordination system that provides a precise map of the physical world, the locations of the moving objects, and the optimized navigation routes. In recent years, many new emerging applications have presented new demands for navigation, e.g., underwater/underground navigations where no GPS or other localizations are available, an un-explored area with no maps, and task-oriented navigations without specific routes. The advances in IoT and AI enable us to design new navigation paradigms, embodied navigation that allows the moving object to interact with the physical world to obtain the local map, localize the objects, and optimize the navigation routes accordingly. We make a systematic and comprehensive review of research in embodied navigation, encompassing key aspects on perceptions, navigation and efficiency optimization. Beyond advancements in these areas, we also examine the emerging tasks enabled by embodied navigation which require flexible mobility in diverse and evolving environments. Moreover, we identify the challenges associated with deploying embodied navigation systems in the real world and extend them to substantial areas. We aim for this article to provide valuable insights into this rapidly developing field, fostering future research to close existing gaps and advance the development of general-purpose autonomous systems grounded in embodied navigation.