Yuxuan Liu

Yuxuan Liu

Auto-generated profile, based on publicly available information

Yuxuan Liu

Shenzhen International Graduate School, Tsinghua University, Shenzhen, Guangdong, CHINA

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

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

Representative Publications

View All Publications
19 July 2024 in IEEE Journal of Selected Topics in Signal Processing
Verified
Journal Article
Closed access
XCXuecheng Chen
Xuecheng Chen
HWHaoyang Wang
Haoyang Wang
YCYuhan Cheng
Yuhan Cheng
HFHaohao Fu
Haohao Fu
YLYuxuan Liu
Yuxuan Liu
XCXinlei Chen
Xinlei Chen
Delivery drones provide a promising sensing platform for smart cities thanks to their city-wide infrastructure and large-scale deployment. However, due to limited battery lifetime and available resources, it is challenging to schedule delivery drones to derive both high sensing and delivery performance, which is a highly complicated optimization problem with several coupled decision variables. Meanwhile, this complex optimization problem involves multiple interconnected decision variables, making it even more complex. In this paper, we first propose a delivery drone-based sensing system and formulate a mixed-integer non-linear programming problem (MINLP) that jointly optimizes the sensing utility and delivery time, considering practical factors including energy capacity and available delivery drones. Then we provide an efficient solution that integrates the strength of deep reinforcement learning (DRL) and heuristic, which decouples the highly complicated optimization search process and replaces the heavy computation with a rapid approximation. Evaluation results compared with the state-of-the-art baselines show that DDL improves the scheduling quality by at least 46% on average. More importantly, our proposed method could effectively improve the computational efficiency, which is up to 98 times higher than the best baseline.

3 June 2024
Verified
Conference Paper
Open access
YLYuxuan Liu
Yuxuan Liu
HWHaoyang Wang
Haoyang Wang
FMFanhang Man
Fanhang Man
JXJingao Xu
Jingao Xu
FDFan Dang
Fan Dang
XCXinlei Chen
Xinlei Chen
Mobile air pollution sensing methods are developed to collect air quality data with higher spatial-temporal resolutions. However, existing methods cannot process the spatially mixed gas samples effectively due to the highly dynamic temporal and spatial fluctuations experienced by the sensor, leading to significant measurement deviations. We find an opportunity to tackle the problem by exploring the potential patterns from sensor measurements. In light of this, we propose MobiAir, a novel city-scale fine-grained air quality estimation system to deliver accurate mobile air quality data. First, we design AirBERT, a representation learning model to discern mixed gas concentrations. Second, we design a knowledge-informed training strategy leveraging massive unlabeled city-scale data to enhance the AirBERT performance. To ensure the practicality of MobiAir, we have invested significant efforts in implementing the software stack on our meticulously crafted Sensing Front-end, which has successfully gathered air quality data at a city-scale for more than 1200 hours. Experiments conducted on collected data show that MobiAir reduces sensing errors by 96.7% with only 44.9ms latency, outperforming the SOTA baseline by 39.5%.

8 October 2023
Verified
Conference Paper
Open access
HWHaoyang Wang
Haoyang Wang
YLYuxuan Liu
Yuxuan Liu
CZChenyu Zhao
Chenyu Zhao
JHJiayou He
Jiayou He
WDWenbo Ding
Wenbo Ding
XCXinlei Chen
Xinlei Chen
Accurate calibration of low-cost sensors is critical for improving their potential in environmental monitoring. State-of-the-art (SOTA) methods based on supervised learning commonly calibrate sensor measurements with ground truth from the immediate past or future. However, these techniques rely heavily on labeled data which is challenging to obtain in real-world scenarios. Thus, this paper introduces CaliFormer, a novel representation learning model using self-supervised learning to extract time- and spatial-invariant knowledge from unlabeled measurements. Moreover, we propose pre-training enhancements and model architecture modifications to help train CaliFormer. We then fine-tune the calibration model with the learned representations, which is supervised by limited labeled data. Finally, we comprehensively evaluate our calibration model with a dataset collected by low-cost sensors. Results show that our model outperforms other SOTA calibration methods significantly.

11 April 2025 in IEEE Transactions on Mobile Computing
Verified
Journal Article
Closed access
NZNan Zhou
Nan Zhou
YLYuxuan Liu
Yuxuan Liu
HWHaoyang Wang
Haoyang Wang
FMFanhang Man
Fanhang Man
JXJingao Xu
Jingao Xu
XCXinlei Chen
Xinlei Chen
Mobile air pollution sensing methods have emerged to collect air quality data with improved spatial and temporal resolutions. However, existing methodologies struggle to effectively process spatially mixed gas samples due to the highly dynamic fluctuations experienced by sensors, resulting in significant measurement deviations. We identify an opportunity to address this issue by exploring potential patterns within sensor measurements. To this end, we propose CatUA, a novel city-scale fine-grained air quality estimation system designed to deliver accurate mobile air quality data. First, we design AirBERT, a representation learning model specifically aimed at discerning mixed gas concentrations from sensor data. Second, we implement a Prompt-informed Training Strategy that leverages extensive unlabeled and minimal labeled city-scale data to enhance the performance of CatUA. Notably, the Auto-Prompt mechanism allows CatUA to conveniently acquire new knowledge tailored to specific downstream tasks. To ensure the practicality of CatUA, we have invested considerable effort in developing the software stack on our meticulously crafted Sensing Front-end, which has successfully gathered city-scale air quality data for over 1,200 hours. Experiments conducted on the collected data demonstrate that CatUA reduces sensing errors by 96.9% with a latency of only 44.9 ms, outperforming the state-of-the-art baseline by 42.6%.

28 February 2024
Verified
Conference Paper
Open access
HWHaoyang Wang
Haoyang Wang
XLXinyu Luo
Xinyu Luo
CRCiyu Ruan
Ciyu Ruan
XCXuecheng Chen
Xuecheng Chen
WDWenhua Ding
Wenhua Ding
XCXinlei Chen
Xinlei Chen
Background and motivation. The localization and tracking of high-speed dynamic objects find applications across various domains. For example, tracking unmanned aerial vehicles (UAVs) serves a crucial role for regulatory authorities, enabling the swift identification of illicitly operated drones. Related work. Presently, the widespread use of cameras and Radar is observed in the realm of locating and tracking moving objects [1]. An illustrative example is [2], which integrates an RGB camera with Radar for object tracking, yielding follow-up effects in AR applications. Despite their successes, these methods encounter notable latency when tracking high-speed dynamic objects. The latency often extends into the hundreds of milliseconds due to several factors: ① cameras necessitate tens of milliseconds for exposure and may be susceptible to motion blur; ② algorithms may require a sequence of frames for optimizing localization results. This extended latency becomes a critical limitation when dealing with objects moving at speeds surpassing 10m/s, as the system is left with inadequate time for a response. Consequently, current methods fall short of meeting our requirement for precise real-time tracking of dynamic objects. Our Insight. Nowadays, event camera, characterized by its asynchronous and motion-activated nature, boasting microsecond level temporal resolution, is finding application in high-frequency monitoring. This trend serves as a catalyst for our exploration into the integration of these asynchronous measurements with advanced vision techniques, aiming to enhance the tracking for high-speed objects. Challenges. Albeit inspiring, translating it into a practical system is non-trivial and encounters significant challenges: • C1: Event burst and depth obtain delay impair object localization. Event cameras demonstrate enhanced sensitivity to changes in the environment. Crucial events triggered by objects are often overshadowed by the abundance of events triggered by changes in the environment. • C2: Heterogeneous data hinders consistently high-precision tracking. The event camera exhibits a high detection frequency, but its results lack scale information. Simultaneously, the rough 3D localization of objects can be obtained from the depth camera's results, but its frequency is low. Our work. To surmount these challenges, we conceive DO-Tracker, a framework adapt at merging event and depth measurements for the precise localization and tracking of high-speed dynamic objects as illustrated in Fig. 1. To tackle (C1), we devise a clustering and bayes filter-based algorithm, specifically gear towards object detection and object's 2D location tracking. Concurrently, the object undergoes segmentation on the depth map using event camera detection results, effectively addressing scale-related challenges. To address (C2), we introduce the Graph-Instructed Collaborative Localization algorithm, which harmonizes heterogeneous observations from both cameras through joint optimization. We implement DO-Tracker using commercial event and depth cameras and conduct preliminary experiments of tracking a ball, confirming the viability of the approach (Fig. 2).

28 February 2024
Verified
Conference Paper
Open access
YCYuhan Cheng
Yuhan Cheng
XCXuecheng Chen
Xuecheng Chen
YYYixuan Yang
Yixuan Yang
HWHaoyang Wang
Haoyang Wang
YLYuxuan Liu
Yuxuan Liu
XCXinlei Chen
Xinlei Chen
Unforeseen gas leaks can swiftly create a highly flammable atmosphere, which will lead to risks of explosions, health hazards, as well as long-term environmental impacts, especially when the gas is toxic. Hence, localizing the gas source as soon as possible remains paramount to executing subsequent mitigation measures, such as shutting off valves and effectively containing or sealing the leaks. Assigning human operators to engage in these laborious activities poses a risk to their health and lives. Deploying autonomous mobile olfactory robots for gas source localization (GSL) presents a solution with great potential, as they can search the environment efficiently without risking human lives. We task gas leakage in a factory with intricate pipelines as an example as shown in fig. 1(a). Here, a fleet of robots equipped with gas concentration sensors [4] and wind sensors are dispatched to search the emission source. Unfortunately, existing approaches are not able to offer feasible GSL solutions for mobile olfaction-based robots in realistic environments [1]. Specifically, the patchy nature of gas plumes and the discontinuity of the concentration field confuses and traps the robots. As demonstrated by experimental data in fig. 1(b), plumes fragment into disjointed patches spatially. This intermittent characteristic introduces uncertainty in the source direction when robots devise paths using noisy and local sensory data, ultimately leading to their entrapment within these patches. Our work aims to deploy and schedule a team of robots to conduct efficient GSL in realistic patchy gas environments. We develop GasHunter, a collaborative strategy for multiple robots by exchange-based team roles adaptation to balance the search efficiency and localization effectiveness of gas sources. Robots are assigned heterogeneous roles that represent different planning preferences. For example, some robots are responsible for inspecting spurious concentration maxima (exploiter), while others are in charge of discovering new potential source positions (explorer). The robots' roles are adjusted (exchanged) adaptively based on their current positions and the potential to localize the source in the surroundings, thus parallelizing the exploration of the environment and exploitation of collected information in a flexible manner. We validate our design with experiments on a multi-robot testbed and large scale physical feature based simulations. We compared GasHunter with two related state-of-the-art (SOTA) methods, Surge-Cast [2] and Infotaxis [3]. We visualize the trajectories of all the methods in fig. 2 (a)-(c). It is demonstrated that GasHunter produces smoother trajectories than baselines. Surge-Cast yields inefficient zig-zag paths, Infotaxis suffers from trap in patchy areas. fig. 2 (d) illustrates the nearest distance to the source within all robots over time. It's verified that GasHunter navigates towards the source the most efficiently and arrives at the source the most quickly, indicating a decrease in search time by 37.23% and 22.46% in comparison to the two baselines, respectively.