Nana Zhou

Prof. Nana Zhou


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

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

Representative Publications

View All Publications

26 June 2026 in PLOS ONE
Verified
Journal Article
Open access
NZNana Zhou
Nana Zhou
XZXianhua Zhao
Xianhua Zhao
FZFengjun Zhou
Fengjun Zhou
JLJi Li
Ji Li
XYXiuming Yue
Xiuming Yue
Semantic segmentation provides essential scene understanding for unmanned ground vehicles to identify obstacles and plan paths in unstructured environments. Nevertheless, existing methodologies tailored for these settings typically necessitate linear probing or fine-tuning to accommodate novel scenarios, thereby suffering from a deficiency in zero-shot transferability. In response to this limitation, our study introduces a novel framework designed for robust zero-shot transfer in unstructured domains, capitalizing on the superior visual-linguistic alignment capabilities of the EVA-CLIP architecture. To augment segmentation precision, we initially utilize deep prompt tuning to adapt the visual feature extraction efficacy of the EVA-CLIP image encoder to unstructured terrain features. This strategy not only bolsters adaptability to irregular environments but also preserves the intrinsic zero-shot proficiency of the underlying model. Concurrently, we devise an ensemble prompt engineering scheme customized for unstructured settings to further elevate segmentation outcomes. Moreover, the framework optimizes the correspondence between text and images by integrating global and local representations from the respective encoders, thereby maximizing cross-modal alignment for superior segmentation. Empirical evaluations indicate that our methodology surpasses contemporary state-of-the-art techniques, yielding an increase in mIoU ranging from 1.2% to 43.9% on the Robot Unstructured Ground Driving (RUGD) benchmark. Furthermore, evaluations on the Rellis-3D dataset reveal that the model’s cross-domain zero-shot performance rivals that of supervised fine-tuning approaches, demonstrating robust generalization to previously unseen semantic classes.

23 October 2026 in Journal of King Saud University Computer and Information Sciences
Verified
Journal Article
Open access
PDF file
XDXiangli Dong
Xiangli Dong
FZFengjun Zhou
Fengjun Zhou
NZNana Zhou
Nana Zhou
YWYuxing Wang
Yuxing Wang
BWBing Wu
Bing Wu
ZFZhiquan Feng
Zhiquan Feng
In elderly home care, robots face a “safety-autonomy” trust paradox: rigid interventions ensure physical safety but undermine user autonomy. While existing explainable AI (XAI) attempts value alignment, it often assumes perfect perception and ignores the “explanation fatigue” and dynamic cognitive load of elderly users. To address these challenges, we propose the Sequential Trust Alignment framework (STAMM), featuring three core mathematical formulations: (1) a cognitive-modulated dynamic rationality coefficient (βt); (2) a confidence-aware recursive Bayesian trust update (αt) with a forgetting factor to mitigate multimodal sensory noise; and (3) a dynamic explanation fatigue threshold (τt) ensuring “no disturbance unless necessary.” We evaluated STAMM using a dual-dimensional approach: embodied physical simulations in AI2-THOR under complex noise, and a Human-in-the-Loop (HITL) empirical study across high-risk geriatric scenarios. Results show that STAMM effectively alleviates the tension along the safety-autonomy trade-off, maintaining a high physical safety rate (97.5%) while significantly reducing intervention rejections. Furthermore, the HITL study demonstrates that STAMM achieves significantly higher Overall Trust and Perceived Benevolence while reducing Cognitive Load compared to rigid or “Always-On” baselines. By effectively suppressing trust oscillations under perceptual uncertainty, STAMM demonstrates strong robustness, offering an empathetic and practical approach for real-world Human-Robot Interaction.

27 December 2025 in Sensors
Verified
Journal Article
Open access
PDF file
NZNana Zhou
Nana Zhou
FZFengjun Zhou
Fengjun Zhou
CLChangming Li
Changming Li
JZJianning Zhong
Jianning Zhong
There has been significant interest in solving robot path planning problems using fuzzy logic-based methods. Recently, the Genetic Algorithm-based Hierarchical Interval Type-2 Fuzzy (GA-HIT2F) system has been introduced as a novel planner in this domain. However, this method lacks adaptability to changes in target points, and insufficient flexibility can lead to planning failures in local minimum traps, making it difficult to apply to complex scenarios. In this paper, we identify the limitations of the original GA-HIT2F approach and propose an enhanced Q-Learning-aided Adaptive Hierarchical Interval Type-2 Fuzzy (QL-HIT2F) algorithm for path planning. The proposed planner incorporates reinforcement learning to improve a robot’s capability to avoid collisions with special obstacles. Additionally, the average obstacle orientation (AOO) is introduced to optimize the robot’s angular adjustments. Two supplementary robot parameters are integrated into the reinforcement learning action space, along with fuzzy membership parameters. The training process also introduces the concepts of meta-map and sub-training. Simulation results from a series of path planning experiments validate the feasibility and effectiveness of the proposed QL-HIT2F approach.

26 August 2026 in Electronics
Verified
Journal Article
Open access
PDF file
NZNana Zhou
Nana Zhou
RWRui Wu
Rui Wu
XGXueqiang Gao
Xueqiang Gao
LWLi Wang
Li Wang
ZFZhiquan Feng
Zhiquan Feng
BGBin Guo
Bin Guo
Accurate prediction of electric vehicle charging demand is imperative for ensuring grid stability and optimizing urban mobility resources. While the emergence of large language models has introduced translation-based forecasting paradigms, existing methods typically suffer from numerical precision loss due to textual tokenization and fail to capture complex, non-Euclidean spatial dependencies. To address these limitations, this study introduces STAR, a spatio-temporal agentic reasoning framework that fundamentally redefines the forecasting task as a generative reasoning process. STAR integrates three core innovations, beginning with a temporal patching alignment mechanism that projects historical time-series segments into dense semantic vectors to preserve numerical fidelity. This is seamlessly combined with a graph-conditioned spatial context fusion module that empowers the agent to retrieve dynamic spatial dependencies via cross-attention-based topological fusion over an urban knowledge graph, thereby linking temporal dynamics with spatial causality. Finally, the framework employs an agentic chain-of-thought inference engine that mandates the generation of explicit reasoning traces by analyzing trends and synthesizing external factors prior to outputting the final forecast. Extensive experiments on ST-EVCDP, an open benchmark dataset collected from Shenzhen for urban EV charging demand prediction, demonstrate that STAR significantly outperforms state-of-the-art baselines, achieving a 27.3% to 41.9% prediction improvement for 60 min horizons compared to existing methods. Furthermore, the framework exhibits exceptional zero-shot cross-zone transferability across unseen traffic districts, providing interpretable decision support for critical infrastructure management.