diff --git a/hci/undergraduate/content/abstracten.tex b/hci/undergraduate/content/abstracten.tex
index 8d2feaeb..5a659ce6 100644
--- a/hci/undergraduate/content/abstracten.tex
+++ b/hci/undergraduate/content/abstracten.tex
@@ -1,7 +1,6 @@
-%!TEX root = ../csuthesis_main.tex
-\keywordsen{Pedestrian navigation, Reinforcement learning, Virtual simulation, Carla platform, Path planning, Obstacle avoidance}
+\keywordsen{Muscle-driven, Biomechanics, MuJoCo, Human-Computer Interaction, Steering Wheel Control, Precision Test}
 \begin{abstracten}
 
-With the advancement of smart city infrastructure, pedestrian navigation systems face increasing challenges in handling complex and dynamic urban environments. Traditional rule-based planning approaches struggle to adapt to real-time variations in pedestrian behavior. This study proposes a pedestrian navigation and control system based on the Unreal Engine and Carla platform, integrating reinforcement learning to optimize path planning and obstacle avoidance. By simulating realistic traffic scenarios, the system provides a high-fidelity virtual environment for training and evaluating intelligent agents. Reinforcement learning algorithms such as DQN and PPO are applied to improve navigation decision-making, guided by a multi-dimensional reward function that balances goal achievement, safety, and path efficiency. Experiments conducted in various simulation scenarios demonstrate the system's effectiveness in enhancing navigation accuracy, obstacle avoidance success, and overall planning efficiency. The results validate the feasibility of combining virtual simulation with reinforcement learning, offering a scalable and cost-effective approach to intelligent pedestrian navigation, and laying the groundwork for future research in multi-agent collaboration and intelligent traffic systems.
+This thesis focuses on the muscle-driven biomechanical human-computer interaction (HCI) system and designs a steering wheel control system with visual guidance based on the MuJoCo simulation platform. Firstly, the domestic and foreign research status of biomechanical modeling, muscle-driven control, visual guidance and HCI accuracy evaluation is reviewed, and the limitations of existing technologies in generalization, individual adaptation and low-cost simulation are clarified. Secondly, based on the upper limb muscle-skeleton dynamics, a human dual-arm model and a steering wheel physical model are built with MuJoCo, and the coupling drive from muscle activation to steering wheel angle is realized. Then, an adaptive nonlinear mapping algorithm and a real-time visual guidance feedback module are designed to achieve smooth control from muscle force to steering wheel rotation. Finally, the system is verified by functional test, accuracy test, stability test and usability test, using MAE, RMSE, response delay and continuous running stability as indicators. The experimental results show that the mean absolute error of steering wheel angle tracking is less than 3$^\circ$, the response delay is lower than 10 ms, and there is no obvious drift after 1 hour of continuous operation. The system can stably complete the steering wheel control task under visual guidance. This research can provide a lightweight and highly biocompatible HCI solution for driving simulation, rehabilitation training, industrial collaborative manipulation and other scenarios.
 
-\end{abstracten}
+\end{abstracten}
\ No newline at end of file
diff --git a/hci/undergraduate/content/abstractzh.tex b/hci/undergraduate/content/abstractzh.tex
index bc0bba71..88ea047e 100644
--- a/hci/undergraduate/content/abstractzh.tex
+++ b/hci/undergraduate/content/abstractzh.tex
@@ -1,9 +1,6 @@
-%!TEX root = ../csuthesis_main.tex
-% 设置中文摘要
-\keywordscn{行人导航\quad 强化学习\quad 虚拟仿真\quad Carla平台\quad 路径规划\quad 避障}
-%\categorycn{TP391}
+\keywordscn{肌肉驱动\quad 生物力学\quad MuJoCo\quad 人机交互\quad 方向盘控制\quad 精度测试}
 \begin{abstractzh}
 
-智慧城市环境下交通环境对行人导航的要求更高,传统的基于规划路径的方式已经无法满足动态、复杂环境下行人的变化。为了提升行人导航的自适应能力和路径规划效率,本文在虚幻引擎与Carla中模拟一个真实的步行环境,构建一个具备强学习能力的行人控制和行人导航系统,动态仿真真实城市的交通环境,为行人强化学习模型提供测试、训练环境。本文将DQN、PPO等强化学习算法引入行人控制和行人避障,设计多维度奖励函数提升智能体对目标、避障、路径的理解能力,从而提升复杂交通环境下行人的决策能力。对行人导航、避障和路径多场景仿真评估,验证了系统在动态、复杂环境下的鲁棒性。验证了虚拟仿真和强化学习的融合可以拓展智慧交通应用领域,对多智能体协同智能交通未来发展提供可行性方案与路径支撑。
+本课题以肌肉驱动生物力学人机交互为核心，面向视觉引导下的方向盘精准操控需求，设计并实现一套基于 MuJoCo 仿真平台的交互系统。首先梳理生物力学建模、肌肉驱动控制、视觉引导与人机交互精度评估的国内外研究现状，明确现有技术在通用化建模、跨个体适配、低成本仿真等方面的不足。其次以上肢肌肉--骨骼动力学为理论基础，采用 MuJoCo 物理引擎构建人体双臂模型与方向盘物理模型，完成肌肉激活与方向盘转角的耦合驱动。然后设计自适应非线性映射算法与视觉引导实时反馈模块，实现肌肉发力到方向盘转动的平滑控制。最后通过功能测试、精度测试、稳定性测试与易用性测试，以 MAE、RMSE、响应延迟、连续运行稳定性为指标完成系统验证。实验结果表明，本系统方向盘角度跟踪平均绝对误差小于 3$^\circ$，响应延迟低于 10\,ms，连续运行 1 小时无明显漂移，能够稳定完成视觉引导下的方向盘操控任务。本研究可为驾驶模拟、康复训练、工业协同操控等场景提供轻量化、高生物适配性的人机交互方案。
 
-\end{abstractzh}
+\end{abstractzh}
\ No newline at end of file