Research Interests

• Scene Perception and Mapping: Spatial scene reconstruction and understanding based on 3DGS map representation, 3DGS-based SLAM technology, and downstream navigation tasks.
• Embodied Intelligence: Automated generation of Real2Sim datasets and enhancement of embodied datasets.
• Downstream LLM Tasks: Tool calling, prompt engineering, social chat Agents, user profile analysis Agents, multi-agent systems, RAG.
• Binary Security: Reverse engineering, game security, web security, and Windows kernel security.
• Hardware Development: STM32, ESP32, and PCB design.

Publications

OpenGS-Fusion: Open-Vocabulary Dense Mapping with Hybrid 3D Gaussian Splatting for Refined Object-Level Understanding

Dianyi Yang, Xihan Wang, Yu Gao, Liu shiyang, Bohan Ren, Yufeng Yue, Yi Yang*

IROS 2025

Automated 3D-GS Registration and Fusion Via Skeleton Alignment and Gaussian-Adaptive Features

Liu shiyang, Dianyi Yang, Yu Gao, Bohan Ren, Yi Yang*, Mengyin Fu

IROS 2025

RoadsideSplat: Robust 3D Gaussian Reconstruction from Monocular Roadside Surveillance

Zhaoxiang Liang, Wenjun Guo, Bohan Ren, Yi Yang*

IROS 2025

Transferring Prior Thermal Knowledge for Snowy Urban Scene Semantic Segmentation

Xiaodong Guo, Tong Liu, Yefeng Mou, Siyuan Chai, Bohan Ren, Yijin Wang

IEEE TITS

Research on the Implementation of a 3D-GS-Based Depth Visual SLAM Method Using RGB-D Camera

Bohan Ren

Undergraduate Thesis

Projects

A2RL: Abu Dhabi Autonomous Racing League in UAE

Description: A2RL is an all-new extreme racing series that is pushing the envelope of autonomous technology. Conceived by ASPIRE and brought together by the tireless efforts of engineers, scientists, and programmers, it is a STEM-fueled race and global autonomous showcase. Each year, teams from educational and technological institutions from around the world compete to develop the fastest, most capable racing AI.

My Work: One of the main software engineers, global track mapping based on factor graph, global optimal trajectory design and optimization, and sensor calibration.

iLexueHelper: An iOS Assistant App for Beijing Institute of Technology

Description: In this project, I utilized HTTP request analysis techniques to reverse-engineer the API protocols of the university's platform. I then developed a cross-platform application using SwiftUI, ensuring seamless operation across iPhone, iPad, and MacBook. The application also features iCloud integration for data synchronization.

My Work: Full-stack development (front-end and back-end), App Store marketing.

Replicating and Modifying the Stanford AI Town Project

Description: Based on the DeepSeek model, I replicate the Stanford AI Town. It is migrated to high school campus and home scenarios, enabling agents to understand and interact with the environment. I introduce more interactions with the world, such as agent interactions with the stock market, a virtual WeChat Moments, and virtual instant messaging software (like QQ), to observe agent performance in more complex environments.

My Work: Prompt engineering, software framework development.

BeitaRobot: A Forum Commenting and Investigation Bot Based on LLM

Description: I developed a complete set of LLM prompts to enable any back-end LLM to call tool interfaces. The bot can autonomously investigate a user's past comments and posts through several rounds of tool calls, perform self-summarization and user profiling in stages, achieving "infinitely long context" and chain-of-thought user profiling capabilities. The bot can automatically summarize user profile biases based on their historical posting records and generate comments that match the tone of a real netizen, according to the post's content and the bot's persona.

My Work: Prompt engineering, software framework development.

LLMBotForQQ: A Human-like QQ Group Bot Based on LLM

Description: To address the issue of limited or no context memory in previous LLM-based QQ group bots, I designed prompts that enable the LLM to maintain a lifelong user profile for precise replies. A BaseMemory and a PersonalityBias library were designed to describe the personality traits of specified users in the group, allowing the bot to recall this information effectively in subsequent conversations. The Qwen-VLM model is used to understand images within the group.

My Work: Prompt engineering, software framework development.

iDepthMIDI: A MIDI Controller Based on iPhone's TrueDepth Camera

Description: The iPhone's Face ID feature relies on the front-facing infrared TrueDepth camera, which consists of an infrared dot projector and an infrared camera with a working range of about 50cm. In my research, I used the gesture detection interface provided by Apple Intelligence to capture hand movements with the front-facing visible light camera and measured hand distance with the infrared dot projector. This was integrated with a MIDI driver on a computer to achieve contactless gesture control of MIDI channel intensity, which can be used by composers to control the performance intensity of virtual instruments in real-time and can be extended to virtual conducting functions.

My Work: Software development.

BITSuperGPT: A Project to Share a ChatGPT Plus Account Locally

Description: A project that allows sharing a ChatGPT Plus account locally, using the native web version of ChatGPT Plus while avoiding OpenAI's detection and performance degradation.

My Work: Full-stack software development.

BobHCSGOInjector: A CSGO DLL Injector that Bypasses Signature Detection

Description: An injector for Counter-Strike: Global Offensive that bypasses signature detection to inject DLLs.

My Work: Reverse engineering, software development.

blender_mmd_genshin_shader_importer: A Blender Plugin for One-Click Genshin-style Shader Importing

Description: A Blender plugin that imports a Genshin Impact-style rendering shader with one click, supporting automatic material name recognition and binding.

My Work: Software development.

Related Videos