Research
My research agenda is centered on intelligent software systems in the era of large language models, multi-agent collaboration, and embodied intelligence.
Research Vision
My research is organized around intelligent software systems rather than isolated NLP or ML tasks. The goal is to build a closed loop from requirements understanding, code generation, testing, runtime analysis, fault diagnosis, and self-repair to intelligent agents and embodied systems.
以人工智能为核心，以智能软件工程为主阵地，以多智能体与认知机制为方法特色，面向智能机器人和真实应用场景扩展，构建从需求理解、代码生成、运行分析、故障诊断到自我修复的智能软件系统闭环。
Research Areas
Artificial Intelligence & Multi-Agent Systems
This direction studies how LLM-based agents reason, remember, reflect, coordinate, and adapt when solving complex technical tasks. We are especially interested in cognitive multi-agent architectures that divide roles, exchange evidence, recover from failures, and remain trustworthy under uncertainty.
Keywords: Large Language Model Agents, Multi-Agent Collaboration, Cognitive Reasoning, Memory and Reflection, Trustworthy and Adaptive AI
Intelligent Software Engineering
This direction builds AI-driven methods for the full software lifecycle, from requirements analysis to code generation, testing, trace analysis, debugging, repair, and evolution. The emphasis is on verifiable software artifacts, requirement-code-test alignment, and systems that use runtime evidence instead of relying only on generated text.
Keywords: Intelligent Requirements Engineering, Code Generation and Repair, Trace-Driven Debugging, Software Reverse Engineering, Requirements-Code-Test Alignment
Intelligent Robotics & Human-AI Collaboration
This direction extends cognitive and collaborative AI toward embodied systems that perceive environments, plan tasks, make decisions, and collaborate with humans. Students can work on language-to-goal grounding, adaptive task planning, human-robot interaction, and intelligent systems that connect software agents with physical or simulated environments.
Keywords: Task Planning, Environment Perception, Embodied Intelligence, Human-Robot Collaboration, Autonomous Decision Making
Representative Systems
TraceCoder studies trace-driven multi-agent debugging and repair for LLM-generated code. The system follows an observe-analyze-repair loop and uses runtime execution traces, causal analysis, historical lessons, and rollback-based repair.
KGMAF / ChatREQ studies knowledge-guided multi-agent collaboration for intelligent requirements engineering. The framework coordinates role-specialized agents through shared artifacts, injected requirements knowledge, and human-in-the-loop review.
Research Group
The CSMA Research Group is based at Chongqing University of Posts and Telecommunications (CQUPT) and works on artificial intelligence, intelligent software engineering, multi-agent systems, intelligent robotics, and application-driven intelligent systems. The group maintains collaboration and academic exchange with universities and research institutions including Nanyang Technological University, Peking University, Wuhan University, and Beihang University.