Full-Body Motion Reconstruction with Sparse Sensing from Graph Perspective

Feiyu Yao, Zongkai Wu, Li Yi

AAAI 2024

PDF code

What Do You Want? User-centric Prompt Generation for Text-to-image Synthesis via Multi-turn Guidance

Yilun Liu, Minggui He, Feiyu Yao (corresponding author), Yuhe Ji, Shimin Tao, Jingzhou Du, Duan Li, Jian Gao, Li Zhang, Hao Yang, Boxing Chen, Osamu Yoshie

in submission

PDF

Few-shot Domain Adaptation for IMU Denoising

Feiyu Yao, Zongkai Wu, Zhenyu Wei, Donglin Wang

in Arxiv,

PDF

Graph based Environment Representation for Vision-and-Language Navigation in Continuous Environments

Ting Wang, Zongkai Wu, Feiyu Yao, Donglin Wang

ICASSP 2024

PDF

Adapting Large Language Models to Log Analysis with Interpretable Domain Knowledge

Yuhe Ji, Yilun Liu, Feiyu Yao, Minggui He, Shimin Tao, Xiaofeng Zhao, Su Chang, Xinhua Yang, Weibin Meng, Yuming Xie, Boxing Chen, Hao Yang

in submission

PDF

Model reference adaptive tracking control for hydraulic servo systems with nonlinear neural-networks

Zhikai Yao, Jianyong Yao, Feiyu Yao, Qiang Xu, Minrui Xu and Wenxiang Deng

ISA Transactions

Page

LogPrompt: Prompt Engineering Towards Zero-Shot and Interpretable Log Analysis

Yilun Liu, Shimin Tao, Weibin Meng, Feiyu Yao, Xiaofeng Zhao, Hao Yang

ICSE 2024

PDF

SRDC: Semantics-based Ransomware Detection and Classification with LLM-assisted Pre-training

Ce Zhou, Yilun Liu, Weibin Meng, shimin tao, Weinan Tian, Feiyu Yao, Xiaochun Li, Tao Han, Boxing Chen, Hao Yang

AAAI 2025

Enhancing inference capabilities of llms with knowledge graphs

Wei Tang, Xiaosong Qiao, Xiaofeng Zhao, Min Zhang, Chang Su, Yuang Li, Yinglu Li, Yilun Liu, Feiyu Yao, Shimin Tao, Hao Yang, He Xianghui

ACL 2024 Workshop

PDF

Biped Robot Dance

• Brainstormed appearance of organic-looking two-legged robot, and designed the CAD model of all components.

• Fabricated each component of robot using 3D printing and assembled components with the Raspberry Pi, motors, controller board, etc., as entire robot.

• Achieved 1st in of robot speed competition.

• Robot dance show is online: https://www.youtube.com/watch?v=g9lacYcaEdc.

Soft Robot with Genetic Algorithm

• Built virtual physics engine and 3D environment with C++ and OpenGL to simulate the real-world that satisfies the laws of physics.

• Treated each soft robot as a parent genotype and iteratively utilized the selection, mutation, and crossover mechanism to evolve the internal state of soft robots to improve the speed of hopping gait.

• Designed multiple criteria (speed, age, mass center and support plan) and optimized robot population referring to pareto front.

• Best soft robot performance demo: https://www.youtube.com/watch?v=Rx6Fo3JLdqw

Object Detection in Haze

• Combined dehazing network (AECR-NET) and object detection network (YOLO v5) with data augmentation strategy such as rotation, resizing and fine-tuning components on different datasets.

• Won 1st in Class

Object Detection in Haze

• Developed an algorithm for referring video object segmentation task, which aimed to do object segmentation for each frame in video.

• Established causal inference analysis for existing methods and located failure cases on unbalanced object category. distribution in dataset. Applied contrastive learning to reduce image features with no causal relevance.

• The model can get better performance on fine-grained picture with less training time.

Reinforcement Learning in Quadruped Robots

• Controlled quadruped robot to search for specific object.

• Implemented a hierarchical control strategy for quadruped robots, with Soft Actor-Critic reinforcement learning method as upper controller and Optimal Control as lower controller.

• Introduced a SLAM sensor for body control and object detection.

Dynamic Torque Control of Multi-link Robotic Arms

• Controlled the state (i.e., velocity, angle, and torque) of multi-link robotic arms to achieve specific positions in OpenAI Gym.

• Approximated forward dynamic model with multilayer perceptron, trained on the data generated by physical analog. The Robotic arm’s performance will be similar to physical analog only when data is sufficient and target position is not far from initial position).

• Compared performance of traditional control method (Modern Predictive Control) with Reinforcement Learning methods (Deep Q-Network and Proximal Policy Optimization).