Research
Research Projects
Hierarchical Reinforcement Learning for Fragile Object Dexterous Manipulation
- Description: Hierarchical Reinforcement Learning with force feedback for improved sample efficiency in dexterous manipulation for fragile objects
- Machine Learning Keywords: Hierarchical Reinforcement Learning, DQN
- Robot: UR5e Simulated in Gazebo and MuJoCo
- Status/Venue: Accepted @ ICRA2025
Learning from Human Multimodal Feedback
- Description: A novel RLHF framework enabling robots to learn from natural language instructions (via LLM) and human emotions (via computer vision), mediated by a VR system
- Machine Learning Keywords:RLHF, LLM, Computer Vision, Multimodal Interactions
- Robot Embodiment:Demonstrated with a RX250 6 DOF robot arm
- Status/Venue:VR-LLM system highlighted in Viginia Tech News and Center for Human-Computer Interaction News; Presented @ i3CE2025
Multi-agent Intelligent Spatial Planning with Central Knowledge Database
- Description: Built an LLM-RL framework with a GPT-powered data mining pipeline to extract spatio-temporal work plans from the Amazon TEACH dataset, followed by agent-specific task allocation based on partial skill proficiency profiles
- Machine Learning Keywords: LLM-RL, Spatio-Temporal Reasoning, Multi-Agent Task Allocation
- Status/Venue: Presented @ i3CE2025
Knowledge Sharing and Recommendation Agentic AI
- Description: A RAG-based system retrieves text and combines it with SQL queries to extract and complete knowledge from a database. An LLM processes the results to build a tree-structured representation of safety recommendations
- Machine Learning Keywords: Semantic Retrieval, SQL-Augmented LLM, Tree-Based Knowledge Representation, Hierarchical Knowledge Modeling
- Status/Venue: Presented @ i3CE2025
Generalist Robot Learning with Hierarchical Knowledge Graph Extracted from Crowdsourced Internet Data
</div>
- Description: A generalizable embodied AI learning framework that leverages foundational models and hierarchical knowledge representation to process crowdsourced natural language instructions from the internet, minimizing supervision and data complexity for VLA/VLM
- Machine Learning Keywords: Foundational models, VLA/ VLM, World modeling, Spatial intelligence, Generalizable Robot Learning Architecture
- Robot Embodiment:KUKA KR120 Industrial Robot
- Status/Venue: Journal Paper Under Review
Human-Robot Social Behavior Teaching for Close-Proximity Interactions
- Description: Robot interaction behavior profiling and human preference study with potential user/ experienced construction workers
- Machine Learning Keywords: Self-Supervised Learning, Imitation Learning
- Robot Embodiment: KUKA KR120 Industrial Robot
- Status/Venue: Journal Paper Under Review
Sample Efficiency Edge Embodied AI Learning with Hierarhical Knowledge Representations
- Description: A scalable robot programming-by-imitation system that significantly reduces data collection costs through a cloud-based human skill acquisition and management platform, integrated with VR/motion capture-based ergonomic demonstration capture for high-fidelity and reusable micro skill learning
- Machine Learning Keywords: Human Motion Understanding and Representation Learning, Scalable Robot Learning Infrastructure, Edge AI, Human-in-the-Loop Skill Transfer
- Robot Embodiment: KUKA KR120 Industrial Robot
- Status/Venue: Best Paper Award @ i3CE2024
One-shot Programming for Intent Adaptive Control
- Description: Enabled intent-aware and adaptive robotic handovers via a one-shot/ 60-second learning paradigm with tactile gloves, integrating semi-supervised learning for data efficiency and precise human state estimation
- Machine Learning Keywords: One-shot Learning, Semi-Supervised Learning, Internet of Things, LSTM, Sparse Input
- Robot Embodiment: KUKA KR120 Industrial Robot
- Status/Venue: Published in Automation in Construction (2023, Impact Factor 12.7)