About me
I am a Ph.D. Candidate in RoboLAND at the University of Southern California (USC), advised by Dr. Feifei Qian. I build legged robots that use legs and bodies as active force sensors, enabling real-time terrain estimation and adaptive navigation beyond vision or LiDAR for wild applications, supported by NASA projects such as LASSIE and TRUSSES.
I also collaborate with Dr. Gaurav S. Sukhatme on embodied proactive agents for human-robot teaming, developing foundation model frameworks that combine fast–slow reasoning to enable robots to proactively assist humans in real time.
My work sits at the intersection of embodied intelligence, active sensing, and human–robot collaboration. I connect foundation models, planning under uncertainty, and proprioceptive-inspired sensing into a unified vision of proactive, adaptive, and collaborative robotic co-scientists in the field.
Reasoning & Planning using foundation model
Submitted to ACM Transactions on Human-Robot Interaction (THRI)
Workshop on Multimodal AI Collaborations @ AAAI 2025
- - Submitted to ACM Transactions on Human-Robot Interaction (THRI)
- - Workshop on Multimodal AI Collaborations @ AAAI 2025
- - Novel LLM-powered framework with DAG architecture for adaptive communication
- - Dual-module system: Coordinator (strategic) + Manager (tactical) for human-robot teaming
- - Dynamic transition between passive and active interaction modes based on task complexity
Language Models for Planning Workshop @ AAAI 2025
- - Submitted to Language Models for Planning Workshop @ AAAI 2025
- - Novel fine-grained benchmark with 22 layouts to assess proactive adaptability and temporal responsiveness
- - MonTA framework using hierarchical LLM approach (fast monitor + slow adapters)
- - Real-time adaptation capabilities enabling proactive decision-making in collaborative scenarios
- - 75% reasonability in human expert evaluations of adaptation plans and instructions
- - Significant performance improvements in low-teaming-fluency scenarios
Robot locomotion & navigation
LPSC 2025
- - In submission
- - Scout-rover cooperation for online terrain strength mapping and traversal risk estimation
- - Planetary-analog exploration applications
- - Multi-institutional collaboration across multiple institutions
Multi-Objective Optimization Workshop @ RSS 25
- - Multi-Objective Optimization Workshop @ RSS 25
- - Safe active navigation and exploration using proprioceptive measurements
- - Joint work towards robust planetary navigation policies
Arxiv
- - Bio-inspired tail oscillation for fast crawling on deformable granular terrains
- - Demonstrates effective locomotion strategies leveraging body-tail interactions
RSS (Robotics: Science and Systems), 2025
- - RSS (Robotics: Science and Systems), 2025
- - Novel proprioceptive sensing method to estimate mud properties through actuator signals
- - RFT-based approach for characterizing substrate strength and force responses
- - Flipper-driven robot design for adaptive locomotion on varying muddy terrains
- - Real-time adaptation prevents locomotion failures in complex, deformable natural environments
IEEE International Conference on Robotics and Automation (ICRA), 2024 & IEEE Robotics and Automation Letters (RAL)
- - IEEE International Conference on Robotics and Automation (ICRA), 2024 & IEEE Robotics and Automation Letters (RAL)
- - Novel force modeling approach measuring horizontal (shear) and vertical (extraction) forces
- - Mudskipper-inspired robot design for studying terrestrial locomotion on muddy substrates
- - Identified two distinct failure mechanisms: high water content (slippage) and low water content (entrapment)
- - Non-monotonic performance dependence on mud water content with optimal performance at 25%-26%
- - Adaptation strategies increased robot speed by more than 200%
In preparation
- - Novel approach to discrete multi-agent path finding using categorical diffusion models
- - Constraint projection methods for ensuring feasible solutions
- - Discrete optimization techniques for multi-agent coordination
Human behavior modelling
ACM/IEEE International Conference on Human-Robot Interaction (HRI 2024)
- - Proceedings of the 2024 ACM/IEEE International Conference on Human-Robot Interaction (HRI '24)
- - 🏆 Best Paper Finalist (5%)
- - Developed models to understand expert sampling strategies in scientific exploration
- - Created web interfaces for interactive exploration of sampling strategies
ACM Transactions on Human-Robot Interaction (THRI)
- - ACM Transactions on Human-Robot Interaction (THRI)
- - Presented at ICRA Workshop, May 2022
- - Presented at American Geophysical Union Fall Meeting, New Orleans, LA, 2021
- - Developed frameworks for understanding human sampling objectives in scientific exploration
