PhD Student, Carnegie Mellon University
I am a 4th-year PhD student in Electrical and Computer Engineering affiliated with the Robotics Institute at Carnegie Mellon University, working with Prof. Guanya Shi (LeCAR Lab) and Prof. Guannan Qu.
My research interests lie at the intersection of physics-based robot data generation and the understanding of generative models in control.
* denotes equal contribution; † denotes equal advising.
ICLR 2026
TL;DR: Generative control policies benefit from supervised iterative refinement and suitable stochasticity. A simple two-step regression policy matches flow-based policies on behavior-cloning benchmarks, challenging explanations centered on multimodal action distributions.
arXiv 2026
TL;DR: FS-MPC combines feedback-guided sampling with local and global search to improve sample efficiency and control unstable, contact-rich systems, including real humanoid locomotion and manipulation.
TMLR 2026
TL;DR: We analyze sampling-based optimization through diffusion-style smoothing and develop DIDA, an annealed algorithm with a global convergence guarantee under the stated assumptions.
ICRA 2025 Best Paper Finalist
TL;DR: DIAL-MPC uses diffusion-style annealing to refine sampled control trajectories, enabling real-time, training-free torque control with full quadruped dynamics, including precise real-world jumping with payloads.
L4DC 2024
TL;DR: We analyze MPPI convergence and use the results to design an optimal sampling covariance schedule, improving sampling-based control in simulation and real-world agile quadrotor tasks.
RSS 2025
TL;DR: ASAP learns a residual action model from real-world data to correct simulation mismatch, then fine-tunes motion-tracking policies to enable more agile and coordinated humanoid skills.
IROS 2023
TL;DR: Tactile sensing builds an estimate of unknown object shape and pose, while Bayesian optimization balances exploration with reorientation for insertion, reducing exploration time in simulation.
ICRA 2023
TL;DR: Symmetry-aware actor-critic learning and object-centric goal relabeling help two arms coordinate multi-object handovers and rearrangement, with demonstrations on real robots and in human-robot collaboration.
EUSIPCO 2022
TL;DR: Multi-agent reinforcement learning with a Hausdorff-distance reward trains a decentralized policy for flexible robot formations without a shared global coordinate system, validated in simulation and on mobile robots.
Trajectory optimization for quadrotor drones carrying slung payloads, enabling aggressive aerial transportation maneuvers through explicit time optimization and collision-aware planning.
A wearable headgear system that guides visually impaired users through head-conditioned reflex signals for intuitive, hands-free obstacle avoidance navigation.
Research Intern · San Francisco, USA
Working on robot data curation and annotation for pretraining. Supervised by Guanya Shi and Rocky Duan.
Research Science Intern · Pittsburgh, USA
Developed universal human-to-robot physics-based retargeting system for converting raw human interaction trajectories to feasible robot trajectories. Supervised by Francois Hogan.
PhD Student · Pittsburgh, USA
Developing generative models for control, bridging learning-based generative models with model-based control for contact-rich real-world tasks. Supervised by Prof. Guanya Shi and Prof. Guannan Qu.
Summer Intern · Stanford, USA
Implemented tactile-based in-hand manipulation system using Bayesian optimization without object shape prior or vision information. Supervised by Prof. Jeannette Bohg.
Research Assistant · Beijing, China
Developed bimanual coordination system for handover and rearrangement tasks using structured reinforcement learning. Supervised by Prof. Yi Wu.
Research Assistant · Beijing, China
Designed decentralized formation control system for mobile robots using multi-agent RL and Hausdorff distance. Supervised by Prof. Yuan Shen.
