Biography
I am a Member of Technical Staff at Amazon FAR (Frontier AI & Robotics).
I received my Master’s degree from the Robotics Institute at Carnegie Mellon University advised by Guanya Shi, and Bachelor’s degree from Shanghai Jiao Tong University advised by Hesheng Wang and Zhongqiang Ren. I also interned at Tsinghua University (IIIS) advised by Huazhe Xu. During my undergraduate career, I led the programming group at SJTU VEX Robotics Club.
Current Focus: Scalable learning systems for robots to perceive, reason, and act reliably in the real world.
Robots: I have worked on many types of robots, including humanoids, mobile manipulators, and aerial robots.
News
- [2026.01.22] FALCON is accepted to L4DC 2026 and selected as an Oral!
- [2025.08.01] Hold My Beer is accepted to CoRL 2025.
- [2025.04.11] ASAP and Flying Hand are accepted to RSS 2025.
- [2025.01.29] Catch It! is accepted to ICRA 2025.
Experience
Applied Scientist Intern May 2025 - Feb. 2026 Amazon Frontier AI & Robotics Team Advisor: Prof. Pieter Abbeel and Dr. Rocky Duan 
Research Assistant Oct. 2024 - May 2026 LeCAR Lab, CMU RI Advisor: Prof. Guanya Shi 
Research Assistant Jan. 2024 - Jul. 2024 Tsinghua Embodied AI Lab (TEA Lab) Advisor: Prof. Huazhe Xu Education
Carnegie Mellon University Aug. 2024 - May 2026 M.S. in Robotic Systems Development 
Shanghai Jiao Tong University Sep. 2019 - Jun. 2023 B.Eng. in Automation Research
(* denotes equal contributions, † indicates equal advising)Humanoids
RPL: Learning Robust Humanoid Perceptive Locomotion over Challenging Terrains
In Submission
Links: [arXiv][Website][Twitter]
▶ TL;DR
RPL enables robust humanoid perceptive locomotion through a unified multi-depth policy that handles challenging terrains (slopes, stairs, stepping stones), multi-directional movements, payloads.
Perceptive Humanoid Parkour: Chaining Dynamic Human Skills via Motion Matching
RSS 2026
Links: [arXiv][Website][Twitter]
▶ TL;DR
PHP enables agile and long-horizon humanoid parkour with depth perception and motion matching for chaining skills.
FALCON: Learning Force-Adaptive Humanoid Loco-Manipulation
L4DC 2026 (Oral)
Links: [arXiv][Website][Code][Twitter]
▶ TL;DR
FALCON empowers humanoids with robust locomotion and precise manipulation under significant, unknown 3D end-effector forces, via a novel dual-agent force-adaptive RL framework.
Hold My Beer: Learning Gentle Humanoid Locomotion and End-Effector Stabilization Control
CoRL 2025
Links: [arXiv][Website][Code]
▶ TL;DR
A slow-fast dual-agent RL framework achieving human-level end-effector stability for humanoids.
ASAP: Aligning Simulation and Real-World Physics for Learning Agile Humanoid Whole-Body Skills
RSS 2025
Links: [arXiv][Website][Code][Twitter]
▶ TL;DR
ASAP learns agile whole-body humanoid motions via learning a residual action model from the real world to align sim-to-real physics and achieve athletic motions.
Mobile Manipulation
Catch It! Learning to Catch in Flight with Mobile Dexterous Hands
ICRA 2025CoRL 2024 Workshop (Best Paper Nomination)
Links: [arXiv][Website][Code][Twitter]
▶ TL;DR
We build a mobile manipulator with a dexterous hand, and leverage reinforcement learning to train a whole-body control policy for the robot to catch diverse objects randomly thrown by humans.
ViTaS: Visual Tactile Soft Fusion Contrastive Learning for Reinforcement Learning
ICRA 2026
▶ TL;DR
Introduced ViTaS, a framework using using Soft Fusion Contrastive Learning and a CVAE module to fuse visual and tactile information, achieving state-of-the-art performance with high sample efficiency and low parameter count in robotic manipulation.
Aerial Manipulation / Drones
Flying Hand: End-Effector-Centric Framework for Versatile Aerial Manipulation Teleoperation and Policy Learning
RSS 2025
Links: [Paper][Website]
▶ TL;DR
A unified aerial manipulation framework that enhances precision and versatility through an end-effector-centric interface.
Perception-constrained Visual Servoing Based NMPC for Quadrotor Flight
Undergraduate Thesis
Links: [Code][Video]
▶ TL;DR
Incorporated quadrotor dynamics and visual feature dynamics into NMPC to enable the quadrotor to flight purely based on visual information without localization.
Multi-Agent Path Planning
Multi-Agent Combinatorial Path Finding with Heterogeneous Task Duration
SoCS 2024
Links: [arXiv][Code][Video]
▶ TL;DR
Proposed two conflict-based search methods--CBSS-TPG (conflict-free) and CBSS-D (conflict-free and minimal cost) to solve the multi-agent combinatorial path finding problem with target duration.
Competitions
UAV Intelligent Perception Technology Competition Autonomous drone racing and perception
Team Lead National Third Prize (Top 10%)
National University IOT Design Competition HarClass intelligent classroom system
National University ICT Competition (Innovation) Edge intelligence and autonomous vehicle deployment
Leadership
SJTU-VEX Programming Team Leader
- 2021 National VEX Robotics Competition: Tournament Champions & Skills Champion (World Record)
- 2021 VEX Robotics Competition Asian Open: Tournament Champions VEXU; Excellence Award
- 2021 VEX Robotics Competition China Final: Tournament Champions VEXU; Excellence Award
Awards
- Outstanding Paper Nomination in LFDM Workshop at CoRL 2024
- SJTU Outstanding Graduate (top 3%)
- SJTU Merit Student (top 3%)
- SJTU Academic Progress Scholarship
Reviewer Service
IROS 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems
CoRL 2025 Conference on Robot Learning
ICRA 2025 IEEE International Conference on Robotics and Automation
IROS 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems