Curriculum Vitae

Zhejiang University, Hangzhou 2024.09 - 2027.06

• Lab: FastLab (Fire Group)
• Supervisors: Prof. Yanjun Cao and Prof. Chao Xu
• Research Focus: Trajectory Planning & Optimization, Reinforcement Learning, End-to-End Autonomous Navigation
• Honors: Outstanding Graduate Student of Zhejiang University (2024-2025); Outstanding Graduate Student of ZJU Huzhou Institute (2025)

Harbin Institute of Technology, Harbin 2020.09 - 2024.06

• GPA: 93.13 / 100  |  Rank: 5 / 298
• Honors: National Scholarship; First Prize in National Intelligent Car Competition; SMC First-Class Scholarship; People's Scholarship (multiple times)

[1] TOP: Trajectory Optimization via Parallel Optimization towards Constant Time Complexity

Proposed an ADMM-based parallel trajectory optimization framework that decomposes trajectories into independent sub-problems, reducing per-iteration time complexity to O(1). Achieved >10× speedup over serial SOTA on 100-segment trajectories; GPU deployment enables millisecond-level optimization for 1000-segment trajectories.

[2] Learning Safety-enhanced Navigation with Integrated Model Information

Proposed an end-to-end visual navigation framework integrating differentiable physics engine with safety-constrained policy optimization. Leveraged PALM for near-KKT convergence in a single loop. Achieved zero-shot sim-to-real transfer across differential-drive, tracked, and quadruped platforms with zero collision rate in dense environments.

[3] ATRS: Adaptive Trajectory Re-splitting via a Shared Neural Policy for Parallel Optimization

Proposed embedding a shared deep RL agent into the ADMM optimization loop to dynamically restructure trajectory segments, eliminating the optimization bottleneck. Modeled adaptive splitting as a MASP-MDP with a shared Actor-Critic architecture, generalizing across arbitrary segment counts. Achieved 26% fewer iterations and 19.1% less computation time versus baselines; real quadrotor traversing unknown forests in 35 ms.

[4] Whole-body Planning for Any-Shape Robot directly in Point Cloud

Proposed a dual-layer framework for whole-body trajectory optimization of arbitrary-shape robots directly in raw point clouds. Leveraged convex decomposition for differentiable signed-distance constraints with ADMM-based parallel solving. Full GPU pipeline completes in 19.86 ms in narrow mixed environments; real quadrotor with LiDAR navigates unknown environments in real time.

[5] CoNiPA: Cooperative Non-inertial Control Framework with LSTM-Enhanced Predictive Awareness

Proposed an active perception-aware control framework for GPS-denied air-ground cooperation. Unified UAV trajectory and gimbal orientation optimization under non-inertial dynamics via MPC. LSTM-based time-varying IMU prediction compensates for model mismatch. Achieved >98% target visibility and 20 cm tracking accuracy in aggressive maneuvers across simulation and real-world experiments.

Air-Ground Cooperation without Global Information IROS 2025 EXPO

• RoFly and CubeTrack cooperation with CREPES and CoNi-MPC for GPS-denied autonomous air-ground systems.
• Achieved real-time relative pose estimation, non-inertial trajectory tracking, and fleet coordination across multi-scenario real-robot demonstrations.

Autonomous UAV Inspection System for Power Substations Enterprise Project

• Developed a micro UAV (<800 g, >15 min endurance) autonomous inspection system for complex indoor substation environments.
• Implemented GPS-denied real-time visual localization via onboard ORB-SLAM; equipped with thermal and RGB cameras for equipment temperature monitoring and instrument reading detection.
• Achieved full-coverage path planning with interest-point-guided global coverage and local obstacle-avoidance trajectory optimization. Successfully deployed and tested on-site.