ROBOTICS · AI/ML · AUTONOMOUS SYSTEMS

Building intelligent systems
from models to motion.

I’m 张涵俐 Hanli Zhang, a robotics researcher at EPFL LASA under the supervision of Prof. Aude Billard, based in Lausanne, Switzerland.

My work connects learning, planning, perception, and control—then tests the result on real robotic systems.

Explore selected work Robotics résumé
Email LinkedIn GitHub
Hanli Zhang
Open to robotics & applied AI roles
83%lower simulated tracking error
49%lower simulated crash rate
40 FPSon Jetson Xavier NX
±5 mmAGV pose accuracy

SELECTED WORK

Evidence, not just keywords.

Projects spanning aerial robotics, embedded perception, robot learning, and visual SLAM—with demos, code, papers, and reports where available.

Hardware demo
01 Aerial robotics · UPenn GRASP

Drag-Aware Quadrotor Planning & Sim-to-Real Validation

Learned a tracking-cost model from 200,000 simulated trajectories and integrated it into trajectory optimization with an SE(3) controller, then validated the planner on Crazyflie hardware.

83%lower tracking error
49%lower simulated crash rate
PythonJAXTrajectory optimizationSim-to-real
Paper Code Full video
On-car demo
02 Robot perception · UPenn xLAB

F1TENTH Depth-Based Navigation

Built a monocular-depth perception and navigation stack combining MiDaS, rangefinder calibration, and depth-guided Follow-the-Gap control. Optimized inference with ONNX/TensorRT for edge deployment.

40 FPSJetson Xavier NX
ToF + RGBsensor fusion
OpenCVONNXTensorRTJetson
Code Report Full video
03 Visual SLAM

Gaussian Splatting SLAM with Loop Closure

Extended MonoGS with loop closure, Gaussian-aware covisibility, pose-graph optimization, global bundle adjustment, and CUDA rasterizer visibility metrics.

Up to 23%lower ATE RMSE on tested TUM RGB-D sequences
PyTorchCUDAOpenCV3DGS
Read project report
04 LLM planning

Task Planning & Evaluation for Robot Manipulation

Supervised and evaluated a planner using structured action primitives, feedback loops, recovery logic, and retune-vs-replan decisions.

85%success over 20 trials, with failure-mode analysis
LLM agentsEvaluationPlanningRecovery
Read project report

EXPERIENCE

Research depth. Field-tested engineering.

Experience across human-robot interaction, mobile robots, aerial systems, state estimation, and deployment debugging.

EPFL · LASALausanne, Switzerland

Robotics Researcher

Researching coordination-aware modeling for multi-limb human-robot interaction, combining stable dynamical-system motion primitives with cross-limb coupling.

LYF Innovation Ltd., Inc.Philadelphia, USA

Robotic Engineer Intern

Built a VINS-Fusion visual-inertial state-estimation pipeline with panoramic camera and IMU data, reducing outdoor tracking error by 15%.

Suzhou Beacon Robot Technology Co., Ltd.Suzhou, China

Algorithm Engineer

Improved AGV localization to ±5 mm pose accuracy, reduced pose-loss events by 30%, refactored 12 ROS/C++ modules, and resolved 30+ field issues.

TOOLKIT

Built for the full robotics stack.

01

Robotics

ROS1/2, SLAM, localization, mapping, state estimation, sensor fusion

02

Planning & control

Optimal control, motion planning, trajectory optimization, dynamical systems

03

AI & perception

PyTorch, JAX, TensorFlow, OpenCV, LLM planning and evaluation

04

Deployment

C/C++, Python, ONNX, TensorRT, NVIDIA Jetson, Linux, Docker, Git

ROBOTICS RÉSUMÉ

Robotics experience, one focused page.

The project, paper, code, and report links inside the PDF are clickable.

Robotics PDF ↗