Chao Ni

I am a research intern at VILAB at EPFL, where I work on computer vision and reinforcement learning. My supervisor is Amir Zamir.

Before joining VILAB, I finished my master thesis at Robotics System Lab at ETH Zurich supervised by Marco Hutter. During my study at ETH Zurich, I also have collaboration with Roland Siegwart. I obtained my bachelor degrees in Applied Mathematics and Economics at Peking University.

I am interested in the reasearch area bridging perception, learning and control for robotics.

Email  /  CV  /  Twitter  /  Github

profile photo
Ongoing Projects

Here is a few projects that I am currently working on, with some of them avaliable in preprints.

Learning to Walk Over Structured Terrains by Imitating MPC
Chao Ni, Alexander Reske, Takahiro Miki, Ruben Grandia, Jan Carius, Marco Hutter
video / preprint

Leveraging demonstrations from MPC expert, the robot learns to walk over structured terrains.

Learning Sampling Based Exploration Planning
Lukas Schmid*, Chao Ni*, Yuliang Zhong, Sirish Srinivasan,
Cesar Cadena, Roland Siegwart, Olov Andersson
video / preprint

By learning the distribution of optimal samples given a local map, sampling-based exploration planning can be done efficiently.

Past Projects

Here is a list of my past projects, including course projects.

MPC-feedback Trajectory Optimization for Wheeled-legged Robots
Chao Ni
semester thesis
report / video / code

We create a motion primitive library with trajectories generated by modulizable optimizers and use Model Predictive Control to track the trajectory.

Simple Hexapod Robot Control
Chao Ni, Kaiyue Shen, Ji Shi
course project
video / code

Inverse Kinamtics solver for the hexpod robot is implemented, simple PD feedback is used for torque control. An obstacle avoidance algorithm is used to achieve navigation tasks.

Exploiting Effective Representation via Cooperative Learning of Multi-Sensory Robotics Data
Chao Ni
undergraduate thesis

We extract effective representation from the multi-sensory robotics data by self-supervised synchronization and use the latent representation for downstream RL tasks.

Template is credit to source code.