Ri-Zhao (Roger) Qiu

I am a PhD student at UCSD, working with Prof. Xiaolong Wang. Before this, I received my BS/MS in mathematics and computer science at UIUC, where I worked with Prof. Kris Hauser and Prof. Yuxiong Wang.

I love full-stack engineering. During my undergrad, I built things from flip-flops to a RISC-V CPU, from hand-written backprop in C++ to TensorRT, from STM32 to NVIDIA Jetson, and spent a happy year forming the Illini RoboMaster club in 2018.

Google Scholar  /  Twitter  /  Github  /  Cat Senpai

My research interests lie in the intersection of robotics and computer vision. My research aims at enhancing the mobile manipulation capabilities of robots via CV methods.

Egocentric human demonstrations as a data source for humanoid manipulation.

Memory attention to substitute feature distillation.

Long horizon loco-manipulation with LLM planner and learned wholebody skills.

Splatting features enable precise and efficient grasping in dynamic environments.

Synthesizing physics-based particle object motion via feature distillation and MPM.

A sim-to-real framework that enables the robot to grasp different objects in varying surroundings.

Align diffusion models for generation of images and masks of uncommon classes via few-shot conditioning.

Real-time Generalizable Feature Fields enable Mobile Manipulation.

We design a method that enables continual segmentation learning with few annotations via a guided data synthesis process.

We design a disinfection system that can priortize cleaning of high-touch surfaces like human cleaners by co-designing perception and planning algorithms.

We designed and built robots from bolts and metal to perception algorithm. I mainly helped with writing STM32 application software and perception code.

I implemented a codebase for robotics motion planning, which supports common problems like 2D narrow passage, bug traps, and robot arm grasping. It implements a variety of algorithms including RRT, RRT*, bRRT, PRM, FMT. In addition to uniform sampling, it also supports non-uniform sampler using conditional autoencoder.

Borrowed from Jon Barron's source code.