Yu Xia

M.S. Robotics @ UC Riverside · Robot Manipulation · Teleoperation · Deep Reinforcement Learning · Human-Robot Interaction · Extended Reality (XR)

✉ yxia072@ucr.edu 📍 Riverside, CA GitHub
📄 Download CV (PDF)

Education

University of California, Riverside
Master of Science in Robotics
Riverside, CA, USA
Sep 2023 – Jun 2025
Wuhan Huaxia University of Technology
Bachelor of Engineering in Computer Science and Technology
Wuhan, China
Sep 2017 – Jun 2021

Publications

BOLT: Bandwidth Efficient Object State Streaming for XR Robot Teleoperation
Jun 2026
MobiCom 2027 — Under Review
Yu Xia, Hang Qiu
  • Designed an XR 3D scene streaming system that replaces repeated complete-scene transmission with compact object state updates backed by reusable geometry, using mesh-backed object ID/6DoF pose updates and segmented RGB-D fallback until runtime mesh promotion.
  • Built a complete XR teleoperation prototype; across five live runs, single-L515 perception uplink with masked fallback stayed within 0.22–0.34 Mbps, at least 120× below 41.5 Mbps point-cloud streaming for the same task, and GMC reduced median ADD-S from 18.8 mm to 5.5 mm under 50 mm translation noise.
Towards Generalizable Safety in Crowd Navigation via Conformal Uncertainty Handling
Sep 2024
CoRL 2025
Jianpeng Yao, Xiaopan Zhang*, Yu Xia*, Zejin Wang, Amit K. Roy-Chowdhury, Jiachen Li
  • A conformal inference–guided constrained RL framework for crowd navigation, achieving 96.93% success rate and 3.72× fewer collisions with robust out-of-distribution performance.
  • Led the real-robot deployment: built the ROSMASTER X3 sim-to-real stack with LiDAR-based pedestrian detection feeding real-time 3D positions to the RL policy, and conducted crowd-interaction experiments in both sparse and dense environments.
NL2SpaTiaL: Generating Geometric Spatio-Temporal Logic Specifications from Natural Language for Manipulation Tasks
Nov 2025
IROS 2026
Licheng Luo, Kaier Liang, Yu Xia, Mingyu Cai
  • A hierarchical translation framework that converts natural language into verifiable geometric spatio-temporal logic (SpaTiaL) specifications for manipulation tasks.
  • Conducted early-stage real-robot validation on ALOHA 2, evaluating π0-generated manipulation trajectories against translated SpaTiaL specifications to verify spatio-temporal constraint satisfaction.

Robot Hardware Experience

TurtleBot 4
Differential-Drive Mobile Robot
  • Differential-drive mobile robot with iRobot Create 3 base, Raspberry Pi 4, OAK-D stereo camera, and RPLIDAR A1M8 2D LiDAR.
  • Used in earlier real-world experiments of the crowd navigation work (CoRL 2025), running the same LiDAR-based pedestrian detection pipeline for real-time avoidance in dynamic indoor environments.
ROSMASTER X3
Omnidirectional Mobile Robot
  • Mecanum-wheel omnidirectional mobile robot with Jetson Orin, SLAMTEC LiDAR, and Astra depth camera.
  • Primary sim-to-real platform for the CoRL 2025 crowd navigation deployment: LiDAR-based pedestrian detection provides real-time 3D pedestrian positions to the RL policy for avoidance in both sparse and dense crowds.
ALOHA 2
Bimanual Teleoperation System
  • Low-cost bimanual teleoperation system with 4× Trossen ViperX/WidowX 6-DoF arms, leader-follower control, and multi-view RGB cameras.
  • Used for early-stage real-robot evaluation in the NL2SpaTiaL project (IROS 2026): action trajectories generated by π0 are evaluated against translated SpaTiaL formulas to verify whether executed manipulations satisfy the spatio-temporal specifications.
Kinova Gen3
7-DoF Collaborative Arm
  • 7-DoF ultra-lightweight collaborative arm (8.2 kg, 902 mm reach, 4 kg payload) with integrated torque sensors and 2D/3D vision module.
  • Serving as the manipulation execution end for an ongoing VR teleoperation project. Integrating Kortex API with WebRTC-based remote control for real-time manipulation over WAN.
Hexapod (DIY)
Legged Locomotion Platform
  • Custom-built six-legged robot for RL-based locomotion research, including gait training and flip-recovery behavior.
  • Trained RL locomotion and fall-recovery policies in simulation. Developed the real-robot controller for deployment; sim-to-real transfer pending due to hardware actuator limitations.

Project Experience

Robotics Fabrication: Ball Balancing Car
Apr 2024
  • Role: Perception & Control Developer. Built subsystems for sensing, wireless communication, and PID motor control for real-time ball balancing.
ArduinoPIDIMUBluetooth
UAV Path Planning and Obstacle Avoidance
Dec 2023
  • Role: Simulator & Communication Developer. Set up simulator, configured UAV sensors, established ROS–PX4 communication for RL training.
  • Developed a rule-based baseline for navigating obstacles; contributed to RL reward function design to accelerate convergence.
ROSPX4GazeboPPO
Mask Wearing Detection with YOLOv5
Sep 2021
  • Trained and optimized YOLOv5 for face mask verification, achieving 85% accuracy on SF-MASK via iterative hyperparameter tuning.
YOLOv5PyTorchOpenCV
Image Classification System Design
May 2021
  • Enhanced AlexNet architecture and achieved 85% accuracy on CIFAR-10 through comparative experiments using TensorFlow.
TensorFlowAlexNetCIFAR-10

Technical Skills

Programming

PythonC/C++Java MATLABHTML5SQL

Robotics & Sim

ROS / ROS 2MoveIt 2Gazebo MuJoCoIsaac SimPX4Unity

AI / ML

PyTorchTensorFlowOpenCV MediaPipeYOLO

Tools

DockerGitLinux (Ubuntu) WebRTCSolidWorksBlenderPhotoshop

Robot Platforms

TurtleBot 4ROSMASTER X3ALOHA 2 Kinova Gen3Hexapod (DIY)UAV (PX4)