Tianyu Tu

I'm a Master's student at Stanford University, pursuing a degree in Mechanical Engineering with a focus on robot perception, tactile sensing, and AI for robotics. I'm currently working with Professor Mark Cutkosky at the Biomimetics & Dexterous Manipulation Laboratory (BDML) on whisker-inspired tactile sensing for underwater robotics.

My research interests lie in enhancing robot perception through bio-inspired sensors, active vision, and learning-based methods. I focus on enabling robots to extract meaningful insights from limited sensory data and adapt to changing environments—much like how biological systems perceive and respond to their surroundings.

I received my B.E. in Mechanical Engineering from Shanghai Jiao Tong University as part of the prestigious Tsien Hsue-Shen Honor Program. Born and raised in southern China, I maintain a balance between technical studies and artistic pursuits—I've been a tenor section leader in the university choir and continue to play piano and guitar.

Email  /  CV  /  Bio  /  GitHub

Research

My research focuses on robot perception—teaching robots to "see" and "understand" their environment using sensors like cameras and tactile arrays. I'm particularly interested in active vision systems that adjust to improve perception, bio-inspired sensors that mimic natural sensing mechanisms, and learning-based methods that enable robots to make quick, reliable decisions in dynamic environments. By ensuring robots can effectively interpret limited information, we empower them to adapt and respond with precision.

Publications

A wearable fiber Bragg grating (FBG) tactile sensor system that enables seamless skill transfer from human demonstrations to robotic manipulation through tactile-based learning.

A novel whisker-inspired tactile sensing modality using fiber Bragg grating optical strain sensors for underwater flow detection and hydrodynamic perception in aquatic robots.

Research Projects

Developing a novel tactile sensing modality inspired by animal whiskers, integrating FBG optical strain sensors with compliant beams for underwater flow detection. Achieved >8× improvement in signal-to-noise ratio through experimental design and signal processing. Designed and built an embedded experimental platform with stepper-motor actuation and real-time data acquisition.

Developed a robust large-range real-time mapping technique for rotorcraft robots using probability grid mapping, YOLOv5, and DBSCAN clustering to represent obstacles. Introduced a multi-agent collaborative triangulation method for distant obstacle mapping. The system automatically adapted to RGB-D camera parameters during dynamic flight and was validated through simulations and real-world flight tests.

Analyzed calendering force effects on electrode sheets via Abaqus simulations and literature review. Developed three innovative corrugation mitigation methods inspired by belt gearing and hydraulics. Designed a novel multi-stage calendering apparatus with differential roller diameters for precise velocity control—now patent-pending.

Course Projects

Preprocessed athlete motion data with traditional computer vision algorithms (Hough transform, polynomial fitting) to extract interpretable features from raw video. Combined visual features with audio spectrogram filtering to identify key frames. Achieved >20% accuracy improvement in shot direction prediction by incorporating spline-fitted foot trajectories as logistic regression model inputs.

Implemented an end-to-end embedded architecture using ESP32-CAM for real-time data capture, UDP protocol for low-latency transmission, and NVIDIA TX2 for model inference. Trained and deployed a YOLOv5 gesture-recognition model, bridging perception algorithms with embedded hardware systems.

Designed a vision pipeline using Sobel-based lane boundary detection and non-uniform centerline sampling to compute lateral offset. Integrated perception with PD steering control and speed regulation, with fallback bang-bang control for steep slopes. Secured 3rd place in campus camera group (8 teams).

Ski-Teaching Robot: Conducted motion analysis to design a robot mimicking skiing techniques. Engineered a mechanism with linkages and planetary gears for accurate arm movement replication.
Jumping Robot: Developed a novel jumping mechanism using a silicone gel bowl to enhance takeoff dynamics, validated through calculations, finite element analysis, and testing.

Website template from Jon Barron. Last updated: October 2025.