Steven Oh

Email: oh.steven [at] fuji.waseda.jp

I am a fourth-year undergraduate student at Waseda University, majoring in Mechanical Engineering with a minor in Computer Science. I am currently a visiting scholar at Carnegie Mellon University, working with Prof. Deepak Pathak and Prof. Nancy Pollard. At Waseda University, I conduct undergraduate research under the supervision of Prof. Tetsuya Ogata and Prof. Shigeki Sugano.

Previously, I was a robotics research intern at OMRON SINIC X, and a visiting scholar at Imperial College London, advised by Dr. Dandan Zhang. During my high school years, I participated in a robotics research project at RIKEN under the guidance of Prof. Shingo Shimoda.

LinkedIn /

Life Updates

[November 2025] -- I will be joining CMU as a visiting scholar working with Prof. Nancy Pollard and Prof. Deepak Pathak.

Research

I am broadly interested in compliance control, tactile sensing, and contact-rich manipulation via behavior cloning.

	Refinement of Accelerated Demonstrations via Incremental Iterative Reference Learning Control for Fast Contact-Rich Imitation Learning Koki Yamane, Cristian C. Beltran-Hernande, Steven Oh, Masashi Hamaya, Sho Sakaino Under review at ICRA 2026 paper We propose I2RLC to safely refine accelerated demonstrations for fast, contact-rich manipulation by progressively increasing execution speed while iteratively reducing tracking error.
	CLAW: A Soft Wrist with Anisotropic and Selective Stiffness for Safe Robot Learning in Contact-rich Manipulation Steven Oh, Tomoya Takahashi, Cristian Camilo Beltran-Hernandez, Yuki Kuroda, Masashi Hamaya Under review at RAL CoRL 2025 Open Hardware Workshop paper / website CLAW is a novel soft wrist mechanism providing large 6-DoF deformation and mode-switchable anisotropic stiffness for robust contact-rich manipulation.
	TacEva: A Performance Evaluation Framework for Vision-Based Tactile Sensors Qingzheng Cong, Steven Oh*, Wen Fan, Shan Luo, Kasper Althoefer, Dandan Zhang Advanced Intelligent Systems (JCR Q1, IF: 6.1) arxiv / website TacEva is a performance evaluation framework for vision-based tactile sensors that provides standardized benchmarks and metrics for assessing sensor capabilities in contact-rich manipulation tasks.
	MicCheck: Repurposing Off-the-Shelf Pin Microphones for Easy and Low-Cost Contact Sensing Steven Oh, Tai Inui, Magdeline Kuan, Jia-Yeu Lin IEEE International Automatic Control Conference* Best Presentation Award paper We repurpose low-cost pocket mic for robot learning and material classfication. We show imitation learning system equipped with pocket mic achieves higher success rate in contact-rich tasks.

Projects

This includes coursework, side projects, and engineering work.

	Fine tuning VLA models on Torobo Humanoid Robots Internship project with Tokyo Robotics video Developed a ROS-based data collection pipeline and an AWS SageMaker training pipeline, and fine-tuned Octo for pick-and-place manipulation tasks.
	Unsupervised Slide Quality Assessment via Designer Cue Augmentation Course Project arxiv An unsupervised slide quality assessment pipeline combining seven expert-inspired visual design metrics with CLIP-ViT representations.
	Robot sketching system with vector flow Research project at Toyota-RIKEN collaboration center Designed robotic pencil sketching system using vector flow-based algorithm to process internet images and torque control to draw high-fidelity pencil sketch.

Life Updates

Research

Refinement of Accelerated Demonstrations via Incremental Iterative Reference Learning Control for Fast Contact-Rich Imitation Learning

CLAW: A Soft Wrist with Anisotropic and Selective Stiffness for Safe Robot Learning in Contact-rich Manipulation

TacEva: A Performance Evaluation Framework for Vision-Based Tactile Sensors

MicCheck: Repurposing Off-the-Shelf Pin Microphones for Easy and Low-Cost Contact Sensing

Projects

Fine tuning VLA models on Torobo Humanoid Robots

Unsupervised Slide Quality Assessment via Designer Cue Augmentation

Robot sketching system with vector flow