Kelin Yu

Afford2Act: Affordance-Guided Automatic Keypoint Selection for Generalizable and Lightweight Robotic Manipulation

Anukriti Singh, Kasra Torshizi, Khuzema Habib, Kelin Yu, Ruohan Gao, Pratap Tokekar

Paper / Website / In Submision

AFFORD2ACT, an affordance-guided framework that distills a minimal set of semantic 2D keypoints from a text prompt and a single image. AFFORD2ACT follows a three-stage pipeline: affordance filtering, category-level keypoint construction, and transformer-based policy learning with embedded gating to reason about the most relevant keypoints, yielding a compact 38-dimensional state policy that can be trained in 15 minutes, which performs well in real-time without proprioception or dense representations.

ControlTac: Force- and Position-Controlled Tactile Data Augmentation with a Single Reference Image

Dongyu Luo*, Kelin Yu*, Amir-Hossein Shahidzadeh, Cornelia Fermuler, Yiannis Aloimonos, Ruohan Gao, (Dongyu is my undergraduate mentee)

ICCV CDEL Workshop 2025 ( Oral )
Paper / Website / In Submission

Vision-based tactile sensing has been widely used in perception, reconstruction, and robotic manipulation. However, collecting large-scale tactile data remains costly due to the localized nature of sensor-object interactions and inconsistencies across sensor instances. Existing approaches to scaling tactile data, such as simulation and free-form tactile generation, often suffer from unrealistic output and poor transferability to downstream tasks. To address this, we propose ControlTac, a two-stage controllable framework that generates realistic tactile images conditioned on a single reference tactile image, contact force, and contact position.

GenFlowRL: Shaping Rewards with Generative Object-Centric Flow in Visual Reinforcement Learning

Kelin Yu*, Sheng Zhang*, Harshit Soora, Furong Huang, Heng Huang, Pratap Tokekar, Ruohan Gao

ICCV 2025
ICRA FMNS Workshop 2025 ( Spotlight, Best Paper Nomination )
Paper / Website / Video

Recent advances have demonstrated the potential of video generation models to guide robot learning by deriving effective robot actions through inverse dynamics. However, these methods heavily depend on the quality of generated data and struggle with fine-grained manipulation due to the lack of environment feedback. While video-based reinforcement learning improves policy robustness, it remains constrained by the artifacts of generated video and the challenges of collecting large-scale in-domain robot datasets for training diffusion models. Motivated by the above, we propose GenFlowRL, which derives shaped rewards from generated flow trained from cross-embodiment datasets.

MimicTouch: Leveraging Multi-modal Human Tactile Demonstrations for Contact-rich Manipulation

Kelin Yu*, Yunhai Han*, Qixian Wang, Vaibhav Saxena, Danfei Xu, Ye Zhao

CoRL 2024
NeurIPS Touch Processing Workshop 2023 ( Best Paper Award )
CoRL Deployable Workshop 2023
Paper / Website / Video

We introduce MimicTouch, a novel framework for learning policies directly from demonstrations provided by human users with their hands. The key innovations are i) a human tactile data collection system which collects multi-modal tactile dataset for learning human's tactile-guided control strategy, ii) an imitation learning-based framework for learning human's tactile-guided control strategy through such data, and iii) an online residual RL framework to bridge the embodiment gap between the human hand and the robot gripper. Through comprehensive experiments, we highlight the efficacy of utilizing human's tactile-guided control strategy to resolve contact-rich manipulation tasks.

Continual Robot Learning via Language-Guided Skill Acquisition

Zhaoyi Li*, Kelin Yu*, Shuo Cheng*, Danfei Xu

ICLR AGI Workshop 2024 / ICLR LLMAgent Workshop 2024 / Website / Under Review by RA-L

We developed LG-SAIL (Language Models Guided Sequential, Adaptive, and Incremental Robot Learning), a framework that leverages Large Language Models (LLMs) to harmoniously integrate TAMP and DRL for continuous skill learning in long-horizon tasks. Our framework achieves automatic task decomposition, operator creation, and dense reward generation for efficiently acquiring the desired skills.

Learning Generalizable Vision-Tactile Robotic Grasping Strategy for Deformable Objects via Transformer

Yunhai Han*, Kelin Yu*, Rahul Batra, Nathan Boyd, Chaitanya Mehta, Tuo Zhao, Yu She, Seth Hutchinson, Ye Zhao

Transactions on Mechatronics 2024 (Long Version)
AIM 2023 (Short Version)
Paper / Code

Reliable robotic grasping with deformable objects remains a challenging task due to underactuated contact interactions with a gripper, unknown object dynamics, and variable object geometries. In this study, we propose a Transformer-based robot grasping framework for rigid grippers that leverage tactile information from a GelSight sensor for safe object grasping. The Transformer network learns physical feature embeddings from visual & tactile feedback and predict a final grasp through a multilayer perceptron (MLP) with grasping strength. Using these predictions, the gripper is commanded with an optimal force for safe grasping tasks.

Evolutionary Curriculum Training for DRL-Based Navigation Systems

Kelin Yu*, Max Asselmeier*, Zhaoyi Li*, Danfei Xu

RSS MultiAct Workshop 2023 / Website / Paper /

We introduce a novel approach called evolutionary curriculum training to tackle these challenges. The primary goal of evolutionary curriculum training is to evaluate the collision avoidance model's competency in various scenarios and create curricula to enhance its insufficient skills. The paper introduces an innovative evaluation technique to assess the DRL model's performance in navigating structured maps and avoiding dynamic obstacles. Additionally, an evolutionary training environment generates all the curriculum to improve the DRL model's inadequate skills tested in the previous evaluation. We benchmark the performance of our model across five structured environments to validate the hypothesis that this evolutionary training environment leads to a higher success rate and a lower average number of collisions.

Temporal Video-Language Alignment Network for Reward Shaping in Reinforcement Learning

Kelin Yu*, Ziyuan Cao*, Reshma Ramachandra*

Technical Report 2022 / Paper /

Designing appropriate reward functions for Reinforcement Learning (RL) approaches has been a significant problem. Utilizing natural language instructions to provide intermediate rewards to RL agents in a process known as reward shaping can help the agent in reaching the goal state faster. In this work, we propose a natural language-based reward-shaping approach that maps trajectories from Montezuma's Revenge game environment to corresponding natural language instructions using an extension of the LanguagE-Action Reward Network (LEARN) framework. These trajectory-language mappings are further used to generate intermediate rewards which are integrated into reward functions that can be utilized to learn an optimal policy for any standard RL algorithms.

Work Experience

Robotics Software Engineering intern

Amazon Robotics AI /

Designed and built Calibration Drift Detector for our Industrial manipulator with Python, Open3D, OpenCV, and machine learning classifier. Each multi-pick detected as a single pick costs 12$, and MEP/DEP package costs $0.05. My system saves potential thousands of dollars every day. Used AWS tool (S3) to get past images and point clouds and implemented advanced computer vision algorithms and applied ML classifier to detect calibration drift.

Projects

iValet An Intelligent Parking Lot Management System and Interface

Kelin Yu*, Faiza Yousuf*, Wei Xiong Toh*, Yunchu Feng*,

Senior Design 2022 / Website /

The iValet intelligent parking lot management system automatically directs drivers to the nearest vacant parking spot upon entering a crowded parking lot. The system consists of a camera, machine learning development board, a PostgreSQL server, and a user interface (web application). The camera is used to take photos of the entire parking lot, the development board runs segmentation and classification algorithms on those photos, the SQL server contains data about each parking spot that is written to by the image processing models, a path-planning algorithm, and the UI, while the web application shows end users the directions to the empty parking spots based on the location of the parking lot entrance.

Awards

ICRA FMNS Workshop Best Paper Nomination 2025, Dean’s Fellowship, University of Maryland, College Park, 2024-2026,

NeurIPS TouchProcessing Workshop Best Paper Award 2023,

CoRL Travel Grant 2023,

Graduate with Highest Honor at Georgia Tech 2022,

Professional Service

ICCV 2025, Reviewer

NeurIPS 2025, Reviewer

RSS 2025, Reviewer

ICML 2025, Reviewer

ICRA 2025, Reviewer

ICLR 2025, Reviewer

NeurIPS 2024, Reviewer

CoRL Deployable Workshop, 2023, Reviewer

Teaching Experience

CMSC 848M: Multimodal Computer Vision, University of Maryland, Spring 2025 CMSC 421, Introduction to Artificial Intelligence, University of Maryland, Fall 2024 CS 4476/6476, Computer Vision, Georgia Tech, Spring 2024, Fall 2023, Spring 2023, Fall 2022 ECE 3741, Instrumentation and Electronics Laboratory, Georgia Tech, Spring 2021