Robotics · Sensing · Learning
M.S.E. in Mechanical Engineering and Applied Mechanics
University of Pennsylvania
I work across tactile sensing, deep learning, and robotic systems. My experience spans data collection, model training and evaluation, hardware prototyping, and efficient LLM inference.
I want to build robots that use sensory feedback to manipulate more reliably, adapt to changing conditions, and recover from mistakes.
Tactile ManipulationLearning to manipulate with tactile feedback.
Reinforcement Learning for Adaptive Robot SkillsLearning motion and manipulation skills for changing conditions.
IROS 2026 · Accepted · Second author
Estimating a soft robot’s changing shape from a flexible sensor array, without camera input during inference.
My work. Sensor acquisition software, experimental data collection, geometric preprocessing, model training, and evaluation.
4.7° bending / 4.9° twisting mean angle error · reported in the paper
Less storage for the KV cache
73.4%reduction vs. FP16 (approx.)
Undergraduate thesis · Independent research
Reducing LLM inference storage while preserving attention behavior and long-context retrieval.
My work. Independently developed the calibration, quantized cache, Triton decoding kernels, and evaluation pipeline.
Role-aware calibration for keys and values · evaluated across six instruction models
Pressure sensing, Arduino control, and prototype integration.
STM32 motor control, PID tuning, and team leadership. Second in the course competition.
Robot fabrication and mechanism assembly. National Third Prize with the SCUT team.
Accepted to IEEE/RSJ IROS 2026.
Published in IPMLP 2025 · Proceedings published by ACM.
