HTTP/2 200
server: GitHub.com
content-type: text/html; charset=utf-8
last-modified: Thu, 06 Nov 2025 21:40:12 GMT
access-control-allow-origin: *
strict-transport-security: max-age=31556952
etag: W/"690d15bc-1c1b"
expires: Sun, 28 Dec 2025 15:02:22 GMT
cache-control: max-age=600
content-encoding: gzip
x-proxy-cache: MISS
x-github-request-id: 797E:1F53DD:7BB928:8AB1BA:69514425
accept-ranges: bytes
age: 0
date: Sun, 28 Dec 2025 14:52:22 GMT
via: 1.1 varnish
x-served-by: cache-bom-vanm7210074-BOM
x-cache: MISS
x-cache-hits: 0
x-timer: S1766933542.909965,VS0,VE220
vary: Accept-Encoding
x-fastly-request-id: c2ef39f32162d276e8ec37a20cddfd8bb3991fa6
content-length: 2238
Kevin Kim
Kevin Kim
kimkj [at] usc [dot] edu
I am a third-year undergraduate student at the University of Southern California, studying Computer Science and Applied Mathematics. My research is done under the guidance of Professors Erdem Bıyık and Daniel Seita.
Policy-agnostic Extraction of Essential Keypoints for zero-shot robot manipulation. Fine-tunes VLMs to predict unified point-based intermediate representations: end-effector paths and task-relevant masks, enabling 41.4× real-world improvement for 3D policies.
Fast robot adaptation via hand path retrieval from task-agnostic play data. Uses human hand demonstrations to retrieve relevant robot behaviors, enabling real-time learning of tasks in under four minutes with 2x improvement over baselines.