Joochan Kim

I am an first year PhD student at Purdue ECE, advised by Ziran Wang. I was a research intern at KIST in Seoul, working on Embodied AI, with a focus on Vision-Language-Action models, under the mentorship of Hwasup Lim and Tackgeun You.

I received my M.S. from SNU, advised by Byoung-Tak Zhang. During my master's program, I also interned at SIMTech, A*STAR, where I was mentored by Haiyue Zhu. I received my B.S. from Yonsei University.

Profile photo of Joochan Kim

Research

I'm interested in Embodied AI, Multimodal AI, and data-centric AI. Most of my research is about developing embodied AI from internet AI. Some papers with highlight are papers with main contributions.

Go2-DrivoR real-world trajectory prediction example
Taming an End-to-End Autonomous Driving Policy for Urban Navigation of Quadruped Robots
Joochan Kim*, Chanuk Yang*, Tackgeun You, Ziran Wang, Hwasup Lim
IROS Workshop on AI Meets Autonomy, 2026
Paper Poster

We present Go2-DrivoR, a goal-conditioned adaptation of the end-to-end autonomous driving planner DrivoR for urban quadrupedal navigation. Trained exclusively on TartanGround simulation data, it improves goal-conditioned planning on unseen simulated environments and transfers zero-shot to open-loop real-world Unitree Go2 trajectory prediction.

The Losing Winner paper figure
The Losing Winner: An LLM Agent that Predicts the Market but Loses Money
Youwon Jang*, Joochan Kim*, Byoung-Tak Zhang
NeurIPS Workshop on Generative AI in Finance, 2025
Paper Poster

Fine-tuning an LLM for Bitcoin market state prediction improves accuracy but paradoxically worsens trading returns, exposing the dangers of proxy objectives and reward hacking in financial AI.

Continual Vision-and-Language Navigation paper figure
Continual Vision-and-Language Navigation
Seongjun Jeong, Gi-Cheon Kang, Seongho Choi, Joochan Kim, Byoung-Tak Zhang
BMVC, 2025
Paper

We propose Continual Vision-and-Language Navigation (CVLN) paradigm along with two methods for CVLN: Perplexity Replay (PerpR) and Episodic Self-Replay (ESR).

Ordinal bias in action recognition paper figure
Exploring Ordinal Bias in Action Recognition for Instructional Videos
Joochan Kim, Minjoon Jung, Byoung-Tak Zhang
ICLR Workshop on Spurious Correlation and Shortcut Learning, 2025
arXiv Poster

Ordinal bias leads action recognition models to over-rely on dominant action pairs, inflating performance and lacking true video comprehension even when challenged by action masking and sequence shuffling.

BM-DETR paper figure
Background-aware Moment Detection for Video Moment Retrieval
Minjoon Jung, Youwon Jang, Seongho Choi, Joochan Kim, Jin-Hwa Kim, Byoung-Tak Zhang
WACV, 2025
arXiv Code

We propose Background-aware Moment Detection TRansformer (BM-DETR), which carefully adopts a contrastive approach for robust prediction. BM-DETR achieves state-of-the-art performance on various benchmarks while being highly efficient.

VLN-CM paper figure
Zero-Shot Vision-and-Language Navigation with Collision Mitigation in Continuous Environment
Seongjun Jeong, Gi-Cheon Kang, Joochan Kim, Byoung-Tak Zhang
CVPR Workshop on Embodied AI, 2025
arXiv

We propose the zero-shot Vision-and-Language Navigation with Collision Mitigation (VLN-CM), which takes low-level actions as an output while considering possible collisions.

MPGN paper figure
Modal-specific Pseudo Query Generation for Video Corpus Moment Retrieval
Minjoon Jung, Seongho Choi, Joochan Kim, Jin-Hwa Kim, Byoung-Tak Zhang
EMNLP, 2022
arXiv

We propose a self-supervised learning framework: Modal-specific Pseudo Query Generation Network (MPGN). First, MPGN selects candidate temporal moments via subtitle-based moment sampling. Then, it generates pseudo queries exploiting both visual and textual information from the selected temporal moments.

Miscellanea

Academic Service
Teaching
Teaching Assistant, M1522.000300 Spring 2023