Research

Publications & competitions

• m2sv: A Scalable Benchmark for Map-to-Street-View Spatial Reasoning — Yosub Shin, Michael Buriek, Igor Molybog. CTB @ ICML 2026 Workshop. arXiv:2601.19099
• What Matters in Data Curation for Multimodal Reasoning? Insights from the DCVLR Challenge — Yosub Shin, Michael Buriek, Boris Sobolev, Pavel Bushuyeu, Vikas Kumar, Haoyang Xu, Samuel Watson, Igor Molybog. EMM-QA @ ICML 2026 Workshop. arXiv:2601.10922
• NeurIPS 2025 — DCVLR (Data Curation for Vision-Language Reasoning) Competition: Winner (1st place). Curated a vision-language reasoning dataset achieving top performance across benchmarks; presented at the NeurIPS competition workshop. Press coverage
• Beyond Audio and Pose: A General-Purpose Framework for Video Synchronization — Yosub Shin, Igor Molybog. arXiv preprint, 2025. arXiv:2506.15937

Selected projects

• Dexterous cable insertion — Intrinsic AI for Industry Challenge,
  1. End-to-end LeRobot pipeline (Gazebo, ROS 2) for contact-rich cable insertion; 1K+ tier-filtered episodes via DART corrective rollouts; ~50% full-insertion rate in sim.
• Bilateral teleoperation on SO-101 — dual-arm LeRobot extension with haptic contact feedback, motion assist, and gravity hold on low-cost servos. github.com/YosubShin/lerobot/tree/bilateral-teleop
• SO-101 field notes — the two-part series on this site: training a diffusion policy on a $300 arm and measuring the evaluation itself.
• koa-cli — open-source multi-cluster Slurm submission tool. github.com/YosubShin/koa

Resume (PDF)