Hojin Lee

Research profile

Research summary

Hojin Lee is a robotics researcher at the Ulsan National Institute of Science and Technology (UNIST), South Korea, working at the intersection of robot learning and field robotics, with a recent focus on uncertainty-aware perception for agile platforms. Two first-author arXiv preprints from 2026 anchor the Evolve catalog: a kernel-metric-regularized deep kernel learning approach for predicting opponent vehicle trajectories with uncertainty in 1/10-scale autonomous racing, and a generative-replay continual learning method for traversability prediction that limits catastrophic forgetting without storing raw data. Both were reviewed by Evolve as methodologically plausible with medium reproducibility readiness (mean 55 across 2 papers). The OpenAlex record shows an h-index of 8 and i10-index of 7, 10 works in the last three years, and an upward output trend (2 works in 2023, 4 in 2024, 6 in 2025). Topical coverage is broad, spanning distributed control of multi-agent systems, wireless networks and protocols, network traffic and congestion control, reinforcement learning in robotics, and topic modeling, with the field distribution split evenly between field robotics/robot learning and autonomous racing/motion prediction. Collaboration is compact: 5 distinct co-authors, with recurring partners including Cheolhyeon Kwon, Yunho Lee, Daniel A Duecker, Youngim Nam and Sanghun Lee. Output and impact dimensions are not yet scored due to missing verified aggregates.

Research keywords

Highlights

• Uncertainty-aware opponent trajectory prediction for autonomous racing

  First-author 2026 arXiv work regularizing deep kernel learning with kernel-metric learning to predict opponent vehicle trajectories with uncertainty on a 1/10-scale racing platform; Evolve notes strong hardware results but thin statistics (reproducibility readiness: medium).

• Continual learning for traversability prediction without storing data

  First-author 2026 arXiv work using generative replay to cut catastrophic forgetting in traversability prediction, with uncertainty-aware adaptation for off-road navigation; Evolve judges real-robot gains plausible but flags navigation trials at n=3 and unreported λ/τ hyperparameters (reproducibility readiness: medium).

Timeline

1. 20244 works recorded for this profile in OpenAlex during 2024, up from 2 in 2023.
2. 20256 works recorded in 2025 — the highest single-year count in the available countsByYear series (1998–2025).
3. 2026First-author arXiv preprint on kernel-metric-regularized deep kernel learning for uncertain opponent vehicle trajectory prediction in 1/10-scale autonomous racing; Evolve notes strong hardware results but thin statistics.
4. 2026First-author arXiv preprint on generative-replay continual learning for traversability prediction with uncertainty-aware adaptation; Evolve judges real-robot gains plausible but flags n=3 navigation trials and unreported λ/τ.

Papers in the Evolve catalog

• 2026-09-15first author

  Kernel-Based Metrics Learning for Uncertain Opponent Vehicle Trajectory Prediction in Autonomous Racing

• 2026-09-15first author

  Continual Learning for Traversability Prediction with Uncertainty-Aware Adaptation

Frequent co-authors