HYHyeonseo Yang

Hyeonseo Yang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Hyeonseo Yang has a total of 7 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Towards object-centric spatial abstraction for probabilistic environment understanding

    by Won-Seok Choi, Dong-Sig Han, Suhyung Choi, Hyeonseo Yang, Byoung-Tak Zhang

  • Open Access

    CraftGround: A High-Performance Reinforcement Learning Environment Based on Minecraft

    by Hyeonseo Yang, Minsu Lee, Byoung-Tak Zhang

    Recent trends in reinforcement learning have focused on developing agents that can operate in open-ended, multimodal environments and can be generalized to real-world tasks. Minecraft offers a rich, procedurally generated world that supports hierarchical skill learning, long-term planning, and complex decision-making, making it a compelling testbed for advancing generalizable reinforcement learning research. However, existing Minecraft-based frameworks such as Malmö, despite their early contributions, suffer from slow simulation speeds, outdated dependencies, and limited extensibility–posing significant challenges for modern, large-scale reinforcement learning research. To address these limitations, we introduce CraftGround, a next-generation high-performance Minecraft-based reinforcement learning benchmark environment. Rather than simply improving upon Malmö, CraftGround is designed from the ground up for scalability, multimodal interaction, and experimental flexibility. It achieves up to 2.79× faster simulation performance and supports sound subtitles for richer perception. It also provides a unified action configuration that allows seamless switching between discrete and human-like controls without reconfiguration. These advancements make CraftGround a robust and future-ready platform for complex and realistic reinforcement learning research. We further assess its usability through two RL experiments, showing that CraftGround supports practical learning in representative settings. The environment is open-sourced under the GNU General Public License v3.0 and publicly available at https://github.com/yhs0602/CraftGround.

  • Open Access

    From Sparse to Dense: Toddler-Inspired Reward Transition in Goal-Oriented Reinforcement Learning

    by Junseok Park, Hyeonseo Yang, Min Whoo Lee, Won-Seok Choi, Minsu Lee, Byoung-Tak Zhang

    Reinforcement learning (RL) agents face fundamental challenges in balancing exploration and exploitation, particularly when sparse or dense rewards bias learning toward sub–optimal behaviors. Biological systems, notably human toddlers, naturally navigate this balance through developmental transitions from free exploration to goal-directed behavior. Inspired by this progression, we propose the Sparse-to-Dense (S2D) reward transition for goal-oriented RL tasks. Our approach transitions from sparse rewards encouraging broad exploration to dense rewards guiding efficient exploitation, while preserving optimal policies via potential-based reward shaping throughout. Through extensive experiments on robotic manipulation and egocentric 3D navigation tasks, we demonstrate that S2D transitions significantly enhance sample efficiency, and generalization compared to static reward schemes and intrinsic motivation baselines. Using a novel Cross-Density Visualizer, we reveal that S2D transitions smooth the policy loss landscape, resulting in wider minima associated with improved generalization. Furthermore, reinterpreting Tolman’s maze experiments in modern RL contexts, we show that early sparse-reward exploration establishes robust initial parameters analogous to cognitive maps, facilitating stable learning under subsequent dense rewards. Our findings bridge developmental psychology and machine learning, offering a principled framework for designing adaptive reward structures in complex RL environments.