Yusung Kim

Kyungsoo Kim, Jeongsoo Ha, Yusung Kim

Vision-based reinforcement learning requires efficient and robust representations of image-based observations, especially when the images contain distracting (task-irrelevant) elements such as shadows, clouds, and light. It becomes more important if those distractions are not exposed during training. We design a Self-Predictive Dynamics (SPD) method to extract task-relevant features efficiently, even in unseen observations after training. SPD uses weak and strong augmentations in parallel, and learns representations by predicting inverse and forward transitions across the two-way augmented versions. In a set of MuJoCo visual control tasks and an autonomous driving task (CARLA), SPD outperforms previous studies in complex observations, and significantly improves the generalization performance for unseen observations. Our code is available at https://github.com/unigary/SPD.

Jeongsoo Ha, Kyungsoo Kim, Yusung Kim

Model-based reinforcement learning (MBRL) has been used to efficiently solve vision-based control tasks in highdimensional image observations. Although recent MBRL algorithms perform well in trained observations, they fail when faced with visual distractions in observations. These task-irrelevant distractions (e.g., clouds, shadows, and light) may be constantly present in real-world scenarios. In this study, we propose a novel self-supervised method, Dream to Generalize (Dr. G), for zero-shot MBRL. Dr. G trains its encoder and world model with dual contrastive learning which efficiently captures task-relevant features among multi-view data augmentations. We also introduce a recurrent state inverse dynamics model that helps the world model to better understand the temporal structure. The proposed methods can enhance the robustness of the world model against visual distractions. To evaluate the generalization performance, we first train Dr. G on simple backgrounds and then test it on complex natural video backgrounds in the DeepMind Control suite, and the randomizing environments in Robosuite. Dr. G yields a performance improvement of 117% and 14% over prior works, respectively. Our code is open-sourced and available at https://github.com/JeongsooHa/DrG.git

Seungho Baek, Taegeon Park, Jongchan Park et al.

Existing offline hierarchical reinforcement learning methods rely on high-level policy learning to generate subgoal sequences. However, their efficiency degrades as task horizons increase, and they lack effective strategies for stitching useful state transitions across different trajectories. We propose Graph-Assisted Stitching (GAS), a novel framework that formulates subgoal selection as a graph search problem rather than learning an explicit high-level policy. By embedding states into a Temporal Distance Representation (TDR) space, GAS clusters semantically similar states from different trajectories into unified graph nodes, enabling efficient transition stitching. A shortest-path algorithm is then applied to select subgoal sequences within the graph, while a low-level policy learns to reach the subgoals. To improve graph quality, we introduce the Temporal Efficiency (TE) metric, which filters out noisy or inefficient transition states, significantly enhancing task performance. GAS outperforms prior offline HRL methods across locomotion, navigation, and manipulation tasks. Notably, in the most stitching-critical task, it achieves a score of 88.3, dramatically surpassing the previous state-of-the-art score of 1.0. Our source code is available at: https://github.com/qortmdgh4141/GAS.

Seunghwan Shin, Yusung Kim

In the field of Multi-Person Pose Estimation (MPPE), Radio Frequency (RF)-based methods can operate effectively regardless of lighting conditions and obscured line-of-sight situations. Existing RF-based MPPE methods typically involve either 1) converting RF signals into heatmap images through complex preprocessing, or 2) applying a deep embedding network directly to raw RF signals. The first approach, while delivering decent performance, is computationally intensive and time-consuming. The second method, though simpler in preprocessing, results in lower MPPE accuracy and generalization performance. This paper proposes an efficient and lightweight one-stage MPPE model based on raw RF signals. By sub-grouping RF signals and embedding them using a shared single-layer CNN followed by multi-head attention, this model outperforms previous methods that embed all signals at once through a large and deep CNN. Additionally, we propose a new self-supervised learning (SSL) method that takes inputs from both one unmasked subgroup and the remaining masked subgroups to predict the latent representations of the masked data. Empirical results demonstrate that our model improves MPPE accuracy by up to 15 in PCKh@0.5 compared to previous methods using raw RF signals. Especially, the proposed SSL method has shown to significantly enhance performance improvements when placed in new locations or in front of obstacles at RF antennas, contributing to greater performance gains as the number of people increases. Our code and dataset is open at Github. https://github.com/sshnan7/SOSPE .