Three papers have been accepted to IEEE RA-L 2026
IEEE Robotics and Automation Letters (RA-L), launched by the IEEE Robotics & Automation Society in 2015, is a leading peer-reviewed archival journal that publishes timely and concise reports of innovative research in robotics and automation. According to the 2025 release of Journal Citation Reports, based on 2024 data, RA-L has a Journal Impact Factor of 5.3 and a 5-year Impact Factor of 6.0, and is ranked in Q1 in the Robotics category.
DRIP: Self-Improving Robot State Estimation In-the-Wild via Dense Robot-Fitting in the LooP
Authors: Kim Ohn (KAIST), Kim Youwang (POSTECH), Junuk Cha (KAIST), Tae-Hyun Oh (KAIST)
Project Page | Paper
Accurately estimating robot state, i.e., pose and joint angles, from a single image remains challenging for in-the-wild scenarios, where 3D state labels, calibrated sensor rigs, and joint encoders are often unavailable in practice. Prior learning-based methods rely on synthetic supervision and sparse 2D cues from real images, e.g., keypoints and masks, which provide limited cues for accurate monocular robot state estimation. To mitigate this, we present DRIP, a novel robot-fitting in the loop training framework that learns monocular robot state estimation from real images without 3D state labels. DRIP seamlessly integrates a feed-forward network with iterative robot state refinement in the training pipeline: the network provides an initial robot state prediction, which is then refined through an optimization stage that leverages dense visual cues, e.g., structure, geometry, and silhouette. The refined states are then used as pseudo-labels for subsequent training, forming a self-improving loop that transfers the benefits of refinement to the feed-forward model while preserving efficient single-shot inference at test time. Across multiple real-world robot datasets, DRIP outperforms prior approaches that rely on sparse self-supervision. We show that robot-fitting in the loop training enables effective real-image adaptation for monocular robot state estimation by generating reliable state-level pseudo-labels for stronger supervision.
LiDAR-Anchored Collaborative Distillation for Robust 2D Representations
Authors: Wonjun Jo (POSTECH), Hyunwoo Ha (POSTECH), Kim Ji-Yeon (POSTECH), Hawook Jeong (RideFlux), Tae-Hyun Oh (KAIST)
As deep learning continues to advance, self-supervised learning has made considerable strides. It allows 2D image encoders to extract useful features for various downstream tasks, including those related to autonomous robotic perception systems. Nevertheless, pre-trained 2D image encoders fall short in conducting the task under noisy and adverse weather conditions beyond clear daytime scenes, which require for robust visual perception. To address these issues, we propose a novel self-supervised approach, Collaborative Distillation, which leverages 3D LiDAR as self-supervision to improve robustness to noisy and adverse weather conditions in 2D image encoders while retaining their original capabilities. Our method outperforms competing methods in various downstream tasks across diverse conditions and exhibits strong generalization ability. In addition, our method also improves 3D awareness stemming from LiDARโs characteristics. This advancement highlights our methodโs practicality and adaptability in real world scenarios.
DarkQA: Benchmarking Vision-Language Models on Visual-Primitive Question Answering in Low-Light Indoor Scenes
Authors: Yohan Park (KAIST), Hyunwoo Ha (POSTECH), Wonjun Jo (POSTECH), Tae-Hyun Oh (KAIST)
Vision-Language Models (VLMs) are increasingly used as central reasoning modules for embodied agents, yet existing benchmarks largely evaluate them under well-lit conditions. Robust 24/7 operation, however, requires reliable perception under visual degradations such as low light at night or in dark environments. We present DarkQA, an open-source benchmark for evaluating visual-primitive question answering under controlled multi-level low-light conditions in embodied scenarios. By focusing on visual primitives from egocentric observations, DarkQA isolates low-light perceptual failures before they are confounded by downstream embodied-task complexity. It contains 9.4K deterministically generated and verifiable question--image pairs spanning five visual-primitive families. To ensure physical fidelity, DarkQA models low-light image formation through inverse and forward ISP-based camera processing and RAW-domain sensor-noise injection, and validates the synthesis against real paired low-light camera data. We evaluate representative VLMs with and without Low-Light Image Enhancement (LLIE) preprocessing. Results show consistent VLM degradation under low illumination and sensor noise, while LLIE yields severity-dependent but unstable recovery.