Jaeseung Byun

LG AI Research, Sehyun Chun, Jiye Kim et al.

The extraction of molecular structures and reaction data from scientific documents is challenging due to their varied, unstructured chemical formats and complex document layouts. To address this, we introduce MolMole, a vision-based deep learning framework that unifies molecule detection, reaction diagram parsing, and optical chemical structure recognition (OCSR) into a single pipeline for automating the extraction of chemical data directly from page-level documents. Recognizing the lack of a standard page-level benchmark and evaluation metric, we also present a testset of 550 pages annotated with molecule bounding boxes, reaction labels, and MOLfiles, along with a novel evaluation metric. Experimental results demonstrate that MolMole outperforms existing toolkits on both our benchmark and public datasets. The benchmark testset will be publicly available, and the MolMole toolkit will be accessible soon through an interactive demo on the LG AI Research website. For commercial inquiries, please contact us at \href{mailto:contact_ddu@lgresearch.ai}{contact\_ddu@lgresearch.ai}.

Jaeseung Byun, Simo A. Mäkiharju, Mark W. Mueller

This paper presents a round-trip strategy of multirotors subject to unknown flow disturbances. During the outbound flight, the vehicle immediately utilizes the wind disturbance estimations in feedback control, as an attempt to reduce the tracking error. During this phase, the disturbance estimations with respect to the position are also recorded for future use. For the return flight, the disturbances previously collected are then routed through a feedforward controller. The major assumption here is that the disturbances may vary over space, but not over time during the same mission. We demonstrate the effectiveness of this feedforward strategy via experiments with two different types of wind flows; a simple jet flow and a more complex flow. To use as a baseline case, a cascaded PD controller with an additional feedback loop for disturbance estimation was employed for outbound flights. To display our contributions regarding the additional feedforward approach, an additional feedforward correction term obtained via prerecorded data was integrated for the return flight. Compared to the baseline controller, the feedforward controller was observed to produce 43% less RMSE position error at a vehicle ground velocity of 1 m/s with 6 m/s of environmental wind velocity. This feedforward approach also produced 14% less RMSE position error for the complex flows as well.

Jaeseung Byun, Karan P. Jain, Siddharth H. Nair et al.

We propose a high-level planner for a multirotor to chase a ground vehicle, while simultaneously respecting various state and input constraints. Assuming a minimal kinematic model for the ground vehicle, we use data collected online to generate predictions for our planner within a model predictive control framework. Our solution is demonstrated, both via simulations and experiments on a stable quadcopter platform.