Wenjie Jacky Mo, Qin Liu, Xiaofei Wen, Dongwon Jung, Hadi Askari, Wenxuan Zhou, Zhe Zhao, Muhao Chen. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Training large language models (LLMs) with chain-of-thought (CoT) supervision has proven effective for enhancing their reasoning abilities.However, obtaining reliable and accurate reasoning supervision remains a significant challenge.We propose a scalable method for generating a high-quality CoT supervision dataset by leveraging the determinism of program execution.Unlike existing reasoning dataset generation methods that rely on costly human annotations or error-prone LLM-generated CoT, our approach extracts verifiable, step-by-step reasoning traces from code execution and transforms them into a natural language CoT reasoning.Experiments on reasoning benchmarks across various domains show that our method effectively equips LLMs with transferable reasoning abilities across diverse tasks.Furthermore, the ablation studies validate that our method produces highly accurate reasoning data and reduces overall token length during inference by reducing meaningless repetition and overthinking.1

This paper presents an algorithm for vision-based geolocation of a ground target using on-board gimbal camera of UAV. Assume the initial location of the target is inaccurately given, initial pan/tilt angles of the gimbal camera enable the target to be captured inside the image frame. An object detection and tracking algorithm are incorporated to calculate the offset of the bounding box from the image center, which corrects the pan/tilt angles of the gimbal in order for the target to be exactly centered in the image frame. Once the target is locked on the image frame, the exact position of the target is calculated using the inertial information of the UAV as well as the measured pan/tilt angles of the gimbal camera. As the measurements for geolocationing are vulnerable to disturbances and are very noisy, the extended Kalman filter (EKF) and the unscented Kalman filter (UKF) are adopted to smooth the noisy measurements while providing reliable geolocationing of the target. Based on the extensive verification results over the simulation environment, the proposed algorithm has been implemented on an actual hardware mounted on a UAV test-bed to validate the performance via flight tests.