Multiple motion estimators such as an interacting multiple model (IMM) have been utilized to track target objects such as cars and pedestrians with diverse motion patterns. However, the standard IMM has limitations in combining motion models with different state definitions, so it cannot contain a complementary set of models that accurately work for all motion patterns. In this paper, we propose IMM-based adaptive target tracking with heterogeneous velocity representations and linear/curvilinear motion models. It can integrate four motion models with different state definitions and dimensions to be completely complimentary for all types of motions. We experimentally demonstrate the effectiveness of the proposed method with accuracy for various motion patterns using two types of datasets: synthetic datasets and real datasets. Experimental results show that the proposed method achieves the adaptive target tracking for diverse types of motion and also for various objects such as cars, pedestrians, and drones in the real world.

A GPS is usually installed at the center of a vehicle because with this configuration its data become the position of the vehicle without complex calibration and transformation. In this paper, however, we verify that the off-center arrangement of the GPS improves the position and orientation accuracy theoretically and experimentally. The extended Kalman filter (EKF) formulation and its observability and uncertainty analysis present that the off-centered GPS is better observable and less uncertain. Moreover, our experiments with synthetic data and real sensor dataset confirm the improvement of localization accuracy. The synthetic data generated in various situations reveal the important properties of localization with the off-centered GPS. The public outdoor dataset with real GPS data also supports the advantage of the off-centered GPS.

This paper presents a novel local planning approach called Agile and SAfe Pursuit (ASAP) for autonomous mobile robots. It aims to enable agile path following and safe collision avoidance in cluttered environments, while ensuring computational efficiency for real-time performance in embedded systems with limited computational power. For agile path following, the proposed approach utilizes a local path that includes a line path, arc path, and in-place rotation, and generates a target velocity based on the kinematic constraints of the robot. For safe collision avoidance, the proposed approach uses obstacle information to generate safety corners, which represent points in free space to circumvent obstacles with arbitrary shapes, and selects the best safety corner with the minimum travel time. To reach the target velocity as quickly as possible, the proposed approach uses a normalized velocity space to calculate control velocity that achieves the ratio of the linear and angular components of the target velocity in the shortest possible time period. For end-users to easily adapt a robot’s behavior to different environments, the proposed approach is designed to require only a few tuning parameters. The proposed algorithm’s agile control, rigorous collision avoidance, and computational efficiency were demonstrated through experimental results from hardware-in-the-loop simulations under various scenarios and real-robot tests in cluttered environments. Remarkably, the proposed approach achieves computational speeds that are 25 to 200 times faster than other existing algorithms.

Camera calibration is essential for accurate optical measurement and computer vision applications. The precision of calibration parameters significantly impacts the performance of subsequent computer vision tasks. However, the process is often complicated by the need to select the most suitable camera projection model, determine the optimal number of images, choose appropriate viewpoints, and define criteria for high-quality dataset construction. To address these challenges, we introduce the Generalized Camera Calibration framework, a novel approach that automates dataset creation, considers various camera projection models, and identifies the optimal model and its parameters based on comprehensive model selection criteria. This framework streamlines the calibration process, eliminating the need for manual image and camera model selection before camera calibration. Our method demonstrates outstanding performance on both synthetic and real data. On synthetic datasets, it achieves a remarkable 93.18% accuracy in identifying the correct ground truth model using 40 images, employing BIC for model selection. When applied to real datasets, our method maintains a consistent RMSE of approximately 0.3 pixels across both training and test sets. Extensive validation on synthetic and real data underscores the significant performance enhancements achieved through our approach, making it a powerful tool for simplifying and improving camera calibration in various applications.

GPS is the most popular sensor for outdoor localization. GPS-only localization is the simplest and initial setting, thus it has been employed in many applications. This paper presents more accurate GPS-only localization with two velocity constraints based on Bayesian filtering. GPS-only localization inherently suffers from ambiguity problems in its state variables due to the limitation of position-only observations. These ambiguities lead to incorrect or diverged state estimates, which are commonly observed in cases of violating assumptions in motion and observation models. Since two proposed velocity constraints can resolve the ambiguity problems, EKF localization with two additional constraints can achieve more accurate localization and demonstrate better recovery from broken state estimates. We quantitatively validated such improvements in localization accuracy and recovery using synthetic data with various GPS trajectories and configurations. Experimentally, the constant velocity model with two velocity constraints exhibited around 25% less position error and 70% less orientation error on average compared to the original constant velocity model. We also qualitatively observed similar results with two real-world datasets. In our experiments with real-world datasets, two velocity constraints successfully resolved state ambiguities after abrupt motion and severely incorrect GPS measurements. Our basic implementation is available at https://github.com/mint-lab/filtering_tutorial.