Jiwon Seok

Passenger satisfaction in an autonomous vehicle depends on individual differences in the vehicle motion a passenger is willing to accept. The personalization literature commonly assumes that satisfaction is maximized when the vehicle reproduces the driving style of its passenger. In this work we propose a personalized Model Predictive Control (MPC) scheme for autonomous trajectory tracking and use it to test that assumption. We first conducted a literature review to derive the key personality parameters that characterize driving behavior, then acquired naturalistic driving data from seventeen drivers on a fixed urban and highway combined course and modeled the distribution of their driving styles statistically, condensing the parameters into a scalar Driving Style Metric (DSM) by principal component analysis. A real-time personality updater identifies the driving style online and maps it to the constraint set of the controller. The proposed MPC controller and personality estimator were verified in both simulation and real-world vehicle experiments, and the resulting driving styles were assessed in a ride-along user study. The subjective ratings do not support the style-matching hypothesis: no driver group rated the Dynamic style highest, not even the participant whose own driving defined its limits. What the habitual driving style predicts is not the preferred motion but the tolerated one, as tolerance for the Dynamic style increased monotonically with habitual aggressiveness (2.83, 3.07 and 4.44 on a five-point scale for Comfort-, Normal- and Dynamic-style drivers). We therefore reinterpret the extracted personality parameters as an individual acceptability bound, that is, the upper limit of the personalization search space rather than its optimum. The personalized style, which places the constraint ceiling at that individual bound, obtained the highest overall satisfaction (4.53, against 4.46 for Comfort and 3.45 for Dynamic). The user study covers six participants and all reported values are descriptive.

Learning-based motion planners infer driving behavior from data, reducing reliance on hand-crafted rules. However, closed-loop deployment remains difficult because their outputs can fluctuate across planning cycles, violate vehicle dynamics and friction limits, and degrade ride comfort through unnecessary acceleration and steering oscillations. Prior safety filters and low-level control corrections can mitigate immediate violations, but they do not provide a unified mechanism that enforces trajectory-level consistency, executability, and risk-aware validation at the planner output. This paper proposes a refinement-and-verification trajectory processing pipeline that improves closed-loop execution without replacing the underlying planner. The pipeline combines a Kalman filter–based consistency refinement across planning cycles, an NMPC-based Dynamic–Safety–Comfort refinement that enforces vehicle dynamics together with safety and comfort constraints, and a risk-aware verification and selection stage based on friction limits, collision risk, and driving-quality metrics. On the NuPlan closed-loop benchmark, the full pipeline improves ride-comfort metrics by more than 10% and increases collision- and TTC-related safety metrics by about 4%, yielding a consistent gain in the overall planning score. Additional evaluation in CarMaker and real-vehicle experiments shows that the refined trajectories remain dynamically feasible and can be stably tracked without sacrificing task-level driving performance. These results support the role of a structured trajectory-level intermediate layer for bridging learning-based planning and real-world execution. Supplementary real-vehicle comfort demonstration videos are available at https://github.com/ailab-hanyang/ML-Planning-Refinement-Verification.

An electronically controlled suspension with road preview enhances ride comfort and driving stability. For detecting road irregularities such as speed bump, look-ahead sensors such as camera and Light Detection And Ranging (LiDAR) sensors have been used. However, these preview systems perform poorly in night and adverse weather conditions. To overcome these limitations, this paper proposes a novel machine learning-based speed bump prediction system that predicts remaining distance to and height of speed bumps using in-vehicle data, which is robust to external environment and represent driving pattern of driver. The proposed system consists of three components: machine learningbased bump predictor, automatic training dataset generator, and cloud-based continuous learning system. The effectiveness of the proposed system was confirmed through tests with varied datasets in different drivers and locations. The prediction of the distance to the speed bump demonstrated a Root Mean Square Error (RMSE) of 6.37 meters, while the height prediction showed an RMSE of 0.9 cm. Additionally, after applying cloud-based continuous learning for driver-specific personalization, the RMSE for the distance prediction improved to 5.96 meters. Furthermore, experiments conducted under challenging conditions showed that our system outperformed existing systems.

This paper proposes a Light Detection And Ranging (LiDAR) based system for estimating road surface height profiles and detecting bumps and dips for preview suspension systems. The proposed system includes a method for ground extraction to remove obstacles and applies Adaptive Smoothing method to generate a continuous and accurate road height profile. Additionally, a bump-dip detection technique is introduced to identify road irregularities directly from the generated profile. We conducted a quantitative analysis of the system's performance through simulations and proving ground experiments, considering various LiDAR installation configurations and bump-dip environments. The results verify the accuracy and robustness of the proposed algorithm, and real-world vehicle experiments further confirm the reliability of the road height profile estimation.

Accurate localization is crucial for autonomous vehicle navigation. In particular, there is active research in odometry that estimates a vehicle's transformation over time. To achieve this, various environmental perception sensors are utilized, among which radar sensors stand out for their costeffectiveness and robustness against adverse weather conditions compared to LiDAR sensors. However, traditional radar sensors provide only 2D spatial information (x, y,doppler). Recent technical advancements have introduced 4D imaging radar, providing 3D spatial (x, y,z,doppler) information. Nonetheless, radar data remains sparse and noisy when compared to LiDAR data, making it challenging to apply conventional LiDAR-based odometry algorithms. To address this challenge, we propose a radar point cloud odometry system that leverages Doppler, representing relative velocity, and Radar cross-section (RCS), a unique measure of an object's reflective ability. The IMU-free ego-motion estimation step utilizes Doppler data to generate an initial guess for registration. Additionally, to effectively extract meaningful points, we propose a polar-grid-based feature extraction algorithm utilizing RCS. To overcome the sparsity of radar point cloud data, we perform RCS-weighted accumulated scans-to-submap matching, where weights for point cloud registration are modeled based on RCS values to achieve robust odometry. Our proposed algorithm was experimentally evaluated using the View-of-Delft dataset. The results demonstrate that our algorithm outperforms existing methods, providing superior odometry performance.

The importance of precise localization technology for the autonomous driving of industrial mobile robots is steadily increasing. Notably, research into enhancing accuracy and robustness by fusing multiple systems is actively conducted rather than relying on a single localization system. We highlight the use of track-to-track (T2T) fusion, which takes the localization results of independent systems as input. This approach eliminates system adjustments with sensor changes, offering benefits for industrial mobile robots. However, existing T2T-based fusion methods suffer from overlooking slowly changing biases that can gradually increase over time due to sensor drift errors, map biases, etc. Since biases have different values and frequencies for each system, they are challenging for conventional T2T methods to handle. This article proposes a localization fusion framework that tackles such slowly varying biases. First, estimating the distinct biases inherent to each system poses a challenging problem; therefore, we align them to a single common bias. Second, localization estimates with a common bias are fused using a split covariance intersection filter, one of the T2T fusion techniques, considering the independence and correlation within each system to ensure fusion consistency. The proposed method has been validated in both simulation and real-world environments, confirming superior performance compared to existing algorithms.