Two drivers exhibit distinct preferences. Each driver avoids the unpreferred region (red shaded area in the front camera view) based on their preferences. To achieve personalized driving behavior, we optimize over feasible trajectories that avoid the resulting unpreferred region.
Abstract
Driving without considering a user’s preferred separation distance from surrounding vehicles can cause discomfort and reduce trust in autonomous driving systems. We propose a preference-aware planning framework that explicitly incorporates user-preferred clearance margins into an optimal-control planner. A purpose-built questionnaire captures interaction-relevant preferences over vehicle size, relative speed, lane, relative position, and maneuvers of both the ego and surrounding vehicles. The responses define preference-space constraints for trajectory optimization. Because directly encoding all possible scenarios in a single optimal control problem is computationally intractable for real-time driving, we decompose the original problem into scenario-specific subproblems, solve them in parallel, and select a trajectory using the original objective. Simulations in HighwayEnv show that the proposed planner better aligns with user-preferred clearances than preference-agnostic baselines, while also producing distinct behaviors for conservative and aggressive driver profiles.
Method overview
The proposed framework to achieve personalized driving. Further details are provided in our paper.
Our framework consists of three stages, illustrated by the magenta, cyan, and orange shaded regions, respectively.
The first stage (questionnaire stage) collects the user-preferred clearance margin through a questionnaire. The questionnaire is available here.
The second stage (parallel trajectory optimization stage) takes the questionnaire responses and the current driving scenario as inputs. The resulting preferences are incorporated as clearance constraints in the optimal control problem, which is then decomposed into scenario-specific subproblems.
Finally, in the third stage (selection stage), these subproblems are solved in parallel, and one is selected considering the global objective.
Real-vehicle experiments
We compare two representative driver profiles: an aggressive driver and a conservative driver. The aggressive driver accepts smaller clearance margins and therefore plans more assertive trajectories that pass closer to surrounding vehicles when this improves progress. The conservative driver requires larger clearance margins, leading to more cautious trajectories that keep greater separation from neighboring vehicles. As a result, the same driving scenario produces different planning outcomes.
Planning results for two representative driver profiles.
As shown in the figure above, the aggressive driver plans a more assertive trajectory that passes closer to surrounding vehicles, whereas the conservative driver plans a trajectory with larger clearance margins. These differences arise from the questionnaire-derived preference constraints and demonstrate personalized planning behavior under the same driving scenario. In addition, we conduct vehicle experiments to demonstrate the effectiveness of the proposed method.
Additional comparisons and quantitative evaluations are provided in the paper.
BibTeX citation
@inproceedings{lim2026personalized, title = {Personalized Autonomous Driving via Optimal Control with Clearance Constraints from Questionnaires}, author = {Lim, Yongjae and Kim, Dabin and Kim, H. Jin}, booktitle = {IEEE International Conference on Robotics and Automation (ICRA)}, year = {2026}}