[Paper Review] Improving Driver Satisfaction with a Driving Function Learning from Implicit Human Feedback -- a Test Group Study

Robin Schwager, Andrea Anastasio+7 more|arXiv (Cornell University)|Feb 14, 2026

Human-Automation Interaction and Safety0 citations

TL;DR

The paper presents an iterative Speed Profile Adjustment Algorithm (SPAA) that personalizes a Predictive Longitudinal Driving Function (PLDF) using drivers’ implicit interventions, showing increased satisfaction and reduced interventions in a driving-simulator study.

ABSTRACT

During the use of advanced driver assistance systems, drivers frequently intervene into the active driving function and adjust the system's behavior to their personal wishes. These active driver-initiated takeovers contain feedback about deviations in the driving function's behavior from the drivers' personal preferences. This feedback should be utilized to optimize and personalize the driving function's behavior. In this work, the adjustment of the speed profile of a Predictive Longitudinal Driving Function (PLDF) on a pre-defined route is highlighted. An algorithm is introduced which iteratively adjusts the PLDF's speed profile by taking into account both the original speed profile of the PLDF and the driver demonstration. This approach allows for personalization in a traded control scenario during active use of the PLDF. The applicability of the proposed algorithm is tested in a driving simulator-based test group study with 43 participants. The study finds a significant increase in driver satisfaction and a significant reduction in the intervention frequency when using the proposed adaptive PLDF. Additionally, feedback by the participants was gathered to identify further optimization potentials of the proposed system.

Motivation & Objective

• Motivate personalization of driving functions by leveraging driver-initiated interventions within the PLDF's ODD.
• Propose an iterative SPAA that blends the base PLDF speed profile with driver demonstrations to learn preferences.
• Evaluate whether SPAA increases driver satisfaction and reduces intervention frequency in a simulator study.
• Explore limitations and suggest avenues for real-vehicle deployment and long-term evaluation.

Proposed method

• Describe the Predictive Longitudinal Driving Function (PLDF) and how it respects speed limits and road curvature.
• Define driver interventions as pedal or set speed adjustments and explain their recording and use within SPAA.
• Detail the Speed Profile Adjustment Algorithm (SPAA), including: stretching intervention profiles with factor alpha=0.5, aligning with driver profiles, applying a Savitzky-Golay smoothing, and forming v_mean(d) = (v_PLDF(d) + v_prepro(d))/2.
• Explain iterative training where the adjusted speed profile becomes the new PLDF baseline for subsequent drives.
• Differentiate pedal interventions (merged with PLDF profile) and set-speed interventions (taken directly) in SPAA.
• Describe the test-group simulator setup, track, participants, and procedure for comparing baseline PLDF (System A) with adaptive PLDF using SPAA (System B).

Experimental results

Research questions

• RQ1 Does the adaptive PLDF with SPAA increase driver satisfaction compared to the baseline PLDF?
• RQ2 Does SPAA reduce driver intervention frequency during driving with the PLDF?
• RQ3 How do drivers perceive the speed-profile, and which aspects of the speed profile are most improved?
• RQ4 What are qualitative participant observations about communication and learnability of the adaptive system?
• RQ5 What are limitations and future directions for real-vehicle deployment and arbitration of interventions?

Key findings

• System B yields higher general satisfaction (mean 4.27) than System A (mean 3.69) with p = 3.85e-5 (paired t-test).
• Speed-profile satisfaction is higher for System B (mean 4.14) than System A (mean 3.06) with p = 3.83e-9 (Wilcoxon).
• Combined intervention rate drops from 54.68% (System A) to 22.97% (System B) with p = 4.55e-13 (Wilcoxon).
• Pedal intervention rate decreases from 22.32% (System A) to 12.04% (System B) with p = 9.78e-8 (t-test).
• Set-speed intervention rate decreases from 39.76% (System A) to 12.42% (System B) with p = 1.38e-7 (Wilcoxon).
• One iteration of SPAA already significantly reduces intervention rates, though complete convergence to 0% was not observed within the study duration.