# Comfort-oriented driving: performance comparison between human drivers and motion planners

arXiv: https://arxiv.org/abs/2301.10538 · HTML: https://mlsanity.com/paper/2301.10538/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Imitation-learning end-to-end planners, one of 9 directions in End-to-end autonomous driving.
- Builds on 4 of its 45 references that are in the map.
- First posted on arXiv 25 Jan 2023.

## Placed in

- End-to-end autonomous driving / [Imitation-learning end-to-end planners](https://mlsanity.com/driving/imitation-learning-end-to-end-planners/)

## Builds on (most cited first, 4 of 4)

- [openDD: A Large-Scale Roundabout Drone Dataset](https://mlsanity.com/paper/2007.08463/) (2020)
- [Coordinating Vehicle Platoons for Highway Bottleneck Decongestion and Throughput Improvement](https://mlsanity.com/paper/1907.13049/) (2019)
- [Vehicle Platooning Impact on Drag Coefficients and Energy/Fuel Saving Implications](https://mlsanity.com/paper/2001.00560/) (2020)
- [Mitigating Motion Sickness with Optimization-based Motion Planning](https://mlsanity.com/paper/2301.07977/) (2023)

## Abstract (arXiv)

Motion planning is a fundamental component in automated vehicles. It influences the comfort and time efficiency of the ride. Despite a vast collection of studies working towards improving motion comfort in self-driving cars, little attention has been paid to the performance of human drivers as a baseline. In this paper, we present an experimental study conducted on a public road using an instrumented vehicle to investigate how human drivers balance comfort and time efficiency. The human driving data is compared with two optimization-based motion planners that we developed in the past. In situations when there is no difference in travel times, human drivers incurred an average of 23.5% more energy in the longitudinal and lateral acceleration signals than the motion planner that minimizes accelerations. In terms of frequency-weighted acceleration energy, an indicator correlated with the incidence of motion sickness, the average performance deficiency rises to 70.2%. Frequency-domain analysis reveals that human drivers exhibit more longitudinal oscillations in the frequency range of 0.2-1 Hz and more lateral oscillations in the frequency range of up to 0.2 Hz. This is reflected in time-domain data features such as less smooth speed profiles and higher velocities for long turns. The performance difference also partly results from several practical matters and additional factors considered by human drivers when planning and controlling vehicle motion. The driving data collected in this study provides a performance baseline for motion planning algorithms to compare with and can be further exploited to deepen the understanding of human drivers.