Master thesis: Reckless Riding Detection for Safer Micromobility
How can we increase safety by detecting reckless riding?
-
In my Master Thesis Project you will get the chance to work with real-world micromobility data from Voi’s operational fleet and contribute to next-generation rider safety systems with real deployment potential.
Marco Capuccini
Senior Machine Learning Engineer
The thesis
E-scooter safety has become a key challenge as shared micromobility expands across cities. While most riders behave responsibly, a small share engage in reckless behaviors—such as swerving, tandem riding, or single-handed riding—that significantly increase crash risk, as shown by Pai and Dozza (2025). Building on this foundation, Capuccini, Pai, and Carlsson (2025) introduced a testing-by-betting approach for detecting anomalies in ride data, providing a statistically grounded and computationally efficient method for onboard event detection. However, current detection methods still struggle to capture subtle, context-dependent riding patterns. This thesis aims to address these limitations by integrating richer sensing and learning-based models for more accurate and robust detection of reckless riding.
YOUR MISSION
The goal of this thesis is to advance reckless riding detection by leveraging rich sensor data and modern hypothesis testing and learning techniques. Possible directions include:
- Integrating sensor data, such as accelerometer and gyroscope, to improve detection accuracy and robustness.
- Exploring machine learning techniques such as contrastive learning–based sensor representations, LLM-based zero/few-shot detection, and statistical methods like testing-by-betting for theory-grounded guarantees.
- Utilizing forward-facing camera data available on a subset of Voi’s fleet to complement sensor-based detection, using privacy-preserving methods.
- Evaluating model performance under realistic deployment conditions, focusing on accuracy, computational efficiency, and robustness.
The exact focus will be shaped according to the student’s interests and background.
STUDENT PROFILE
We are looking for students that have:
- Background in statistics, signal processing, or machine learning.
- Programming skills in Python; familiarity with cloud computing platforms is beneficial.
- Interest in applying data-driven methods to urban mobility and safety.
- The project scope will be adapted to the student’s skills, experience, and interest.