Anomaly Detection for Safety Navigation
This project explores vision-based safety systems for outdoor navigation, with a focus on detecting and distinguishing between hazardous and non-hazardous anomalies in sidewalk environments. By combining unsupervised learning through a Variational Autoencoder (VAE) with a machine learning classifier, the system learns what normal environments look like and flags anything unusual. The goal is to develop intelligent navigation aids that can reliably identify potential obstacles and support safe movement for users in complex, real-world settings.
Researchers: Edgar Guzman
Sponsors:
Terrain Transition for Assistive Navigation
By combining RGB-D camera data with IMU signals, we detect upcoming staircases, estimate when users will transition onto them, and accurately measure step dimensions. This information enables adaptive control of lower-limb assistive devices, such as exoskeletons or prosthetics, helping users navigate complex terrain more safely and intuitively. Our system bridges the gap between environmental awareness and human motion prediction, offering a foundation for smarter, more responsive wearable technologies.
Researchers: Edgar Guzman
