Minimalist Visual Inertial Odometry

Researchers proposed a minimalist visual-inertial odometry system for differential-drive robots using four downward-facing photodiodes, learned optical Gabor masks, and an IMU.

A temporal convolutional network estimates speed and heading for planar motion, avoiding full camera-based vision.

The system was tested on a robot in both indoor and outdoor settings without real-world fine-tuning.

The results suggest accurate navigation is possible with very low-cost, low-data sensing and lower compute demands.