Unsupervised surface classification to enhance the control performance of a UGV

Unmanned Ground Vehicles (UGVs) currently serve a wide array of purposes — from mapping remote locations to planetary exploration — where surface conditions are rarely well defined ahead of time. The unknown nature of these surfaces is problematic since optimal performance of a ground vehicle depends greatly on the properties of the surface the vehicle finds itself on. In this work, an unsupervised learning algorithm that classifies surfaces and adapts driving control accordingly is presented. This project uses a modified density-based spatial clustering of applications with noise (DBSCAN) algorithm with data from a 2-D LiDAR sensor and RGB-D Camera to classify different surfaces. A control algorithm then uses acceleration data from an IMU to control a ground vehicle's traversal across different surfaces by limiting the maximum forward velocity of the vehicle based on the vibrations it experiences. The maximum velocity for each classified surface is stored, allowing the system to quickly revert to optimal operation when returning to surfaces it has previously identified.