Robot Quadruped Material Database

We present the RQMD dataset, a comprehensive collection of diverse material samples aimed at advancing computer vision and machine learning algorithms in terrain classification tasks. This dataset contains RGB images of 5 different terrains, such as Asphalt, Brick, Grass, Gravel, and Tiles, captured using an 8-megapixel Raspberry Pi camera from a top-view perspective. Notably, the dataset encompasses images taken at different times of the day, introducing variations in lighting conditions and environmental factors.

To facilitate friction force extraction, we conducted an experiment using a rubber ball, simulating the foot of a small-scale robot. The experiment involved scratching the rubber ball on the different terrains using a DC motor with an encoder, allowing us to measure the friction force acting on the ball. This information contributes to a more comprehensive understanding of the terrain properties.

In summary, RQMD serves as a valuable resource for researchers and practitioners seeking to enhance their computer vision and machine learning models for terrain classification tasks, and the friction force extraction experiment adds an additional dimension to the dataset, enabling deeper insights into terrain characteristics.