Low-cost Mapping of surface terrain features in two dimensions through Machine Learning and Sensor fusion techniques

A novel low-cost, surface relief mapping capable mobile robot for the purpose of space exploration and mapping is developed and presented. With the help of machine learning post-processing of obtained data in IMU (Inertial Measurement Unit), a machine learning classifier using the random forest algorithm with 100 trees is developed which can classify terrain based on IMU inputs to further develop more complex surface maps. The ML classifier data is compared to the raw input data from the sensor on the basis of a rigid range based frequency of sensor data and the frequencies obtained from the classifier and an analysis of the classifier’s accuracy is performed. Based on the accuracy of the machine learning model in predicting the type of terrain and comparison with the rule-based rigid system, an approximate of the machine learning model’s accuracy can be obtained. The project was developed with hardware cost under ${\$}$30.