Optimized Machine Learning Model for Autonomous Robot Navigation Prediction

The primary problem at hand revolves around the domain of autonomous robot navigation. It is framed within the specific scenario of a robot, the SCITOS G5 navigating along the four walls of an enclosed room in clockwise direction for 4 rounds, equipped with 24 Ultrasound sensors circularly positioned around its waist, the dataset for which is acquired from Kaggle. This task, although seemingly simple is a non-linearly separable classification task and thus, linear classifiers are unable to learn and command the robot around the room without collisions. To address this challenge, a decision tree classifier model is employed, trained on a wealth of sensor data collected during the robot’s navigation endeavours. The objective is to harness the power of machine learning and predictive analytics to enable the robot to and accurately predict its upcoming turns, thereby enhancing its navigation capabilities. The model’s strengths emanate from its ability to expound upon decision trees, scrutinize the importance of features, and capture complex nonlinear relationships. This research serves as a conduit connecting predictive modelling with applications in robotics, providing valuable insights and avenues for enhancing the precision and resilience of turn prediction models. Prospects for forthcoming research include the exploration of ensemble methods, the integration of temporal dependencies, advanced methodologies for feature engineering, considerations for real-time deployment, and assessments spanning a variety of environmental conditions.