Robots Find a Better Way: A Learning Method for Mobile Robot Navigation in Partially Unknown Environments

This paper represents a method for mobile robot navigation in environments where obstacles are partially unknown. The method uses a path selection mechanism that creates innovative paths through the unknown environment and learns to use routes that are more reliable. This approach is implemented on Khepera robot and verified against shortest path following by wave transform algorithms. Based on the experimental data, we claim that robot’s trajectory planned by wave transform algorithms is very difficult to predict and control unless the environment is completely modeled and the localization errors are small. We show that even small unmodeled obstacles can cause large deviation from the preplanned path. Our complementary approach of path selection decreases the risk of path following as well as increases the predictability of robot’s behaviour.