Experimental setup for collision avoidance algorithms for mobile robots

Collision avoidance is a topic of utmost relevance in mobile robot navigation, where robots should be able to reach a goal and to avoid obstacles on their way autonomously, to operate consistently in a real-life environment. For this purpose, several algorithms have been developed over the last years. This work aims at providing a real-time collision-free path for mobile robots, implementing in real world a pre-existent collision avoidance algorithm, which has been tested so far in a simulation setting only. Specifically, the chosen technique improves the classical artificial potential fields by considering local attractors in addition to repulsors, in order to drive the robot towards a goal, while avoiding obstacles along preferred directions. The first phase consisted in reproducing the conditions of the simulated environment in a laboratory setup made of a single obstacle and a mobile robot controlled by the algorithm, that has been implemented in ROS (Robot Operating System). In this first step, the pose of the obstacle was known and fixed and the robot pose was estimated by odometry sensors in terms of position and orientation. At the end of this process, it was possible to identify potential improvements with the purpose of perfectioning position measurements and obtaining an online obstacle detection. This was achieved by using a color camera and ArUco markers, opportunely placed in order to track the pose of the relevant elements: robot, obstacle and final pose. With the results achieved in the first phase, the effort focused in studying a significant application for mobile robots, where obstacles can either be objects or humans. Thus, the final step of this work consisted of different tests to prove the effectiveness of the proposed technique. Each test is characterized by the same initial and final pose of the robot, but different obstacle poses, so that the robot approaches to the object from different directions. Results show that the robot is able to pass on the desired side with respect to the obstacle, i.e. the side where the local attractor is placed. Future works will involve the study of some aspects that are still to be developed to obtain a reliable result and bring this technique in real life, such as dynamic obstacle and multiple obstacles.