Next-Best-Sense: our latest paper is published on RA-L!

It is our delight to communicate that our latest paper entitled “Next-Best-Sense: a multi-criteria robotic exploration strategy for RFID tags discovery” has been accepted and published by IEEE Robotics and Automation Letters.

In this work, we aim to solve the problem of searching for, finding and localizing a set or Radio Frequency IDentification (RFID) tags randomly distributed in an environment while using a mobile robot equipped with an antenna. This work is relevant for any inventory management applications (e.g. Amazon warehouses) but it can be useful for any search applications, from human-rescue to nuclear radioactive source identification.

Next-Best-Sense (NBS) fits in the robotic exploration domain, in particular among the Next-Best-View approaches which take a local decision, only within the already sensed area in the robot surroundings. NBS draws from the Information Theory field and adapts Multi-Criteria Decision Making for taking intelligent decisions regarding the next robot position. Combining multiple exploration criteria in a utility function, all the possible new destinations are evaluated and the one with the highest associated utility values is chosen.

This method proved to quickly and successfully cover large environments while finding all the tags with a localization’s precision less than one meter. An example of the tag localization process is given in the following images. Here, The red arrow is the robot pose, the green circle is the ground truth position of one RFID tag. White cells are associated with a high probability of containing the tag. At the beginning (left), all the cells have a uniform probability. After emitting a signal, if the robot does not receive a reply, the surrounding cells within the detection field and in the line-of-sight are updated with a lower value, while the probability the tag is located elsewhere in the map is increased. Differently, if a reply is received, the probability is increased in the surrounding of the robot and decreased elsewhere. Please note how the probability moves towards the correct map position at different exploration stages (from top-left to bottom-right).

For more information on this work, please have a look at the paper at the following link. We also released our code on our Github page under MIT Licence, please be free to play with it 🙂