Robot following and escaping for human interfacing using LIDAR-UWB based robot dog

Abstract

Robots are increasingly becoming part of our everyday lives. Various efforts in commercial and research settings have made robots more accessible and capable of enhancing our daily lives with more comfort. Robots as a companion, in particular, is an idea that has been championed in mass media, and is now gradually becoming a reality. The Go1 Robot by Unitree is an example of a robot dog that seeks to bridge the human-robot gap, and will be the main robot of use in this project.

In this project, an interactive human-robot behaviour was developed, where the robot dog was able to follow a human subject. Ultra-Wideband (UWB) tags and Light Detection And Ranging (LiDAR) sensor were used in this project to achieve this capability. Basic behaviours are derived from the application of a slew of algorithms such as FAST-LIO2, A* global path planner, and DWA local path planner algorithms. These algorithms mainly work with the LiDAR to enable the robot to localise and navigate in the global coordinate frame. On the other hand, trilateration algorithms and Kalman trackers using ranging distances obtained from UWB will be used to localise the human subject.

To aggregate the behaviours, behaviour trees are used to build on these basic behaviours, creating complex behaviours that were used to explore different algorithms of subject tracking. The first behaviour explores using a single ranging distance to perform subject tracking using an Unscented Kalman Filter. The second behaviour, which had the benefit of four ranging distances, did not necessitate non-linear methods to correctly localise and track the subject, and was successful in its purpose to track and follow the Subject.