Forschungsthemen
[MA] Self-modelling the Driving Behavior of Robots
Robotic systems are used in many areas ranging from industrial settings over search and rescue settings and even at home. A particularly challing area is the use of robots in unreachable areas (moon, mars, or areas with high radiation). The robot should autonomously perform a given task, e.g., building a map of the environment and taking samples or measurements. A key problem in such a setting is that you cannot assume that the robot moves correctly, because the how the robot moves depends on the environment which is unknown at design time. For example, if the robot shall move one meter straight by simply letting his tires make a certain amount of cycles it will most likely not reach the exact target position. This is because the floor could have different characteristics (slippery, sandy, gluey). As the robot cannot be remotely controlled, it needs to learn the floor characteristics by itself and has to use this model to properly navigate. This capability is called self-modelling~\cite{B17}, i.e., the ability of a system to build and use models of the environment and the itself by itself. Initial thoughts on how to add this capability to driving robots are summarized in~\cite{wang2021}. The task of this thesis is to realize the self-modelling capability for driving robots. For this, Gazebo~\footnote{https://classic.gazebosim.org/} shall be used as simulation environment and the robot operating system (ROS)~\footnote{https://www.ros.org/} to implement the robot's software. For demonstration purposes, an appropriate world shall be created in Gazebo and the robot shall provide a graphical dashboard showing the self-modelled runtime model during the simulation. The research question to answer is: is it possible to enable a robot to automatically learn the floor characteristics whilst performing an exploration task and to utilize the self-modelled knowledge. The tasks in detail are:
- Getting acquainted to Gazebo and ROS
- Literature review on self-modelling for robotic systems
- Creating a Gazebo world with different types of floors (at least standard, slippery/icy and gluey)
- Realization of a robot with self-modelling capability to automatically learn the floor characteristics, to show them in a dashboard and to utilize the knowledge for navigation
- The approach has to evaluated using three example scenarios within the created Gazebo world
- Initial thoughts on how to generalize the approach to other types of movement (e.g., drones) have to be described.
Betreuer: Sebastian Götz