Nowadays highly accurate and mobile robots are able to perceive and manipulate their environment. Such robots can be used to assist humans in everyday tasks to reduce their workload. As a consequence, communication and interaction between humans and robots is becoming more important. In this thesis we present a system for human-robot collaboration for on-table tasks. The system is implemented for the PR2 research and development platform. Using the robot's depth sensor the objects on a table are detected and tracked during task execution. We analyzed on-table tasks in general and used the system to implement three different example tasks: a task that requires the human and the robot to order blocks by color, a task where a bridge is built and a task with a joint action. Additionally, we implemented five action-selection strategies for the robot: proactive, autonomous, reactive, human-requested and human-commands. We conducted a user study to compare the interaction modes during the bridge task and to evaluate the usage of the system in a real scenario. The answers of the participants indicate that for the bridge-building task, people prefer to interact with a robot using the proactive, autonomous or human-commands action-selection strategy rather than with the reactive or the human-requested strategy. The user study demonstrates that the system is useful for tabletop human-robot interaction with a real robot. Furthermore, the system can serve as a basis for further investigations in the area of human-robot interaction.