Heterogeneous robot systems are characterized by the robots' diversity, each contributing different capabilities that can be utilised to complete tasks. Such diversity does not only emerge from differences in hardware, but also from distinct robot software and algorithms. Due to the large variety of robotic systems available today, there is a need for flexible software architectures enabling easy integration of new robot configurations into existing systems. Platform-independence is highly desirable. One step in achieving platform-independent software systems is to identify common functionalities for which interfaces can be designed such that specific implementations may vary among platforms, while interaction between the common modules remains platform-independent. A robot capability may be such a functionality which can be provided by several different robots, regardless varying implementations. A platform-independent model defining such robot capabilities can also include a measure of quality or extent for each capability, describing the individual robot's skill level for a capability. Once it is determined which capabilities are required to complete a task, such quality measures provide a means for evaluating and comparing the robots' suitability for the task. Finding which capabilities are required to solve a task is part of the task planning step. When capabilities are defined such that a whole task can be described by them (e.g., reach, grasp, lift, then move), a planner can find task solutions consisting of robot capabilities required to achieve a goal. This thesis proposes a platform-independent model of robot capabilities and develops a framework for task planning among teams of robots based on this capability model. Individual capabilities are examined in detail, discussing possible quality measures and establishing required data flow between capabilities. The task planner can find and evaluate a plan based on various degrees of approximate information available at planning time. Existing planning techniques need to be extended in order to reflect concurrent data flow in a robotic system. For execution of the tasks on the robots, a framework based on the Robot Operating System (ROS) implementing the capability model has been developed. Results show that the capability model can provide an integrated framework for task planning and execution. The work is evaluated by empirical results in simulation and on a real robot.