The vision

The overall requirement on any service robot system is to be capable of responding intelligently to situations that have not been specified in the robotics system design. In this project we focus on solutions by developing new methods to cope with such novelty in manipulation tasks. The overall aim of the GeRT project is to enable a robot to autonomously generalise its manipulation skills from a set of known objects to previously un-manipulated objects in order to achieve an everyday manipulation task.

To achieve this we shall adopt the working hypothesis that the roles of objects and actions in an abstract description of a task should be acquired from specific examples of that kind of task. Our vision is to start with an existing control program for each instantiation of an abstract task (i.e. a specific set of objects). These control programs work only for their specific objects. Such an example control program would in fact consist of a number of smaller programs, such as a high level task plan, grasp points for specific objects, impedance controllers, trajectories, and programs for grasping objects, or performing manipulations with them (such as twisting the lid off a jar). These programs then constitute a database of prototype programs for that class of task. When confronted with a novel instance of the same task the robot needs to autonomously create the new control program by generalizing from these prototype programs, establishing appropriate correspondences between objects and actions in the prototypes and their counterparts in the novel tasks instance. In this way, prototypical task plans are adapted to generate a new plan that is suitable for handling different geometric, kinematic, and dynamic task settings, hence solving a task that is physically substantially different but similar at an abstract level. This kind of knowledge transfer or generalization will not just be restricted to the most abstract layer: low-level motor primitives as well as high-level logical robotic actions and operators will be adapted to the new instance.