The aim of this work is to discuss how Logic Programming (LP) can meet the increasing requirements of robot applications, with respect to programming models, techniques, and tools. A new approach to the integration between object-oriented, knowledge-based and logic programming is introduced, by exploiting an extension to LP called Contextual Logic Programming, which allows users to define, both statically and dynamically, extendible software components, and to promote incremental design and development of (declarative) software systems. A support for task-level robot programming (called CARA, Contextual Agent Robot Architecture) is introduced as an extension/specialization of a general-purpose Contextual-LP environment. By providing high-level abstractions such as backtrackable objects and logic theories as communication devices, CARA seems to effectively help reducing the gap between low-level and high-level software layers, and coordinating the interaction of different, even heterogeneous, agents. In order to show CARA performance in real applications, a case-study is discussed where a robot has to perform a set of tasks of increasing complexity, from pure reactive behaviours to tasks involving high-level forms of reasoning.