Cognitive robots are all over the place. Well, at least that is what it looks like in research lab agendas on the planet.Some basic and fundamental problems in controlling mobile robots have been quite well understood in recent years: Localization and map building are chief examples. Now, as autonomous mobility as such is much less of a challenge, research resumes an activity that has started about half a century ago in mainstream AI; that got out of the AI mainstream for a while, as other challenges turned out to offer solutions faster; that got a little mocked for some time ("good old-fashioned AI", gofai); yet that has remained a fascinating scientific problem, and potentially a rich source of application projects: algorithmic models for goal-directed physical action in the world.So in a sense, very recent and ambitious projects in cognitive robotics, or cognitive technical systems, or robot cognition, or whatever you prefer to call them, are taking up that thread, which, I would say, goes back to the pioneering shakey project in the AI of the 1960s, and has remained unresolved ever since. However, there are a number of interesting differences between thinking about a cognitive robot in the 1960s and doing so today: Our modern computers are nearly infinitely more powerful; sensors are much better; decent robot hardware has become affordable even for a normal university lab; and, the proper share of AI: 50 years of AI research has yielded quite a number of wellunderstood algorithms and formalisms for planning, knowledge representation and reasoning that were unknown in the early days.And still, getting a cognitive robot to work is a challenge -which is why cognitive robots are not yet all over the place, and the cognition in those that are may fail to impress. Among a zoo of problems from "just" technical (integration!) to big AI problem (symbol grounding!) is a specimen that has to do with the proper AI share of the story: Using a planner or reasoner within the control software of an autonomous cognitive robot is different from using them in an off-line or human-in-the-loop system. Now this is the point where Ronny Hartanto's thesis comes into play. A robot working, for example, in a mundane household domain, would need to
May 2011Joachim Hertzberg
PrefaceAction planning has been used in the field of robotics for solving long-running tasks. In the robot architectures field, it is also known as the deliberative layer. However, there is still a gap between the symbolic representation on the one hand and the low-level control and sensor representation on the other. In addition, the definition of a planning problem for a complex, real-world robot is not trivial. The planning process could become intractable as its search spaces become large. Since the defined planning problem determines the complexity and the computability for solving the problem, it should contain only relevant states. In this work, a novel approach which amalgamates description logic (dl) reasoning with hierarchical task network (htn) planning i...