Multiagent environments are often highly dynamic and only partially observable which makes deliberative action planning computationally hard. In many such environments, however, agents can take a more proactive approach and suspend planning for partial plan execution, especially for active information gathering and interaction with others. This paper presents a new algorithm for Continual Collaborative Planning (CCP) that enables agents to deliberately interleave planning, acting, perception and communication. Our implementation of CCP has been evaluated with MAPSIM, a tool that automatically generates multiagent simulations from formal multiagent planning (MAP) domains. For different such simulations, we show how CCP leads to collaborative planning and acting and, despite minimal linguistic capabilities, to fairly natural dialogues between agents.

Situated dialogue is usually tightly integrated with behavior planning, physical action and perception. This paper presents an algorithmic framework, Continual Collaborative Planning (CCP), for modeling this kind of integrated behavior and shows how CCP agents naturally blend physical and communicative actions. For experiments with conversational CCP agents we have developed MAPSIM, a software environment that can generate multiagent simulations from formal multiagent planning problems automatically. MAPSIM permits comparison of CCP-based dialogue strategies on a wide range of domains and problems without domain-specific programming. Despite their linguistic capabilities being limited MAPSIM agents can already engage in fairly realistic situated dialogues.

The article presents a cross-modal approach to modeling spatio-temporal comprehension in situated dialogue. The article proposes an approach for representing spatiotemporal-causal structure at the level of linguistically conveyed meaning, adopting the notion of event nucleus presented [?]. In the approach, basic tense, aspect and modality can be captured, as well as aspectual coercion, and temporal sequencing. The article then discusses how the incremental construction of such linguistic representations can be combined with continuous action planning. Through cross-modal integration of action planning representations into linguistic processing, the article explores how action planning can prime selectional attention in utterance comprehension by disambiguating linguistic analyses on the basis of plan availability, and by raising expectations what action(s) may be talked about next. Furthermore, planning can complement linguistic analyses with information about spatiotemporal-causal structure established in planning inferences. This makes such inferences available for future referencing in the discourse context, yet lessening the load on dialogue comprehension for having to establish them.

For a robot to be able to first understand and then achieve a human's goals, it must be able to reason about a) the context of the current situation (with respect to which it must interpret the human's commands) and b) the future world state (as intended by the human) and how to achieve it. Since humans express their intentions and plans using qualitative symbolic representations, robots must be enabled to reason and interact on the same representational level. In this paper, we describe the use of classical AI Planning techniques for situation-aware interpretation and execution of human commands. We show how, based on a Planning domain, a robot can be enabled to understand commands in natural language, plan for their situation-dependent realization and revise its plans based on new perceptions. We show the effectiveness of this approach in several HRI scenarios modeled as Planning domains as well as with examples from a real robot system developed in the EU-funded CoSy project.

In this paper, we consider the problem of online failure detection and isolation for mobile robots. The goal is to enable a mobile robot to determine whether the system is running free of faults or to identify the cause for faulty behavior. In general, failures cannot be detected by solely monitoring the process model for the error free mode because if certain model assumptions are violated the observation likelihood might not indicate a defect. Existing approaches therefore use comparably complex system models to cover all possible system behaviors. In this paper, we propose the mixed-abstraction particle lter as an efcient way of dealing with potential failures of mobile robots. It uses a hierarchy of process models to actively validate the model assumptions and distribute the computational resources between the models adaptively. We present an implementation of our algorithm and discuss results obtained from simulated and real-robot experiments.

In highly dynamic environments, e.g. multiagent systems, finding optimal action plans is practically impossible since individual agents lack important knowledge at planning time or this knowledge has become obsolete when a plan is executed. It is often more practical in such environments to enable agents to actively extend their knowledge as part of their plans and then revise their decisions in light of these update. In this paper, we describe a new principled approach to Continual Planning, i.e. the integration of Planning, Execution and Monitoring. The algorithm deliberately postpones parts of the planning process to later stages in an agent's plan-act-monitor cycle and automatically determines when to switch back to refining or revising a partly executed plan. To evaluate our (and others') Continual Planning techniques we have developed a simulation environment where formal MA Planning domains are not only used by planning agents but also as the basis of the simulation model such that agents can not only plan, but execute actions and perceive their environment. Our experiments show that, using continual planning techniques, deliberate action planning can be used efficiently even in complex multiagent environments.