An Account of Intensional and Extensional Actions, and its Application to Belief, Nondeterministic Actions and Fallible Sensors

In general, an agent may have incomplete and inaccurate knowledge about its environment. As well, actions may not turn out as intended or may have nondeterministic effects, and sensors may on occasion give incorrect results. We present a general, qualitative approach to reasoning about action and change in such a setting. The approach is expressed as an extension to basic action theories in the situation calculus, where an agent's epistemic state is modelled by a set of situations, where each situation is assigned a non-negative integer representing its plausibility. The agent's epistemic state is updated by modifying these plausibility values after the execution of an action, taking into account the possibility of unexpected results. To this end, we consider actions to have an intensional aspect, under the control of and determined by the agent, and an extensional aspect, not directly accessible to the agent and controlled by "nature". This leads to two distinct but related related notions of belief, an extensional "bird's eye" view which models an agent's beliefs wrt actually-executed actions, and an intensional view representing beliefs from the agent's point of view. We argue that the approach is significantly more general and comprehensive than previous accounts, and leads to a unified view of failed actions and nondeterminism with respect to physical and sensing actions.

Belief Change without Compactness

Dealing with dynamics is a vital problem in Artificial Intelligence (AI). An intelligent system should be able to perceive and interact with its environment to perform its tasks satisfactorily. To do so, it must sense external actions that might interfere with its tasks, demanding the agent to self-adapt to the environment dynamics. In AI, the field that studies how a rational agent should change its knowledge in order to respond to a new piece of information is known as Belief Change. It assumes that an agent’s knowledge is specified in an underlying logic that satisfies some properties including compactness: if an information is entailed by a set X of formulae, then this information should also be entailed by a finite subset of X. Several logics with applications in AI, however, do not respect this property. This is the case of many temporal logics such as LTL and CTL. Extending Belief Change to these logics would provide ways to devise self-adaptive intelligent systems that could respond to change in real time. This is a big challenge in AI areas such as planning, and reasoning with sensing actions. Extending belief change beyond the classical spectrum has been shown to be a tough challenge, and existing approaches usually put some constraints upon the system, which are either too restrictive or dispense some of the so desired rational behaviour an intelligent system should present. This is a summary of the thesis “Belief Change without Compactness” by Jandson S Ribeiro. The thesis extends Belief Change to accommodate non-compact logics, keeping the rationality criteria and without imposing extra constraints. We provide complete new semantic perspectives for Belief Change by extending to non-compact logics its three main pillars: the AGM paradigm, the KM paradigm and Non-monotonic Reasoning.

Hybrid conditional planning for robotic applications

Robots who have partial observability of and incomplete knowledge about their environments may have to consider contingencies while planning, and thus necessitate cognitive abilities beyond classical planning. Moreover, during planning, they need to consider continuous feasibility checks for executability of the plans in the real world. Conditional planning is concerned with reaching goals from an initial state, in the presence of incomplete knowledge and partial observability, by considering all contingencies and by utilizing sensing actions to gather relevant knowledge when needed. A conditional plan is essentially a tree of actions where each branch of the tree represents a possible execution of actuation actions and sensing actions to reach a goal state. Hybrid conditional planning extends conditional planning by integrating feasibility checks into executability conditions of actions. We introduce a parallel offline algorithm, called HCPlan, for computing hybrid conditional plans. HCPlan relies on modeling deterministic effects of actuation actions and non-deterministic effects of sensing actions in the causality-based action language [Formula: see text]. Branches of a hybrid conditional plan are computed in parallel using a SAT solver, where continuous feasibility checks are performed as needed. We develop a comprehensive benchmark suite and introduce new evaluation metrics for hybrid conditional planning. We evaluate HCPlan with extensive experiments in terms of computational efficiency and plan quality. We perform experiments to compare HCPlan with other related conditional planners and approaches to deal with contingencies due to incomplete knowledge. We further demonstrate the applicability and usefulness of HCPlan in service robotics applications, through dynamic simulations and physical implementations.

KABouM: Knowledge-Level Action and Bounding Geometry Motion Planner

For robots to solve real world tasks, they often require the ability to reason about both symbolic and geometric knowledge. We present a framework, called KABouM, for integrating knowledge-level task planning and motion planning in a bounding geometry. By representing symbolic information at the knowledge level, we can model incomplete information, sensing actions and information gain; by representing all geometric entities--objects, robots and swept volumes of motions--by sets of convex polyhedra, we can efficiently plan manipulation actions and raise reasoning about geometric predicates, such as collisions, to the symbolic level. At the geometric level, we take advantage of our bounded convex decomposition and swept volume computation with quadratic convergence, and fast collision detection of convex bodies. We evaluate our approach on a wide set of problems using real robots, including tasks with multiple manipulators, sensing and branched plans, and mobile manipulation.

An Epistemic Event Calculus for ASP-based Reasoning About Knowledge of the Past, Present and Future

We present a generalisation of the Event Calculus, specified in classical logic and implemented in ASP, that facilitates reasoning about non-binary-valued fluents in domains with non-deterministic, triggered, concurrent, and possibly conflicting actions. We show via a case study how this framework may be used as a basis for a "possible-worlds" style approach to epistemic and causal reasoning in a narrative setting. In this framework an agent may gain knowledge about both fluent values and action occurrences through sensing actions, lose knowledge via non-deterministic actions, and represent plans that include conditional actions whose conditions may be initially unknown.

Hybrid conditional planning using answer set programming

We introduce a parallel offline algorithm for computing hybrid conditional plans, called HCP-ASP, oriented towards robotics applications. HCP-ASP relies on modeling actuation actions and sensing actions in an expressive nonmonotonic language of answer set programming (ASP), and computation of the branches of a conditional plan in parallel using an ASP solver. In particular, thanks to external atoms, continuous feasibility checks (like collision checks) are embedded into formal representations of actuation actions and sensing actions in ASP; and thus each branch of a hybrid conditional plan describes a feasible execution of actions to reach their goals. Utilizing nonmonotonic constructs and nondeterministic choices, partial knowledge about states and nondeterministic effects of sensing actions can be explicitly formalized in ASP; and thus each branch of a conditional plan can be computed by an ASP solver without necessitating a conformant planner and an ordering of sensing actions in advance. We apply our method in a service robotics domain and report experimental evaluations. Furthermore, we present performance comparisons with other compilation based conditional planners on standardized benchmark domains.

Decreasing Uncertainty in Planning with State Prediction

In real world environments the state is almost never completely known. Exploration is often expensive. The application of planning in these environments is consequently more difficult and less robust. In this paper we present an approach for predicting new information about a partially-known state. The state is translated into a partially-known multigraph, which can then be extended using machine-learning techniques. We demonstrate the effectiveness of our approach, showing that it enhances the scalability of our planners, and leads to less time spent on sensing actions.

Replanning in Domains with Partial Information and Sensing Actions

Replanning via determinization is a recent, popular approach for online planning in MDPs. In this paper we adapt this idea to classical, non-stochastic domains with partial information and sensing actions, presenting a new planner: SDR (Sample, Determinize, Replan). At each step we generate a solution plan to a classical planning problem induced by the original problem. We execute this plan as long as it is safe to do so. When this is no longer the case, we replan. The classical planning problem we generate is based on the translation-based approach for conformant planning introduced by Palacios and Geffner. The state of the classical planning problem generated in this approach captures the belief state of the agent in the original problem. Unfortunately, when this method is applied to planning problems with sensing, it yields a non-deterministic planning problem that is typically very large. Our main contribution is the introduction of state sampling techniques for overcoming these two problems. In addition, we introduce a novel, lazy, regression-based method for querying the agent's belief state during run-time. We provide a comprehensive experimental evaluation of the planner, showing that it scales better than the state-of-the-art CLG planner on existing benchmark problems, but also highlighting its weaknesses with new domains. We also discuss its theoretical guarantees.

ALPprolog – A new logic programming method for dynamic domains

Logic programming is a powerful paradigm for programming autonomous agents in dynamic domains as witnessed by languages such as Golog and Flux. In this work we present ALPprolog, an expressive, yet efficient, logic programming language for the online control of agents that have to reason about incomplete information and sensing actions.