We address the problem of computing a policy for fully observable non-deterministic (FOND) planning problems. By focusing on relevant aspects state world, we introduce series improvements to previous art and extend applicability our planner, PRP, work in an online setting. The use relevance allows be exponentially more succinct representing solution FOND some domains. Through introduction new techniques avoiding deadends determining sufficient validity conditions, PRP has potential compute...

Many AI applications involve the interaction of multiple autonomous agents, requiring those agents to reason about their own beliefs, as well other agents. However, planning involving nested beliefs is known be computationally challenging. In this work, we address task synthesizing plans that necessitate reasoning We plan from perspective a single agent with potential for goals and actions non-homogeneous co-present observations, ability one if it were another. formally characterize our...

Temporally extended goals are critical to the specification of a diversity real-world planning problems. Here we examine problem non-deterministic with temporally specified in linear temporal logic (LTL), interpreted over either finite or infinite traces. Unlike existing LTL planners, place no restrictions on our formulae beyond those necessary distinguish from interpretations. We generate plans by compiling into instances described Planning Domain Definition Language that solved...

Planning is the branch of Artificial Intelligence (AI) that seeks to automate reasoning about plans, most importantly goes into formulating a plan achieve given goal in

LTL synthesis is the task of generating a strategy that satisfies Linear Temporal Logic (LTL) specification interpreted over infinite traces. In this paper we examine problem LTLf synthesis, variant where behaviour generate finite traces -- similar to assumption make in many planning problems, and important for business processes other system interactions duration. Existing approaches transform into deterministic finite-state automata (DFA) reduce DFA game. Unfortunately, transformation...

Temporal logics are useful for providing concise descriptions of system behavior, and have been successfully used as a language goal definitions in task planning. Prior works on inferring temporal logic specifications focused "summarizing" the input dataset - i.e., finding that satisfied by all plan traces belonging to given set. In this paper, we examine problem describe differences between two sets traces. We formalize concept such contrastive explanations, then present BayesLTL Bayesian...

Planning with sensing actions under partial observability is a computationally challenging problem that fundamental to the realization of AI tasks in areas as diverse robotics, game playing, and diagnostic solving. Recent work on generating plans for partially observable domains has advocated online planning, claiming offline are often too large generate. Here we push envelope this problem, proposing technique conditional (aka contingent) offline. The key our planner's success reliance...

We achieved a new milestone in the difficult task of enabling agents to learn about their environment autonomously. Our neuro-symbolic architecture is trained end-to-end produce succinct and effective discrete state transition model from images alone. target representation (the Planning Domain Definition Language) already form that off-the-shelf solvers can consume, opens door rich array modern heuristic search capabilities. demonstrate how sophisticated innate prior we place on learning...

We consider the problem of deriving formulas that capture traps, invariants, and dead-ends in classical planning through polynomial forms preprocessing. An invariant is a formula true initial state all reachable states. A trap conditional invariant: once reached makes true, states are from it will sat- isfy as well. Finally, satisfied make goal unreachable. introduce preprocessing algorithm computes traps k- DNF form exponential k parameter, show how can be used to precompute invariants...

Partial-order plans (POPs) are attractive because of their least-commitment nature, which provides enhanced plan flexibility at execution time relative to sequential plans. Current research on automated generation focuses producing plans, despite the appeal POPs. In this paper we examine POP by relaxing or modifying action orderings a optimize for criteria that promote flexibility. Our approach relies novel partial weighted MaxSAT encoding supports minimization deordering reordering actions....

We address the class of probabilistic planning problems where objective is to maximize probability reaching a prescribed goal. The complexity makes it difficult compute high quality solutions for large instances, and existing algorithms either do not scale, or so at expense solution quality. leverage core similarities between fully observable non-deterministic (FOND) construct sound, offline planner, ProbPRP, that exploits algorithmic advances from state-of-the-art FOND PRP, compact policies...

Partial-order plans (POPs) have the capacity to compactly represent numerous distinct plan linearizations and as a consequence are inherently robust. We exploit this robustness do effective execution monitoring. characterize conditions under which POP remains viable regression of goal through structure POP. then develop method for monitoring via structured policy, expressed an ordered algebraic decision diagram. The policy encompasses both state evaluation action selection, enabling agent...

Recent advances in fully observable non-deterministic (FOND) planning have enabled new techniques for various applications, such as behaviour composition, among others. One key limitation of modern FOND planners is their lack native support conditional effects. In this paper we describe an extension to PRP, the current state art planning, that supports generation policies domains with effects and non-determinism. We present core modifications PRP planner enhanced functionality without...

Generating complex multi-turn goal-oriented dialogue agents is a difficult problem that has seen considerable focus from many leaders in the tech industry, including IBM, Google, Amazon, and Microsoft. This large part due to rapidly growing market demand for capable of behaviour. Due business process nature these conversations, end-to-end machine learning systems are generally not viable option, as generated must be deployable verifiable on behalf businesses authoring them. In this work, we...

We investigate agent supervision, a form of customization, which constrains the actions an so as to enforce certain desired behavioral specifications. This is done in setting based on Situation Calculus and variant ConGolog programming language allows for nondeterminism, but requires remainder program after execution action be determined by resulting situation. Such programs can fully characterized set sequences that they generate. Hence operations like intersection difference become...