We present some techniques for planning in domains specified with the recent standard language PDDL2.1, supporting 'durative actions' and numerical quantities. These are implemented LPG, a domain-independent planner that took part 3rd International Planning Competition (IPC). LPG is an incremental, any time system producing multi-criteria quality plans. The core of based on stochastic local search method graph-based representation called 'Temporal Action Graphs' (TA-graphs). This paper...

The treatment of exogenous events in planning is practically important many real-world domains where the preconditions certain plan actions are affected by such events. In this paper we focus on temporal with that happen at known times, imposing constraint must be executed during some predefined time windows. When have durations, handling constraints adds an extra difficulty to planning. We propose approach these which integrates constraint-based reasoning into a graph-based framework using...

Reasoning about temporal information is an important task in many areas of Artificial Intelligence. In this paper we address the problem scalability reasoning by providing a collection new algorithms for efficiently managing large sets qualitative relations. We focus on class relations forming Point Algebra (PA-relations) and major extension to include binary disjunctions PA-relations (PA-disjunctions). Such add great deal expressive power, including ability stipulate disjointness intervals,...

We propose some domain-independent techniques for bringing well-founded partial-order planners closer to practicality. The first two are aimed at improving search control while keeping overhead costs low. One is based on a simple adjustment the default A* heuristic used by UCPOP select plans refinement. other preferring ``zero commitment'' (forced) plan refinements whenever possible, and using LIFO prioritization otherwise. A more radical technique use of operator parameter domains prune...

We study classical planning for temporally extended goals expressed in Pure-Past Linear Temporal Logic (PPLTL). PPLTL is as expressive Linear-time on finite traces (LTLf), but shown this paper, it computationally much better behaved planning. Specifically, we show that can be encoded into with minimal overhead, introducing only a number of new fluents at most linear the goal and no spurious additional actions. Based these results, implemented system called Plan4Past, which used along...

The paper introduces a novel polynomial compilation technique for the sound and complete removal of conditional effects in classical planning problems. Similar to Nebel's effects, our solution also decomposes each action with into several simpler actions. However, it does so more effectively by exploiting actual structure given effects. We characterise such using directed graph leverage significantly reduce number additional atoms required, thereby shortening size valid plans. Our...

While several powerful domain-independent planners have recently been developed, no one of these clearly outperforms all the others in every known benchmark domain. We present PbP, a multi-planner which automatically configures portfolio by (1) computing some sets macro-actions for planner portfolio, (2) selecting promising combination and relative useful macro-actions, (3) defining running time slots their round-robin scheduling during planning. The configuration relies on knowledge about...

When designing state-of-the-art, domain-independent planning systems, many decisions have to be made with respect the domain analysis or compilation performed during preprocessing, heuristic functions used search, and other features of search algorithm. These design can a large impact on performance resulting planner. By providing alternatives for these choices exposing them as parameters, systems in principle configured work well different domains. However, planners are typically default...

Multi-task learning approaches have shown significant improvements in different fields by training related tasks simultaneously. The multi-task model learns common features among where they share some layers. However, it is observed that the approach can suffer performance degradation with respect to single task of natural language processing tasks, specifically sequence labelling problems. To tackle this limitation we formulate a simple but effective combines transfer learning. We use...

This paper studies an approach to planning with PDDL3 constraints involving mixed propositional and numeric conditions, as well metric time constraints. We show how the whole instantaneous actions can be compiled away into a problem without constraints, enabling use of any state-of-the-art planner that is agnostic existence PDDL3. Our solution exploits concept regression. In addition basic compilation, we present optimized variant based on observation it possible make compilation sensitive...

Fast plan adaptation is important in many AI applications. From a theoretical point of view, the worst case adapting an existing to solve new problem no more efficient than complete regeneration plan. However, practice can be much generation, especially when adapted obtained by performing limited amount changes original In this paper, we investigate domain-independent method for that modifies replanning within temporal windows containing portions need revised. Each window associated with...