Initial State Prediction in Planning (original) (raw)

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 machinelearning 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.

Automated Planning With Invalid States Prediction

IEEE Access, 2021

The increase of automated systems in space missions raises concerns about safety and reliability in operations carried out by satellites due to performance degradation. There have been several studies about the automatic planning process, but many approaches are generated with invalid states. The invalid state can be understood as a prohibited, degraded or risky scenario for the domain. This paper proposes an automated planning process with restrictions that enables automatic planners to not generate plans with invalid states. We implement a validator method for the planner software which proves that plan generation matches the restrictions imposed on the domain. In the experiments, we test an automatic planning process that is specific to the aerospace area, where a knowledge base with invalid states is available in the context of the operation of a satellite. Our proposal to carry out the verification of invalid states in automatic planning, can contribute to plans being generated with higher quality, ensuring that the goal of a plan is only achieved through valid intermediate states. It is also expected that the generated plans will be executed with better performance and will require less computational resources, since the search space is reduced. INDEX TERMS Automated planning, domain rule learning, machine learning, PDDL.

Planning with imperfect information

2004 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (IEEE Cat. No.04CH37566), 2004

We describe an efficient method for planning in environments for which prior maps are plagued with uncertainty. Our approach processes the map to determine key areas whose uncertainty is crucial to the planning task. It then incorporates the uncertainty associated with these areas using the recently developed PAO* algorithm to produce a fast, robust solution to the original planning task.

Planning with Incrementally Created Graphs

A framework for planning algorithms that is both optimal and complete will be presented. The algorithm allows for the planning of optimal paths through a multi-level hierarchy of annotated graph spaces. A rich world model that contains multi-resolution attributes, costs, and predicates controls the integrated incremental construction and evaluation of the planning graph in dynamic environments. The properties of this framework are proven and a number of example problem domains are presented that show how this framework can incorporate both hard and soft constraints during both graph construction and arc evaluation.

Improving Determinization in Hindsight for Online Probabilistic Planning

2010

Abstract Recently,'determinization in hindsight'has enjoyed surprising success in on-line probabilistic planning. This technique evaluates the actions available in the current state by using non-probabilistic planning in deterministic approximations of the original domain. Although the approach has proven itself effective in many challenging domains, it is computationally very expensive. In this paper, we present three significant improvements to help mitigate this expense.

Planning under uncertainty in dynamic domains

1997

Planning, the process of nding a course of action which can be executed to achieve some goal, is an important and well-studied area of AI. One of the central assumptions of classical AI-based planning is that after performing an action the resulting state can be predicted completely and with certainty. This assumption has allowed the development of planning algorithms that provably achieve their goals, but it has also hindered the use of planners in many real-world applications because of their inherent uncertainty. Recently, several planners have been implemented that reason probabilistically about the outcomes of actions and the initial state of a planning problem. However, their representations and algorithms do not scale well enough to handle large problems with many sources of uncertainty. This thesis introduces a probabilistic planning algorithm that can handle such problems by focussing on a smaller set of relevant sources of uncertainty, maintained as the plan is developed. This is achieved by using the candidate plan to constrain the sources of uncertainty that are considered, incrementally considering more sources as they are shown to be relevant. The algorithm is demonstrated in an implemented planner, called Weaver, that can handle uncertainty about actions taken by external agents, in addition to the kinds of uncertainty handled in previous planners. External agents may cause many simultaneous changes to the world that are not relevant to the success of a plan, making the ability t o determine and ignore irrelevant e v ents a crucial requirement for an e cient planner. Three additional techniques are presented that improve the planner's e ciency in a n umber of domains. First, the possible external events are analyzed before planning time to produce factored Markov c hains which can greatly speed up the probabilistic evaluation of the plan when structural conditions are met. Second, domainindependent heuristics are introduced for choosing an incremental modi cation to apply to the current plan. These heuristics are based on the observation that the failure of the candidate plan can be used to condition the probability that the modi cation will be successful. Third, analogical replay is used to share planning e ort across branches of the conditional plan. Empirical evidence shows that Weaver can create high-probability plans in a planning domain for managing the clean-up of oil spills at sea.

Closing the learning-planning loop with predictive state representations

2010

A central problem in artificial intelligence is that of planning to maximize future reward under uncertainty in a partially observable environment. In this paper we propose and demonstrate a novel algorithm which accurately learns a model of such an environment directly from sequences of action-observation pairs. We then close the loop from observations to actions by planning in the learned model and recovering a policy which is near-optimal in the original environment. Specifically, we present an efficient and statistically consistent spectral algorithm for learning the parameters of a Predictive State Representation (PSR). We demonstrate the algorithm by learning a model of a simulated high-dimensional, vision-based mobile robot planning task, and then perform approximate point-based planning in the learned PSR. Analysis of our results shows that the algorithm learns a state space which efficiently captures the essential features of the environment. This representation allows accurate prediction with a small number of parameters, and enables successful and efficient planning.

Symbolic generalization for on-line planning

2002

Symbolic representations have been used successfully in off-line planning algorithms for Markov decision processes. We show that they can also improve the performance of online planners. In addition to reducing computation time, symbolic generalization can reduce the amount of costly real-world interactions required for convergence. We introduce Symbolic Real-Time Dynamic Programming (or sRTDP), an extension of RTDP. After each step of on-line interaction with an environment, sRTDP uses symbolic modelchecking techniques to generalizes its experience by updating a group of states rather than a single state. We examine two heuristic approaches to dynamic grouping of states and show that they accelerate the planning process significantly in terms of both CPU time and the number of steps of interaction with the environment.

State Agnostic Planning Graphs: Deterministic, Non-Deterministic, and Probabilistic Planning

Artificial Intelligence, 2009

Planning graphs have been shown to be a rich source of heuristic information for many kinds of planners. In many cases, planners must compute a planning graph for each element of a set of states, and the naive technique enumerates the graphs individually. This is equivalent to solving a multiple-source shortest path problem by iterating a single-source algorithm over each source.

Planning with State Uncertainty via Contingency Planning and Execution Monitoring

This paper proposes a fast alternative to POMDP planning for domains with deterministic state-changing actions but probabilistic observation-making actions and partial observability of the initial state. This class of planning problems, which we call quasi-deterministic problems, includes important realworld domains such as planning for Mars rovers.The approach we take is to reformulate the quasi-deterministic problem into a completely observable problem and build a contingency plan where branches in the plan occur wherever an observational action is used to determine the value of some state variable. The plan for the completely observable problem is constructed assuming that state variables can be determined exactly at execution time using these observational actions. Since this is often not the case due to imperfect sensing of the world, we then use execution monitoring to select additional actions at execution time to determine the value of the state variable sufficiently accurately. We use a value of information calculation to determine which information gathering actions to perform and when to stop gathering information and continue execution of the branching plan. We show empirically that while the plans found are not optimal, they can be generated much faster, and are of better quality than other approaches.