Landmark Generation in HTN Planning Revisited

Hierarchical task network planning with resources and temporal constraints

In this paper, we address the question of how to create episodic memory based long-term affective interactions with a human-like robot. A key challange for long-term interaction is the recall of past important events during conversation. We suggest that episodic memory is a core concept for realizing this intelligence. In this paper, we propose a model of episodic memory and integrate it with a decision making module based on a Hierarchical Task Network (HTN) planner. Plans generated by the HTN planner are executed by a Finite-state-machine (FSM) based dialogue system in order to produce appropriate responses. Finally, we present a prototype implementation demonstrating the preliminary results we obtained.

Hierarchical task network (HTN) planning is well-known for being an efficient planning approach. This is mainly due to the success of the HTN planning system SHOP2. However, its performance depends on hand-designed search control knowledge. At the time being, there are only very few domain-independent heuristics, which are designed for differing hierarchical planning formalisms. Here, we propose an admissible heuristic for standard HTN planning, which allows to find optimal solutions heuristically. It bases upon the so-called task decomposition graph (TDG), a data structure reflecting reachable parts of the task hierarchy. We show (both in theory and empirically) that rebuilding it during planning can improve heuristic accuracy thereby decreasing the explored search space. The evaluation further studies the heuristic both in terms of plan quality and coverage.

When an actor executes a plan, action failures and exogenous events may lead to unexpected states that require replanning from the middle of plan execution. In Hierarchical Task Network (HTN) planning, unless the HTN methods have been carefully written to work well in unexpected states, replanning may either fail or produce plans that perform poorly.
 To overcome this problem, we introduce IPyHOP, a reentrant version of GTPyhop (a SHOP-like HTN planner), and Run-Lazy-Refineahead, a modification of the Run-Lazy-Lookahead actor that utilizes IPyHOP's reentrant replanning capability to replan during plan execution. In our experiments, Run-Lazy-Refineahead and IPyHOP expend less search effort (fewer decompositions and fewer iterations), find revised plans with fewer actions and lower total action cost, and finish execution with fewer failures.

Goal-driven automated composition of software components is an important problem with applications in Web service composition and stream processing systems. The popular approach to address this problem is to build the composition automatically using AI planning. However, it is shown that some of these planning approaches may neither be feasible nor scalable for many large-scale flow-based applications. Recent advances have proven that the automated composition problem can take advantage of expert knowledge describing the many ways in which different reusable components can be composed. This knowledge can be represented using an extensible composition template or pattern. In prior work, a flow pattern language called Cascade and its corresponding specialized planner have shown the best performance in these domains. In this paper, we propose the use of Hierarchical Task Network (HTN) planning for the composition of stream processing applications. To this end, we propose an automated approach of creating an HTN-based problem from the Cascade representation of the flow patterns. The resulting technique not only allows us to use the HTN planning paradigm and its many advantages including added expressivity but also enables optimization and customization of composition with respect to preferences and constraints. Further, we propose and develop a lookahead heuristic and show that it significantly reduces the planning time. We have performed extensive experimentation with stream processing applications and evaluated applicability and performance of our approach.

Action planning has been used in the field of robotics for solving long-running tasks. However, the definition of a planning problem for a complex, real-world robot is not trivial. The planning process could become intractable as its search spaces become large. The dissertation presented in this article introduces a novel approach which amalgamates Description Logic (dl) reasoning with Hierarchical Task Network (htn) planning. The planning domain description as well as fundamental htn planning concepts are represented in dl and can therefore be subject to dl reasoning; from these representations, concise planning problems are generated for htn planning.

This article presents a concept for scalable autonomy of UAVs applicable to the field of reconnaissance missions. Level of autonomy can be defined along the amount of human independence. Low human independence allows detailed control, while adding to the workload. Higher human independence provides functionalities inflicting less workload, but hides information from the pilot or inhibits from interacting with the system in depth. We present a concept to facilitate a range of human independence provided by an intelligent agent onboard the UAV. This enables the interaction of the human with the automation on different echelons of control. Instead of offering access to the different automation functions directly, the operator is guided along a hierarchical task tree. This allows the system to deduce the overall purpose of human interactions and facilitates support by informing about flaws in the plan and by suggesting alternatives. The concept was implemented resulting in an agent architecture, which is based on Hierarchical Task Network (HTN) planning and reasoning with the use of the Drools rule engine. The agent is commanded using the concept of task-based guidance, which represents the highest human independence currently implemented in the system. The research for guiding multiple UAVs on a team level is planned as a next step. We conducted tests with the implemented system in a simulated environment.

To scale-up to real-world problems, planning systems must be able to replan in order to deal with changes in problem context. In this paper we describe hierarchical task network and operator-based replanning techniques which allow adaptation of a previous plan to account for problems associated with executing plans in real-world domains with uncertainty, concurrency, changing objectives. We focus on replanning which preserves elements of the original plan in order to use more reliable domain knowledge and to facilitate user understanding of produced plans. We present empirical results documenting the effectiveness of these techniques in a NASA antenna operations application.

It is widely believed, that the expressivity of STRIPS and STRIPS-like planning based on actions is generally lower than the expressivity of Hierarchical Task Network (HTN) and HTN-like planning, based on hierarchical decomposition. This would mean that a HTN-like planner can generally solve more domains than a STRIPS-like planner with the same extensions. In this paper, we show that both approaches, as they are practically used, are identically expressive and can solve all domains solvable by a Turing machine with finite tape (i.e. solvable by a common computer).

Deregulation of telecommunications has meant an increase in third-party service provision, personalized service delivery and integrated networks and media. The efficient allocation of services, without human intervention, to satisfy advanced service requirements spanning several networks is a crucial task. This can be modeled as a winner determination problem in combinatorial auctions where there are multiple services, service providers and winner determination criteria (like cost, bandwidth, delay, etc) but we have shown the problem is NP-complete. This paper describes a new two-stage algorithm for optimal anytime winner determination. In the first stage, a hierarchical task network planner is used to decompose a task into subtasks that can be solved by the available services. In the second stage, a genetic algorithm with heuristics is used to find the optimal combination of service providers to provide the services identified. We show through various experiments that the genetic algorithm finds optimal solutions quicker than a modified depth-first search algorithm.

This paper extends hierarchical task network (HTN) planning with lightweight learning, considering that in robotics, actions have a non-zero probability of failing. Our work applies to A*-based HTN planners with lifting. We prove that the planner finds the plan of maximal expected utility, while retaining its lifting capability and efficient heuristic-based search. We show how to learn the probabilities online, which allows a robot to adapt by replanning on execution failures. The idea behind this work is to use the HTN domain to constrain the space of possibilities, and then to learn on the constrained space in a way requiring few training samples, rendering the method applicable to autonomous mobile robots.

In the context of a business process modeled by commitments, agents enact a protocol by carrying out goals that service their part of commitments. In a competitive or even in a cooperative setting, an agent does not know for sure that its partners will successfully act on their part of the commitments. We introduce uncertainty into a successful recent approach of planning first-order commitment protocols. Probabilities reflect a semantics of the belief of an agent about the successful completion of tasks by other agents within the protocol, capturing notions of trust. We take a deterministic Hierarchical Task Network (HTN) planner, introduce probabilities into the task networks, and derive a protocol enactment which maximizes expected utility from the point of view of one agent. We illustrate our approach on a business scenario in e-commerce.

Three approaches to hierarchical planning have been widely discussed in the recent planning literature: Hierarchical Task Network (HTN) decomposition, model-reduction and operator decomposition. Abstraction is used in different ways in these three approaches and this has significance for both efficiency and expressive power. This paper identifies four issues that arise in the use of abstraction in planning which have been treated in different ways in the three approaches identified above. These issues are discussed with reference to an approach to abstraction which combines elements of the HTN and operator-decomposition approaches. Particular comparison is made with the HTN approach in order to highlight some important distinctions between the task decomposition and operator decomposition planning strategies. The CNF (Common Normal Form) case study, used by Erol to demonstrate certain features of the HTN approach, is used as the basis for this comparison.

Hierarchical Task Networks (HTNs) are a common model for encoding knowledge about planning domains in the form of task decompositions. We present a novel algorithm that uses invariant analysis to construct an HTN from the PDDL description of a planning domain and a single representative instance. The algorithm defines two types of composite tasks that interact to achieve the goal of a planning instance. One type of task achieves fluents by traversing invariants in which only one fluent can be true at a time. The other type of task applies a single action, which first involves ensuring that the precondition of the action holds. The resulting HTN can be applied to any instance of the planning domain, and is provably sound. We show that the performance of our algorithm is comparable to algorithms that learn HTNs from examples and use added knowledge.

This paper proposes a plan creation and execution system used by the agent HTN Fighter in the Fighting ICE game framework. The underlying approach implements a Hierarchical Task Network (HTN) planner and a simple planning domain that focuses on sequences of close-range attacks. The execution process is tightly interleaved with the planning process compensating for the uncertainty caused by the delay of 15 frames which the information about the game world state is provided with. Using an HTN and the proposed execution system, the agent is able to follow high-level strategies staying reactive to changes in the environment. Experiments show that HTN Fighter outperforms the sample MCTS controller and the top three controllers submitted to the 2016 Fighting Game AI Competition.