Proceedings Article•
Planning Domain + Execution Semantics: A Way Towards Robust Execution?
01 Jan 2014-
TL;DR: It is shown that the combined used of causal, temporal and categorical knowledge allows the robot to detect failures even when the effects of actions are not directly observable.
Abstract: Robots are expected to carry out complex plans in real world environments. This requires the robot to track the progress of plan execution and detect failures which may occur. Planners use very abstract world models to generate plans. Additional causal, temporal, categorical knowledge about the execution, which is not included in the planner's model, is often avail- able. Can we use this knowledge to increase robustness of execution and provide early failure detection? We propose to use a dedicated Execution Model to monitor the executed plan based on runtime observations and rich execution knowl- edge. We show that the combined used of causal, temporal and categorical knowledge allows the robot to detect failures even when the effects of actions are not directly observable. A dedicated Execution model also introduces a degree of mod- ularity, since the platform- and execution-specific knowledge does not need to be encoded into the planner.
Citations
More filters
08 Jan 2016
TL;DR: The results of three experimental scenarios in a restaurant domain supported the hypothesis that the robustness and efficiency of such a planning and execution-based system are improved by the addition of physics-based prediction, and questioned how a task-planning based robot system can be improved by prediction derived from physical simulation.
Abstract: The motivation for this work is the today's abstraction gap between the high-level robot control involved in symbolic planning and the low-level, fine-grained control of mobile robot motors. To deal with the complex, changeable environments in which humans live, state-of-the-art robots need to be able to exploit common-sense knowledge (i.e. physical laws) to calculate velocity, acceleration, friction and so on. Physical prediction is typically the domain of a physics simulation, such as Gazebo.
The literature contains examples in which symbolic planning and reasoning methods are extended, such as by adding geometric or temporal extensions or by offline recording of simulated actions.
The goal of this thesis is to question how a task-planning based robot system can be improved by prediction derived from physical simulation and to provide some answers. The goal-directed approach is to integrate a realistic prediction of robot activities so as to allow these activities to be adapted before execution fails. The hypothesis, therefore, is that a system that integrates such prediction is more efficient than a comparable system that does not.
A probabilistic prediction method named "functional imagination" and a system architecture have been developed. The prediction method was integrated into a blackboard-based robot control system and a PR2 robot and Gazebo were used for its evaluation. Experiments verified that simulation is indeed accurate and close to reality. The results of three experimental scenarios in a restaurant domain supported the hypothesis that the robustness and efficiency of such a planning and execution-based system are improved by the addition of physics-based prediction.
The work concludes by summarizing its findings, limitations and possible further research directions and with a future perspective on how parallel and cloud computing will affect the field of robotics in the context of simulation.
Der Ansatz dieser Arbeit ist motiviert von der "Abstraktionslucke" zwischen heutiger symbolischer Aufgabenplanung und Motoransteuerung. Denn gerade heutige Roboter mussen sich in komplexen und sich andernden Umgebungen zurechtfinden konnen. Damit sie das tun konnen, muss eine Verarbeitung von physikalischem Allgemeinwissen moglich sein. Die physikalische Vorhersage von Roboteraktionen ist dabei die Paradedisziplin von Physiksimulationen, wie Gazebo.
In der fachverwandten Literatur werden bereits symbolische Planung und Schlussfolgerung mit geometrischen sowie zeitlichen Komponenten, oder der Aufzeichung von offline simulierten Aktionen, erweitert.
Das Ziel dieser Arbeit ist die Beantwortung der Frage, wie ein planungs-basiertes Robotersystem sein Verhalten durch physik-basierte Vorhersage von Roboteraktionen verbessern kann. Der zielgerichtete Ansatz ist hier die Integration einer physikalisch realistischen Vorhersage um Ausfuhrungsfehler zu verhindern. Erreicht wird das durch die Anpassung oder Veranderung des aktuellen Roboterplans und seiner Aktionen. Die Arbeitshypothese lautet deshalb, dass solch ein integriertes Vorhersagesystem Plane effizienter ausfuhrt als ein System ohne dieses.
Im Verlauf dieser Arbeit wurden eine probabilistische Vorhersagemethode, "Functional Imagination", und eine Systemarchitektur entwickelt. Fur die praktische Evaluation wurde der PR2 Roboter und Gazebo benutzt. Die Vorhersagemethode wurde in ein Blackboard-basiertes Roboter-Kontroll-System integriert. Experimente verifizierten die Simulation als genau und sehr nahe an der Realitat. Die Ergebnisse von drei experimentellen Szenarien in einer Restaurantdomane unterstutzen die Hypothese, dass die physikalische Vorhersage in Kombination mit Planungs- und Ausfuhrungs-Komponenten einem System ohne diese Vorhersage uberlegen ist, hinsichtlich Robustheit und Effizienz.
Am Ende werden die Ergebnisse, Einschrankungen und mogliche zukunftige Forschungsfelder aufgezeigt und ein kurzer Ausblick im Simulationskontext gegeben, wie verteiltes Rechnen und Cloud-Computing das Forschungsfeld der Robotik in Zukunft beeinflussen werden.
5 citations
Book•
21 Mar 2018
TL;DR: A novel approach is proposed to understand, to represent and to execute object manipulation tasks learned from observation by combining methods of data analysis, graphical modeling and artificial intelligence to enable robots to reason about how to solve tasks in dynamic environments and to adapt to unseen situations.
Abstract: Equipping robots with complex capabilities still requires a great amount of effort. In this work, a novel approach is proposed to understand, to represent and to execute object manipulation tasks learned from observation by combining methods of data analysis, graphical modeling and artificial intelligence. Employing this approach enables robots to reason about how to solve tasks in dynamic environments and to adapt to unseen situations. Umfang: X, 236 S. Preis: €47.00 | £43.00 | $83.00
3 citations
Cites methods from "Planning Domain + Execution Semanti..."
...An approach focusing on execution monitoring based on planning is presented by (Konecny et al., 2014)....
[...]
08 Apr 2015
TL;DR: An approach for a robot to conceptualize plan-based robot activity experiences as activity schemata - enriched abstract task knowledge - as well as to exploit them to make plans in similar situations is presented.
Abstract: Learning task knowledge from robot activity experiences has been recognized as an effective approach to improve robot task planning performance. Cognitive capabilities are required to enable a robot to learn new activities from its human partners as well as to refine and improve already learned skills. This paper presents an approach for a robot to conceptualize plan-based robot activity experiences as activity schemata - enriched abstract task knowledge - as well as to exploit them to make plans in similar situations. The experiences are episodic descriptions of plan-based robot activities including environment perceptions, sequences of applied actions and achieved tasks. In this work, the robot activity experiences are obtained through human-robot interaction. The adopted conceptualization approach constructs an activity schema through deductive generalization, abstraction and feature extraction. A high-level task planner was developed to find a solution for a similar task by following an activity schema. The paper proposes a formalization for experience-based planning domains. The proposed learning and planning approach is illustrated in a restaurant environment where a service robot learns how to carry out complex tasks.
3 citations
Cites methods from "Planning Domain + Execution Semanti..."
...During task execution, the Experience extractor extracts a subset of fluents as well as a sequence of applied actions from Working Memory and records them as a plan-based robot activity experience into the Experience Memory [8]....
[...]
...Execution Manager receives the plan generated by the planner and dispatches the planed actions to the robot platform as well as records success or failure information into the Working Memory [11]....
[...]
...ConceptualizerExperience Memory Concept Memory Experience Extractor PlannerWorking Memory Execution Manager + Robot CapabilitiesUser Interface Sensor Data/ Action Results World Information/ Instructions Episode ID Experience Experience Teach Task Instruction/ Episode ID Activity Schema Learned Concepts User Instruction/ Occurrences Plan Occurrences Fig....
[...]
Proceedings Article•
01 Jan 2014TL;DR: This paper proposes an Execution Knowledge that encodes the connection between planning models and the actual actions and observations for a given physical system and presents an execution monitoring framework that captures the expectations about physical plan execution.
Abstract: Despite the progress made in planning androbotics, autonomous plan execution on a robot remainschallenging. One of the problems is that (classical) plannersuse abstract models which are disconnected from the sensorand actuation information available during execution. Thisconnection is typically created in a non-systematic way by somesystem-specific execution software. In this paper we proposeto explicitly represent Execution Knowledge that encodes theconnection between planning models and the actual actionsand observations for a given physical system. We present anexecution monitoring framework in which Execution Knowl-edge captures the expectations about physical plan execution.A violation of these expectations indicates an execution failure.
2 citations
Cites background from "Planning Domain + Execution Semanti..."
...Which temporal constraints are suitable to capture (PK) relations between preconditions, actions and effects is discussed in [5]....
[...]
..., cY (lower) instead of ψ = (cY (lower), [5, 10]) for the rest of this paper....
[...]
...ψ = (cY (lower), [5, 10]) specifies that the controller has stopped the lowering movement at time 10....
[...]
TL;DR: A way to build on a general approach for hybrid planning and combine it with hierarchical planning and the decomposition technique from hierarchical planning serves as a powerful heuristic to reduce the huge hybrid search space.
Abstract: will be holding the object. This is the kind of knowledge that classical task planners employ. But the robot should also make sure that the coffee is served hot. This can be encoded with temporal information. Furthermore, resources are important as the robot has to move three objects but the PR2 has only two arms. In addition to that it may even consider information from other sources like estimates of the duration of how long it takes to drive to a given pose, which can be provided by the robot’s path planner. To act robustly in such environments the robot should consider all of these forms of knowledge. Since many of those requirements are in mutual dependency, this should be done in an integrated way. However, any approach to such hybrid planning faces a huge search space for constructing the plan. Therefore, the thesis presents a way to build on a general approach for hybrid planning and combine it with hierarchical planning. This way, the decomposition technique from hierarchical planning serves as a powerful heuristic to reduce the huge hybrid search space. The remainder of the paper first describes the use of a hierarchical planner in a robot control architecture and how robustness can be improved by a closer coupling of planning and plan execution. Afterwards the idea of hierarchical planning and hybrid reasoning are combined in a new hybrid planner (Sect. 3) and Sect. 4 explains how this planner can directly be used for plan execution as well. The paper closes with a conclusion and outlook on possible future work.
2 citations
References
More filters
Proceedings Article•
01 Jan 2009TL;DR: This paper discusses how ROS relates to existing robot software frameworks, and briefly overview some of the available application software which uses ROS.
Abstract: This paper gives an overview of ROS, an opensource robot operating system. ROS is not an operating system in the traditional sense of process management and scheduling; rather, it provides a structured communications layer above the host operating systems of a heterogenous compute cluster. In this paper, we discuss how ROS relates to existing robot software frameworks, and briefly overview some of the available application software which uses ROS.
8,387 citations
"Planning Domain + Execution Semanti..." refers background in this paper
...All modules communicate through the ROS (Quigley et al. 2009) middleware....
[...]
...A common way to execute and monitor a plan in ROS is by using a Finate State Machine architecture (Bohren et al. 2011)....
[...]
TL;DR: The procedure was originally programmed in FORTRAN for the Control Data 160 desk-size computer and was limited to te t ra t ion because subroutine recursiveness in CONTROL Data 160 FORTRan has been held down to four levels in the interests of economy.
Abstract: procedure ari thmetic (a, b, c, op); in t eger a, b, c, op; ¢ o n l m e n t This procedure will perform different order ar i thmetic operations with b and c, put t ing the result in a. The order of the operation is given by op. For op = 1 addit ion is performed. For op = 2 multiplicaLion, repeated addition, is done. Beyond these the operations are non-commutat ive. For op = 3 exponentiat ion, repeated multiplication, is done, raising b to the power c. Beyond these the question of grouping is important . The innermost implied parentheses are at the right. The hyper-exponent is always c. For op = 4 te t ra t ion, repeated exponentiat ion, is done. For op = 5, 6, 7, etc., the procedure performs pentat ion, hexation, heptat ion, etc., respectively. The routine was originally programmed in FORTRAN for the Control Data 160 desk-size computer. The original program was limited to te t ra t ion because subroutine recursiveness in Control Data 160 FORTRAN has been held down to four levels in the interests of economy. The input parameter , b, c, and op, must be positive integers, not zero; b e g i n own i n t e g e r d, e, f, drop; i f o p = 1 t h e n b e g i n a := h-4c; go t o l e n d i f o p = 2 t h e n d := 0; else d := 1; e := c; drop := op 1; for f := I s t e p 1 u n t i l e do b e g i n ari thmetic (a, b, d, drop);
3,848 citations
"Planning Domain + Execution Semanti..." refers background in this paper
...These bounds are updated as a result of temporal constraint reasoning, an operation which can be performed in low-order polynomial time (Dechter, Meiri, and Pearl 1991; Floyd 1962)....
[...]
01 Jan 2005
TL;DR: A formalism for reasoning about actions that is based on a temporal logic allows a much wider range of actions to be described than with previous approaches such as the situation calculus and a framework for planning in a dynamic world with external events and multiple agents is suggested.
Abstract: A formalism for reasoning about actions is proposed that is based on a temporal logic. It allows a much wider range of actions to be described than with previous approaches such as the situation calculus. This formalism is then used to characterize the different types of events, processes, actions, and properties that can be described in simple English sentences. In addressing this problem, we consider actions that involve non-activity as well as actions that can only be defined in terms of the beliefs and intentions of the actors. Finally, a framework for planning in a dynamic world with external events and multiple agents is suggested.
2,631 citations
"Planning Domain + Execution Semanti..." refers methods in this paper
...We model temporal relations between fluents as constraints in Allen’s Interval Algebra (IA) (Allen 1984)....
[...]
TL;DR: In this article, a formalism for reasoning about actions is proposed that is based on a temporal logic, which allows a much wider range of actions to be described than with previous approaches such as the situation calculus.
Abstract: A formalism for reasoning about actions is proposed that is based on a temporal logic. It allows a much wider range of actions to be described than with previous approaches such as the situation calculus. This formalism is then used to characterize the different types of events, processes, actions, and properties that can be described in simple English sentences. In addressing this problem, we consider actions that involve non-activity as well as actions that can only be defined in terms of the beliefs and intentions of the actors. Finally, a framework for planning in a dynamic world with external events and multiple agents is suggested.
2,439 citations
TL;DR: It is shown that the STP, which subsumes the major part of Vilain and Kautz's point algebra, can be solved in polynomial time and the applicability of path consistency algorithms as preprocessing of temporal problems is studied, to demonstrate their termination and bound their complexities.
Abstract: This paper extends network-based methods of constraint satisfaction to include continuous variables, thus providing a framework for processing temporal constraints. In this framework, called temporal constraint satisfaction problem (TCSP), variables represent time points and temporal information is represented by a set of unary and binary constraints, each specifying a set of permitted intervals. The unique feature of this framework lies in permitting the processing of metric information, namely, assessments of time differences between events. We present algorithms for performing the following reasoning tasks: finding all feasible times that a given event can occur, finding all possible relationships between two given events, and generating one or more scenarios consistent with the information provided. We distinguish between simple temporal problems (STPs) and general temporal problems, the former admitting at most one interval constraint on any pair of time points. We show that the STP, which subsumes the major part of Vilain and Kautz's point algebra, can be solved in polynomial time. For general TCSPs, we present a decomposition scheme that performs the three reasoning tasks considered, and introduce a variety of techniques for improving its efficiency. We also study the applicability of path consistency algorithms as preprocessing of temporal problems, demonstrate their termination and bound their complexities.
1,989 citations
"Planning Domain + Execution Semanti..." refers background in this paper
...The STP maintains a time bound [se, sl] for the start of ψ consisting of earliest possible start time se and latest possible start time sl....
[...]
...A STP maintains a lower and an upper bound for each time point (start or finish time of a fluent)....
[...]
...Such a network can be transformed into a Simple Temporal Problem — STP (Dechter, Meiri, and Pearl 1991)....
[...]
...These bounds are updated as a result of temporal constraint reasoning, an operation which can be performed in low-order polynomial time (Dechter, Meiri, and Pearl 1991; Floyd 1962)....
[...]