Enrico Scala

Numeric Kernel for Reasoning about Plans Involving Numeric Fluents

The paper proposes the notion of numeric kernel as a means for reasoning about plans involving numeric state variables, i.e. numeric fluents. A numeric kernel identifies the sufficient and necessary conditions that allow to directly - without any search and any propagation - assess whether a plan is valid in a specific world state. The notion generalizes the propositional kernels defined for the STRIPS language, to support domains involving numeric information as well. A regression method to build such kernels is reported, and its correctness is theoretically proved. To evaluate the numeric kernels contribution, we report two possible repair strategies that can be employed as a direct application of the numeric kernel properties. Results show the promise of the approach both from the computational point of view and in terms of plan quality.

Intelligent Supervision for Robust Plan Execution

The paper addresses the problem of supervising the execution of a plan with durative actions in a just partially known world, where discrepancies between the expected conditions and the ones actually found may arise. The paper advocates a control architecture which exploits additional knowledge to prevent (when possible) action failures by changing the execution modality of actions while these are still in progress. Preliminary experimental results, obtained in a simulated space exploration scenario, are reported.

Landmarks for Numeric Planning Problems

The paper generalises the notion of landmarks for reasoning about planning problems involving propositional and numeric variables. Intuitively, numeric landmarks are regions in the metric space defined by the problem whose crossing is necessary for its resolution. The paper proposes a relaxationbased method for their automated extraction directly from the problem structure, and shows how to exploit them to infer what we call disjunctive and additive hybrid action landmarks. The justification of such a disjunctive representation results from the intertwined propositional and numeric structure of the problem. The paper exercises their use in two novel admissible LP-Based numeric heuristics, and reports experiments on cost-optimal numeric planning problems. Results show the heuristics are more informed and effective than previous work for problems involving a higher number of (sub)goals.

Robust plan execution via reconfiguration and replanning

Acting in real world may be a difficult task for an agent, either software or robotic, because unexpected contingencies may arise at any step of the execution. Previous approaches to robust plan execution consider propositional goals to be achieved and time constraints to be satisfied. However, realistic plans must obey to constraints on continuous/consumable resources, too. To face the complexity in handling these resources, the paper proposes the notion of Multi Modality Action (MMA). The model allows to explicitly express the multiple execution modalities in which a given action can be executed; each execution modality models requirements/consequences on the involved consumable resources when that modality is selected. Relying on the MMA notion, the paper presents how the repair problem can be seen as a problem of reconfiguring actions modalities, and how it can be solved by exploiting a CSP encoding. The MMAs are employed by a new continual planner, FLEX-RR, which, exploiting the synergy from the reconfiguration and a numeric planning mechanism can efficiently repair on the fly the plan keeping it rather stable. An empirical analysis performed on three numeric planning domains, confirms the large benefits of FLEX-RR in terms of competence, efficiency and stability of the repaired plan.

Proactive and reactive reconfiguration for the robust execution of multi modality plans

The paper addresses the problem of executing a plan in a dynamic environment for tasks involving constraints on consumable resources modeled as numeric fluents. In particular, the paper proposes a novel monitoring and adaptation strategy joining reactivity and proactivity in a unified framework. By exploiting the flexibility of a multi modality plan (where each action can be executed in different modalities), reactivity and proactivity are guaranteed by means of a reconfiguration step. The reconfiguration is performed (i) when the plan is no more valid to recovery from the impasse (reactively), or (ii) under the lead of a kernel based strategy to enforce the tolerance to unexpected situations (proactivity). Both mechanisms have been integrated into a continual planning system and experimentally evaluated over three numeric domains, extensions of planning competition domains. Results show that the approach is able to increase the percentage of cases successfully solved while preserving efficiency in most situations.

Numerical Kernels for Monitoring and Repairing Plans Involving Continuous and Consumable Resources

In this paper we introduce a technique for monitoring and repairing a plan dealing with continuous and consumable resources. The mechanism relies on the notion of numerical kernel . Concretely, anumerical kernel establishes the sufficient and necessary conditions for a plan to be valid in a specific state of the system. We employ the mechanism in a continual planning agent and we evaluate experimentally the approach for the Zenotraveldomain. Results show good cpu-time w.r.t. a traditional replanning from scratch.

Robust Execution of Rover Plans via Action Modalities Reconfiguration

Robust execution of exploration mission plans has to deal with limited computational power on-board a planetary rover, and with limited rovers autonomy. Typically, these limitations prevent the rover to synthesize a new mission plan when some unexpected contingency arises. The paper shows that when such deviations refers to anomalies on the consumption of resources, robust execution can be achieved efficiently through an action reconfiguration approach instead of a replanning from scratch. Building up on an extended action model representation, the paper proposes an effective continual planner - ReCon - that, exploiting a general purpose CSP solver, is able to (i) detect violations of mission resource constraints, and (ii) find (if any) a new configuration of actions

Towards Robust Execution of Mission Plans for Planetary Rovers

The paper discusses an on-line control architecture for plan execution whose main aim is to prevent (at least in some cases) the occurrence of action failures. To reach this result, the plan to be executed has been enriched with additional knowledge about the intermediate conditions (i.e., invariant conditions), which must be satisfied during the execution of (durative) actions. This knowledge is used by a Temporal Reasoner to detect anomalous situations that may endanger the safeness of the plan executor. Whenever an anomaly has been detected, the proposed control architecture tries to prevent a failure by changing the execution modality of the action while it is still in progress. Preliminary experimental results, obtained in a simulated space exploration scenario, are reported.

Intelligent Belief State Sampling for Conformant Planning

We propose a new method for conformant planning based on two ideas. First given a small sample of the initial belief state we reduce conformant planning for this sample to a classical planning problem, giving us a candidate solution. Second we exploit regression as a way to compactly represent necessary conditions for such a solution to be valid for the non-deterministic setting. If necessary, we use the resulting formula to extract a counterexample to populate our next sampling. Our experiments show that this approach is competitive on a class of problems that are hard for traditional planners, and also returns generally shorter plans. We are also able to demonstrate unsatisfiability of some problems.

Involving the Human User in the Control Architecture of an Autonomous Agent

The paper presents an architecture for an autonomous robotic agent, which carries on a plan in a partially observable environment. A Supervisor module is in charge of assuring the correct execution of the plan, possibly by inferring alternative recovery plans when unexpected contingencies occur. In the present paper we describe a control strategy where a human user is directly involved in the control loop, and plays the role of advisor by helping the robotic agent both for reducing ambiguity in the robot’s observations, and for selecting the preferred recovery plan.