Deep auto-encoders (AEs) are widely employed deep learning methods in the field of anomaly detection across diverse domains (e.g., cybersecurity analysts managing large volumes of alerts, or medical practitioners monitoring irregular patient signals). In such contexts, practitioners often face challenges of scale and limited processing resources. To cope, strategies such as false positive reduction, human-in-the-loop review, and alert prioritization are commonly adopted. This paper explores the integration of uncertainty quantification (UQ) methods into alert prioritization for anomaly detection using ensembles of AEs. UQ models highlight doubtful classification decisions, enabling analysts to address the most certain alerts first, since higher certainty typically correlates with greater accuracy. Our study reveals a nuanced issue where applying UQ to ensembles of AEs can produce skewed distributions of large reconstruction errors (errors exceeding a pre-defined threshold), which may falsely suggest high uncertainty when standard deviation is used as the metric. Conventionally, a high standard deviation indicates high uncertainty. However, contrary to intuition, large reconstruction errors often reflect AE is strongly confident that an input is anomalous—not uncertainty about it. Moreover, ensembles of AEs generate reconstruction errors with varying ranges, complicating interpretation. To address this, we propose an extension that calibrates the standard deviation distribution of uncertainties, mitigating erroneous prioritization. Evaluation on 10 benchmark datasets demonstrates that our calibration approach improves the effectiveness of UQ methods in prioritizing alerts, while maintaining favorable trade-offs across other key performance metrics.

Inappropriate prescribing of antimicrobials is a major clinical and health concern, as well as a financial burden, in hospitals worldwide. In this paper, we describe a deployed automated antimicrobial prescription surveillance system that has been assisting hospital pharmacists in identifying and reporting inappropriate antimicrobial prescriptions. One of the key characteristics of this system is its ability to learn new rules for detecting inappropriate prescriptions based on previous false alerts. The supervised learning algorithm combines instancebased learning and rule induction techniques. It exploits temporal abstraction to extract a meaningful time interval representation from raw clinical data, and applies nearest neighbor classification with a distance function on both temporal and non-temporal parameters. The learning capability is valuable both in configuring the system for initial deployment and improving its long term use. We give an overview of the application, point to lessons learned so far and provide insight into the machine learning capability.

Possessing a sufficient level of situation awareness is essential for effective decision making in dynamic environments. In video games, this includes being aware to some extent of the intentions of the opponents. Such high-level awareness hinges upon inferences over the lower-level situation awareness provided by the game state. Traditional plan recognizers are completely passive processes that leave all the initiative to the observed agent. In a situation where the opponent's intentions are unclear, the observer is forced to wait until further observations of the opponent's actions are made to disambiguate the pending goal hypotheses. With the plan recognizer we propose, in contrast, the observer would take the initiative and provoke the opponent, with the expectation that his reaction will give cues as to what his true intentions actually are.

Agents in highly dynamic adversarial domains, such as RTS games, must continually make time-critical decisions to adapt their behaviour to the changing environment. In such a context, the planning agent must consider his opponent's actions as uncontrollable, or at best influenceable. In general nondeterministic domains where there is no clear turn-taking protocol, most heuristic search methods to date do not explicitly reason about the opponent's actions when guiding the state space exploration towards goal or high-reward states. In contrast, we are investigating a domain-independent heuristic planning approach which reasons about the dynamics and uncontrollability of the opponent's behaviours in order to provide better guidance to the search process of the planner. Our planner takes as input the opponent's behaviours recognized by a plan recognition module and uses them to identify opponent's actions that lead to low-utility projected states. We believe such explicit heuristic reasoning about the potential behaviours of the opponent is crucial when planning in adversarial domains, yet is missing in today's planning approaches.

Omnidirectional mobile robots, i.e., robots that can move in any direction without changing their orientation, offer better manoeuvrability in natural environments. Modeling the kinematics of such robots is a challenging problem and different approaches have been investigated. One of the best approaches for a nonholonomic robot is to model the robot's velocity state as the motion around its instantaneous center of rotation (ICR). In this paper, we present a motion planner designed to compute efficient trajectories for such a robot in an environment with obstacles. The action space is modeled in terms of changes of the ICR and the motion around it. Our motion planner is based on a Rapidly-Exploring Random Trees (RRT) algorithm to sample the action space and find a feasible trajectory from an initial configuration to a goal configuration. To generate fluid paths, we introduce an adaptive sampling technique taking into account constraints related to the ICR-based action space.

Forces involved in modern conflicts may be exposed to a variety of threats, including coordinated raids of advanced ballistic and cruise missiles. To respond to these, a defending force will rely on a set of combat resources. Determining an efficient allocation and coordinated use of these resources, particularly in the case of multiple simultaneous attacks, is a very complex decision-making process in which a huge amount of data must be dealt with under uncertainty and time pressure. This article presents CORALS (COmbat Resource ALlocation Support), a real-time planner developed to support the command team of a naval force defending against multiple simultaneous threats. In response to such multiple threats, CORALS uses a local planner to generate a set of local plans, one for each threat considered apart, and then combines and coordinates them into a single optimized, conflict-free global plan. The coordination is performed through an iterative process of plan merging and conflict detection and resolution, which acts as a plan repair mechanism. Such an incremental plan repair approach also allows adapting previously generated plans to account for dynamic changes in the tactical situation.

In this paper we describe the Automatic Task Demonstration Generator (ATDG), a system implemented into a software prototype for teaching the operation of a robot manipulator deployed on the International Space Station (ISS). The ATDG combines the use of path planning and camera planning to take into account the complexity of the manipulator, the limited direct view of the ISS exterior, and the unpredictability of lighting conditions in the workspace. The path-planning algorithm not only avoids obstacles in the workspace as is normal for a path-planner, but in addition takes into account the position of corridors for safe operations and the placement of cameras on the ISS. The camera planner is then invoked to find the right arrangement of cameras to follow the manipulator on its trajectory. This allows the on-the-fly production of useful and pedagogical task demonstrations to help the student carry out tasks involving the manipulation of the robot on the ISS. Even if the system has been developed for robotic manipulations, it could be used for any application involving the filming of unpredictable complex scenes.

Behavior-based architectures use behaviors as building blocks for decision-making and action execution processes. Behaviors are distributed and evaluated in parallel for the control of the robot, taking real-time inputs from sensory data and sending real-time commands to effectors. No centralized components exist in these architectures, each module carrying out its own strategy independently, making an overall behavior emerge from the interaction between the concurrently executed modules and the environment. In this paper, we discuss the use of a reactive hierarchical task network (HTN) planner in a behavior-based robot architecture. The planner in this architecture is not a central component on which everything else relies on, but acts as one of the motivational modules recommending tasks to be executed and influencing the selection and configuration of behaviors. The planning module allows the behavior-based architecture to deal with tasks with priorities, flexible time constraints and on-line planning using a simple but very effective reactive planning strategy. We demonstrate our approach in the context of making a robot attend a conference.

Probabilistic roadmaps (PRM) have been demonstrated to be very promising for planning paths for robots with high degrees of freedom in complex 3D workspaces. In this paper we describe a PRM path-planning method presenting three novel features that are useful in various real-world applications. First, it handles zones in the robot workspace with different degrees of desirability. Given the random quality of paths that are calculated by traditional PRM approaches, this provides a mean to specify a sampling strategy that controls the search process to generate better paths by simply annotating regions in the free workspace with degrees of desirability. Second, our approach can efficiently re-compute paths in dynamic environments where obstacles and zones can change shape or move concurrently with the robot. Third, it can incrementally improve the quality of a generated path, so that a suboptimal solution is available when required for immediate action, but get improved as more planning time is affordable

A decision process in which rewards depend on history rather than merely on the current state is called a decision process with non-Markovian rewards (NMRDP). In decision-theoretic planning, where many desirable behaviours are more naturally expressed as properties of execution sequences rather than as properties of states, NMRDPs form a more natural model than the commonly adopted fully Markovian decision process (MDP) model. While the more tractable solution methods developed for MDPs do not directly apply in the presence of non-Markovian rewards, a number of solution methods for NMRDPs have been proposed in the literature. These all exploit a compact specification of the non-Markovian reward function in temporal logic, to automatically translate the NMRDP into an equivalent MDP which is solved using efficient MDP solution methods. This paper presents NMRDPP (Non-Markovian Reward Decision Process Planner), a software platform for the development and experimentation of methods for decision-theoretic planning with non-Markovian rewards. The current version of NMRDPP implements, under a single interface, a family of methods based on existing as well as new approaches which we describe in detail. These include dynamic programming, heuristic search, and structured methods. Using NMRDPP, we compare the methods and identify certain problem features that affect their performance. NMRDPP's treatment of non-Markovian rewards is inspired by the treatment of domain-specific search control knowledge in the TLPlan planner, which it incorporates as a special case. In the First International Probabilistic Planning Competition, NMRDPP was able to compete and perform well in both the domain-independent and hand-coded tracks, using search control knowledge in the latter.

This paper describes the implementation of an intelligent tutoring system for image selection in the context of updating a forest map. It is part of the "Systeme Intelligent de Traitement d'Image" developed at the Universiti de Sherbrooke.

Behavior-based robot control systems have shown remarkable success for controlling robots evolving in real world environments. However, they can fail in different manners due to their distributed control and their local decision making. In this case, monitoring can be used to detect failures and help to recover from them. In this work, we present an approach for specifying monitoring knowledge and a method for using this knowledge to detect failures. In particular we show how temporal fuzzy logic can be used to represent monitoring knowledge and then utilized to effectively detect runtime failures. New semantics are introduced to take into consideration uncertainty and noisy information. There are numbers of advantages to our approach including a declarative semantics for the monitoring knowledge and an independence of this knowledge from the implementation details of the control system. Moreover we show how our system can deal effectively with noisy information and sensor readings. Experiments with two real world robots and the simulator are used to illustrate failure examples and the benefits of failure detection and noise elimination.

In this paper, we present a powerful framework for describing, storing, and reasoning about infinite temporal information. This framework is an extension of classical relational databases. It represents infinite temporal information by generalized tuples defined by linear repeating points and constraints on these points. We prove that relations formed from generalized tuples are closed under the operations of relational algebra. A characterization of the expressiveness of generalized relations is given in terms of predicates definable in Presburger arithmetic. Finally, we provide some complexity results.