Luca F. Bertuccelli

Increasing autonomy of UAVs

This article has presented a tightly integrated systems architecture for a decentralized CSAT mission management algorithm and demonstrated successful implementation in actual hardware flight tests. This CSAT architecture allows each UAV to accomplish a combined search and track mission by conceptualizing the searching aspect as a spare time strategy to be executed optimally over a short time horizon when the agents are not actively tracking a vehicle. This presented a balance between the two conflicting search and track modes and allowed the mission to achieve more than simply searching or tracking alone.

Robust UAV Search for Environments with Imprecise Probability Maps

This paper introduces a new framework for UAV search operations and proposes a new approach to calculate the minimum number of looks needed to achieve a given level of confidence of target existence in an uncertain gridded environment. Typical search theory formulations describe the uncertainty in the environment in a probabilistic fashion, by assigning probabilities of target existence to the individual cells of the grid. While assumed to be precisely known in the search theory literature, these probabilities are often the result of prior information and intelligence, and will likely be poorly known. The approach taken in this paper models this imprecise knowledge of the prior probabilities in the individual cells using the Beta distribution and generates search actions that are robust to the uncertainty. Use of the Beta distribution leads to an analytical prediction of the number of looks in a particular cell that would be needed to achieve a specified threshold in the confidence of target existence. The analytical results are demonstrated in both an expected value setting and a framework that takes into account the variance of the posterior distribution. The effectiveness of the proposed framework is demonstrated in several numerical simulations.

Robust planning for coupled cooperative UAV missions

This paper presents a new formulation for the UAV task assignment problem with uncertainty in the environment. The problem is posed as a task assignment with uncertainty in the cost information, and we apply a modified robust technique that allows the operator to tune the level of robustness in the optimization. This formulation is then used to solve the assignment problem for a heterogeneous fleet of vehicles operating in an uncertain environment. The key aspect of this formulation is that it directly addresses the inherent coupling in deciding how to assign vehicles to perform reconnaissance tasks that provide the most benefit to the strike part of the missions. We demonstrate that the robust solution to this coupled problem can be solved as single mixed-integer linear problem. The paper presents and discusses simulations for the proposed formulation, demonstrating significant improvements over previous ones.

Search for dynamic targets with uncertain probability maps

This paper extends a recently developed statistical framework for UAV search with uncertain probability maps to the case of dynamic targets. The probabilities used to encode the information about the environment are typically assumed to be exactly known in the search theory literature, but they are often the result of prior information that is both erroneous and delayed, and will likely be poorly known to mission designers. Our previous work developed a new framework that accounted for the uncertainty in the probability maps for stationary targets, and this paper extends the approach to more realistic dynamic environments. The dynamic case considers probabilistic target motion, creating uncertain probability maps (UPMs) that take into account both poor knowledge of the probabilities and the propagation of their uncertainty through the environment. A key result of this paper is a new algorithm for implementing UPMs in real-time, and it is shown in various simulations that this algorithm leads to more cautious information updates that are less susceptible to false alarms. The paper also provides insights on the impact of the design parameters on the responsiveness of the new algorithm. Several numerical examples are presented to demonstrate the effectiveness of the new framework

Filter-Embedded UAV Task Assignment Algorithms for Dynamic Environments

This paper presents a modified formulation of the classical task assignment problem that has recently been used to coordinate teams of UAVs. The main contribution is a version of the task assignment problem that can be used to tailor the control system to mitigate the eect of noise in the situational awareness on the solution. The net eect will be to limit the rate of change in the reassignment in a well-defined manner. The approach here is to perform reassignment at the rate that information is updated, which enables immediate reaction to any significant changes in the environment. We demonstrate that the modified formulation can be interpreted as a noise rejection algorithm that can be tuned to reduce the eect of variation in the uncertain parameters in the problem. Simulations are presented to demonstrate the eectiveness of this algorithm.

Developing Operator Models for UAV Search Scheduling

With the increased use of Unmanned Aerial Vehicles (UAVs), it is envisioned that UAV operators will become high level mission supervisors, responsible for information management and task planning. In the context of search missions, operators supervising a large number of UAVs can become overwhelmed with the sheer amount of information collected by the UAVs, making it difficult to optimize the information collection or direct their attention to the relevant data. Novel decisionsupport methods that account for realistic operator performance will therefore be required to aid the operators. This paper considers a decision support formulation for sequential search tasks, and discusses a non-preemptive scheduling formulation for a single operator performing a search mission in a time-constrained environment. The formulation is then generalized to include operator performance obtained from previous human-in-the-loop experiments, and presents one of the principal contributions of the paper. The sensitivity of the proposed model is analyzed in the presence of uncertainty to the operator model and search times, and a comparison is made between the expected performance difference between this scheduling system and a greedy scheduling strategy representative of operator planning. The paper concludes with the design of a human-in-the-loop experiment for a scheduling, replanning task for a simulated UAV mission.

A Robust Approach to the UAV Task Assignment Problem

This paper presents a new formulation for the UAV task assignment problem for uncertain dynamic environments. Uncertainty in this time-varying information directly implies that the optimization data, such as target cost and target-UAV distances, will be uncertain. To mitigate the impact of this uncertainty, the new algorithm combines two key approaches that have been developed to handle the changes in the data of the assignment problem. One approach is to design task assignment plans that are robust to the uncertainty in the data, which reduces the sensitivity to errors in the situational awareness (SA), but can be overly conservative for long duration plans. An alternative strategy is to replan as the SA is updated, which results in the best plan given the current information, but can lead to churning if the updates are too rapid. This paper presents an alternative strategy that combines robust planning with the techniques developed to eliminate churning. The resulting robust filter embedded task assignment (RFETA) uses both proactive and reactive techniques to handle the uncertainty in the information and is shown to improve worst-case behavior of the plans, while at the same time ensuring that limited churning behavior is exhibited by the vehicle responding to noisy measurements. Numerous simulations are shown that demonstrate the performance benefits of this new algorithm

Choice modeling of relook tasks for UAV search missions

This paper addresses human decision-making in supervisory control of a team of unmanned vehicles performing search missions. Previous work has proposed the use of a two-alternative choice framework, in which operators declare the presence or absence of a target in an image. It has been suggested that relooking at a target at some later time can help operators improve the accuracy of their decisions but it is not well understood how - or how well - operators handle this relook task with multiple UAVs. This paper makes two novel contributions in developing a choice model for a search task with relooks. First, we extend a previously proposed queueing model of the human operator by developing a retrial queue model that formally includes relooks. Since real models may deviate from some of the theoretical assumptions made in the requeueing literature, we develop a Discrete Event Simulation (DES) that embeds operator models derived from previous experimental data and present new results in the predicted performance of multi-UAV visual search tasks with relook. Our simulation results suggest that while relooks can in fact improve detection accuracy and decrease mean search times per target, the overall fraction found correctly is extremely sensitive to increased relooks.

Estimation of non-stationary Markov Chain transition models

Many decision systems rely on a precisely known Markov Chain model to guarantee optimal performance, and this paper considers the online estimation of unknown, non-stationary Markov Chain transition models with perfect state observation. In using a prior Dirichlet distribution on the uncertain rows, we derive a mean-variance equivalent of the maximum a posteriori (MAP) estimator. This recursive mean-variance estimator extends previous methods that recompute the moments at each time step using observed transition counts. It is shown that this mean-variance estimator responds slowly to changes in transition models (especially switching models) and a modification that uses ideas of pseudonoise addition from classical filtering is used to speed up the response of the estimator. This new, discounted mean-variance estimator has the intuitive interpretation of fading previous observations and provides a link to fading techniques used in Hidden Markov Model estimation. Our new estimation techniques is both faster and has reduced error than alternative estimation techniques, such as finite memory estimators.

Robust Adaptive Markov Decision Processes: Planning with Model Uncertainty

The ability of autonomous systems to make complex decisions is becoming an increasingly commonplace requirement for many cooperative control operations, including the management of teams of robots such as unmanned aerial vehicles (UAV). Central to this research is the requirement to optimize the vehicle decisions, such as route planning and allocation of team resources, while operating in a dynamic and uncertain environment. Even with the advent of increasingly sophisticated vehicle sensors that can improve the information about the surroundings, uncertainty remains a ubiquitous feature of UAV applications and a key issue in UAV research.