A general overview of verification and autonomous systems and the challenges that autonomous systems pose for verification is presented. The new IEEE Guide for Verification of Autonomous Systems is used as an example of new supports becoming available and examples from the Guide are presented to illustrate the ways in which these supports help address the challenges of verification of maritime autonomous systems. Additional examples of guidance specific to maritime autonomous system verification are included.

A "path-based sensor" produces a single observation along a continuous path. For example, a boolean path-based sensor returns a single "1" if an event of interest is detected at any point along the path and a "0" otherwise. Notably, a "1" provides no direct information about where along the path the event(s) may have occurred. Previous work has demonstrated that observations from multiple path-based sensors can be fused to create a Bayesian belief map over the spatial locations of the underlying event or phenomenon. Moreover, path planning can employ Shannon information theory to accelerate the rate of convergence of the belief map. In this paper, we present a new method to update the belief map based on a path-based sensor observation, and then plan paths to increase information gain. In contrast to prior work that approximates the posterior by averaging over the alternative event histories, we introduce a Bayesian Network (BN) formulation that models the probabilistic relationships between the latent variables and path-based sensor measurements, enabling a more principled Bayesian belief update. We consider static hazard detection in a communication-denied environment as a representative problem setting. The event of a robot returning from its path corresponds to a path-based hazard sensor reading of "0" (hazard not detected), while a robot failing to return corresponds to a reading of "1" (hazard detected). We consider false positives and false negatives. We find that the new method leads to quicker convergence of the belief map than prior work in both single- and multi-robot cases.

With the increasing number of autonomous platforms in everyday life, forming coordinated teams of agents becomes vital. To solve this, we propose BERTeam, an algorithm inspired by Natural Language Processing. BERTeam trains a transformer-based deep neural network to select from a population of agents. It can integrate with coevolutionary deep reinforcement learning, which evolves a diverse set of players to choose from. We evaluate BERTeam in Marine Capture-The-Flag, and find it learns non-trivial team compositions that outperform unknown opponents. In this setting, we find that BERTeam outperforms MCAA, another team selection algorithm.

Station-Keeping short-duration high-altitude balloons (HABs) in a region of interest is a challenging path-planning problem due to partially observable, complex, and dynamic wind flows. Deep reinforcement learning is a popular strategy for solving the station-keeping problem. A custom simulation environment was developed to train and evaluate Deep Q-Learning (DQN) for short-duration HAB agents in the simulation. To train the agents on realistic winds, synthetic wind forecasts were generated from aggregated historical radiosonde data to apply horizontal kinematics to simulated agents. The synthetic forecasts were closely correlated with ECWMF ERA5 Reanalysis forecasts, providing a realistic simulated wind field and seasonal and altitudinal variances between the wind models. DQN HAB agents were then trained and evaluated across different seasonal months. To highlight differences and trends in months with vastly different wind fields, a Forecast Score algorithm was introduced to independently classify forecasts based on wind diversity, and trends between station-keeping success and the Forecast Score were evaluated across all seasons.

This paper considers the problem of controlling a team of camera-equipped robots to ensure a mobile human target is observed by at least one robot at all times. We focus on determining team trajectories over a fixed time horizon such that the human target is in at least one robot's field of view even while the line of sight to the target can be obstructed by obstacles. Our approach to solving this problem lies in using a particle-based Belief Propagation method to estimate the most likely poses for each robot over time such that the visual tracking requirement is satisfied. Our approach is validated in simulation studies and an experimental trial where a human target must be tracked by a pair of miniature autonomous robot blimps.

In this paper, we present a distributed information-gathering algorithm for multi-robot systems that use multiple path-based sensors to infer the locations of hazards within the environment. Path-based sensors output binary observations, reporting whether or not an event (like robot destruction) has occurred somewhere along a path, but without the ability to discern where along a path an event has occurred. Prior work has shown that path-based sensors can be used for search and rescue in hazardous communication-denied environments—sending robots into the environment one-at-a-time. We extend this idea to enable multiple robots to search the environment simultaneously. The search space contains targets (human survivors) amidst hazards that can destroy robots (triggering a path-based hazard sensor). We consider a case where communication from the unknown field is prohibited due to communication loss, jamming, or stealth. The search effort is distributed among multiple robots using an entropy-weighted Voronoi partitioning of the environment, such that during each search round all regions have approximately equal information entropy. In each round, every robot is assigned a region in which its search path is calculated. Numerical Monte Carlo simulations are used to compare this idea to other ways of using path-based sensors on multiple robots. The experiments show that dividing search effort using entropy-weighted Voronoi partitioning outperforms the other methods in terms of the information gathered and computational cost.

This study addresses the challenge of accurate wildfire behavior predictions by developing comprehensive forest attribute databases. Diameter at Breast Height (DBH) values of trees are important to predict fire behavior across forest stands. Diameter at Breast Height (DBH) metrics for are difficult to obtain at a large spatial scale. This paper proposes a multi-agent based methodology for developing a trend function relating area density and tree height to DBH values. In the multi-agent system, an effective tree selection algorithm is essential in developing an accurate trend function while minimizing the number of trees that need to be measured. Gaussian regression functions as a decision-making algorithm by motivating the agent to sample trees that minimize the uncertainty of the developing trend function. Comparing tree selection value algorithms, we find Gaussian regression as a UAV decision making algorithm excels in selecting trees to measure that develop a trend function yielding the most accurate DBH values compared to a threshold-based approach and random tree selection. Gaussian regression is also utilized for the computation of the final trend function due to its effectiveness in capturing non-linear relationships between variables. This research showcases the effectiveness of Gaussian regression in dynamic information foraging problems as well as the effectiveness of utilizing multi-agent systems for DBH collection and predictions.

Unmanned aerial vehicles have an acoustic signature generated by their propulsion systems. Leveraging this passive broadband signal for object-relative range sensing could improve multisensory perception for vehicles at a minimal cost. We present an analytical method to obtain optimal distance estimates for improved accuracy of passive acoustics-based range sensing and demonstrate this approach on data collected with a mounted quad-rotor vehicle in a noisy environment. Additional analysis of uncertainty parameters in the optimal estimation method investigates possible metrics to use for surface detection.

Intermittent actuation for autonomous systems inspired by biological locomotion can be leveraged for swarming platforms tasked with underwater localization and tracking. This paper presents the design of a control strategy for a neutrally buoyant vehicle tracking a moving target using intermittent actuation. Tracking a time-varying reference with intermittent actuation is presented using state-feedback control for the decoupled 1-DOF cases of vertical motion and rotational motion about the vertical axis. Experimental results from tests on the miniature autonomous blimp platform demonstrate the control strategy on a neutrally buoyant vehicle.

Prior work on path-based sensors has assumed that each agent's path is determined before the agent departs and cannot be changed mid-trip. We consider how an agent might adjust its path in response to new information that it gathers en route. Mid-trip path adjustment is non-trivial because it can increase the number of locations at which a missing agent may have been destroyed (from an external observer's point-of-view). We solve this issue by employing backtracking as a particular form of mid-trip path adjustment that avoids the issue of additional potential destruction locations.

In this paper we propose a new algorithm for robots searching a hazardous, communications-denied area to gather information using a robot fleet that has a limited number of agents. The centralized algorithm uses robot survival along search paths as a sensor event for a distributed sensor network. As agents are lost to hazards, the search behavior adjusts to prioritize agent longevity in order to maximize information gain. In the past, related work solving this problem has assumed an infinite number of agents. In contrast, we assume that the number of agents is finite. We use Bayesian inference to update target and hazard belief maps of an area using data from the probability of survival of prior agents' paths as well as sensor readings from the agents along those paths. Using those belief maps, the algorithm can construct paths that maximize information gain, in expectation, while taking into account the predicted decrease in future information collected when losing an agent. This behavior increases the likelihood that agents survive longer, allowing them to collect more data. Using simulations with various fleet sizes and probabilities for hazards disabling agents, we compare our algorithm to work that does not account for attrition. The results show an increase in the longevity of the fleet when hazards are more effective at disabling agents. In nearly all cases, this contributes to an increased rate in information gain when the fleet size is small. Small sized fleets, in our case 10 or less agents, do not meet a threshold of collected information necessary to direct agents away from hazards. Large fleets, over 200 agents in our scenario, collect most of the information before Our algorithm causes a noticeable change in agent behavior (as compared to existing techniques). We find that the proposed method provides the greatest advantage for mid-sized fleets, between 20 and 100 agents, and when hazards have an increased probability of immobilizing agents.

Distributed state estimation is an important tool for coordinated team decision-making and typically involves sharing information between robots in order to outperform individual state estimation. The shared information typically takes the form of relative measurements which allow team members to act as virtual sensors for other robots where the virtual sensor uncertainty is corrupted also by the team member's state uncertainty. However, incorporating relative measurements commonly depends on a pointwise product operation in most distributed estimation techniques which is well defined for continuously-valued distributions but ill-defined for particle-based distributions. We propose a drop-in replacement for the pointwise product based on using the generalized Holder's inequality to upper-bound the product over a series of grid cell sets that discretize the state space. This upper-bound is well defined for particle-based distributions and allows for tighter approximations by decreasing the volume of the sets. We leverage the approach to realize two distributed estimation strategies that use a pointwise product, the Kullback-Leibler Average and Belief Propagation and use these methods in simulations and experiments with a pair of miniature autonomous blimps. We found that after distributed smoothing, we were able to achieve an average improvement of similar to 25.1% and similar to 35.2% in the position tracking error, indicating our distributed smoothing strategy is able to improve the tracking performance of our initial filtering estimates.

In prior research, we analyzed the backwards swimming motion of mosquito larvae, and created a parametrized approximation in a Computational Fluid Dynamics simulation. Since the parameterized swimming motion is replicated from observed larvae, it is not necessarily the most efficient locomotion. In this project, we further optimize this swimming locomotion for the simulated platform, using Reinforcement Learning to guide local parameter updates. Since the majority of the computation cost arises from the Computational Fluid Dynamics model, we additionally train a deep neural network to replicate the forces acting on the swimmer model. We find that this method is effective at performing local search to improve the parameterized swimming locomotion.

When researching robot swarms, many studies observe complex group behavior emerging from the individual agents' simple local actions. However, the task of learning an individual policy to produce a desired group behavior remains a challenging problem. We present a method of training distributed robotic swarm algorithms to produce emergent behavior. Inspired by the biological evolution of emergent behavior in animals, we use an evolutionary algorithm to train a population of individual behaviors to produce a desired group behavior. We perform experiments using simulations of the Georgia Tech Miniature Autonomous Blimps (GT-MABs) aerial robotics platforms conducted in the CoppeliaSim simulator. Additionally, we test on simulations of Anki Vector robots to display our algorithm's effectiveness on various modes of actuation. We evaluate our algorithm on various tasks where a somewhat complex group behavior is required for success. These tasks include an Area Coverage task and a Wall Climb task. We compare behaviors evolved using our algorithm against designed policies, which we create in order to exhibit the emergent behaviors we desire.

Some human-machine systems are designed so that machines (robots) gather and deliver data to remotely located operators (humans) through an interface in order to aid them in classification. The performance of a human as a (binary) classifier-in-the-loop is characterized by probabilities of correctly classifying objects of type $T$ and $F$. These two probabilities depend on the dwell time, $d$, spent collecting information at a point of interest (POI or interchangeably, target). The information gain associated with collecting information at a target is then a function of dwell time $d$ and discounted by the revisit time, $R$, i.e., the duration between consecutive revisits to the same target. The objective of the problem of routing for classification is to optimally route the vehicles and determine the optimal dwell time at each target so as to maximize the total discounted information gain while visiting every target at least once. In this paper, we make a simplifying assumption that the information gain is discounted exponentially by the revisit time; this assumption enables one to decouple the problem of routing with the problem of determining optimal dwell time at each target for a single vehicle problem. For the multi-vehicle problem, we provide a fast heuristic to obtain the allocation of targets to each vehicle and the corresponding dwell time.

Unmanned air vehicles often produce significant noise from their propulsion systems. Using this broadband signal as "acoustic illumination" for an auxiliary sensing system could make vehicles more robust at a minimal cost. We present an acoustic beamforming-based algorithm that estimates object-relative distance with a small two-microphone array using the generated propulsion system noise of a vehicle. We demonstrate this approach in several closed-loop distance feedback control tests with a mounted quad-rotor vehicle in a noisy environment and show accurate object-relative distance estimates more than 2x further than the baseline channel-based approach. We conclude that this approach is robust to several practical vehicle and noise situations and shows promise for use in more complex operating environments.

In aerial flight, lighter-than-air vehicles are an under-utilized form of flight that trades agility and speed for longevity and safety. In this paper, we introduce the Open-Blimp, a lighter-than-air vehicle capable of achieving autonomous navigation using an onboard camera, altitude sensor, and IMU and has wide utility in supporting mobile robotics research. We demonstrate the platform's utility in a wind-field station-keeping problem in which the blimp leverages its onboard camera to perform position estimation and maintain a desired position. State estimates are provided via fiduciary tags in the environment which are observed using our onboard camera. Experimental results using the Open-Blimp are included.