Below are illustrations of various simulations of UASs. The developed simulation is designed to test various capabilities and operations of UASs. The goal of the work is to develop computationally feasible modeling and simulations of various drones working in tangent to meet a goal and objective. The models utilized open-source tool kits to enable the study of multibody dynamics and building models based on derived equations of motion. All the models shown have the basic capability to autonomously navigate toward waypoints without an operator. The more advanced UASs have the capability to also avoid and work with other UASs. Thus, allowing swarming capability based on dynamic co-fields, cooperative motor schema behavior. The movement of the drones are generated by following gradient associated with combined field of all known entities. Lastly, we wanted to demo the ability to simulate many UASs all doing different or cooperative behaviors in one simulation; thus, some models will have many drones. This allows for rapid modeling and adjustments during operations with low computational cost.
Interpretability and explainability are driving factors for utilization and implementation of new technologies. When working with ML and AI, this can be challenging but is at the heart of many researchers. Some approaches have been developed with the aim of better understanding how observations and actions taken by these models occur. The illustration below is an example of one such technique to better understand how observations affect a specific action taken by the AI model. At each time step we highlight the observation that contributes more toward a specific action taken by the AI model. This helps us understand and visually illustrate what observation contributes to the action taken by the AI at any time step.
Density Functional Theory (DFT) is a widely used computational quantum mechanical modeling method that helps investigate the electronic structure and properties of matter. However, it is computationally costly when doing design-based studies for many systems. The developed hybrid physics and ML approach reduced the computational expense in 3 ways:
In the resilience and population effort, we are developing and implementing statistical physics modeling and machine learning approaches to better predict evolving risks to local populations and target relief operations. The datasets used are all open source. It is a very simple model but shows very interesting results. The example shows work where we wanted to understand population, resilience, and model/provide relevant data to first responders. The example shows layers for heat map of population and energy, power systems, cellular towers, and electrical lines. This information can provide first responders with regions of interest.
The demonstration below is a simple analysis of hashtag data. However, the analysis reveals very interesting results and trends. As can be seen, the most talked about topics are expressed in the word cloud, the bigger the word the more prominent it is in the overall dataset. However, this model takes another step to better understand how topics can emerge out of the various categories. For example, for all the conversations around a specific hashtag what other topics arise? The demonstration looks at the hashtag “#AI” and reveals all the other topics that this subgroup cares about and talks around.
The illustration below is a model developed to better understand and experiment with some patterns of underwater acoustics. The model has a source of pings that propagate through the material (water). In the simulation an acoustic object is embedded to reflect the waves propagated. This can be seen in the second figure. As the waves propagate, the instantaneous measurements around the object is recorded, as seen by the sub plots.