Unmanned aerial systems (UAS) are increasingly used in precision agriculture to collect crop health related data. UAS can capture data more often and more cost-effectively than sending human scouts into the field. However, in large crop fields, flight time, and hence data collection, is limited by battery life. In a conventional UAS approach, human operators are required to exchange depleted batteries many times, which can be costly and time consuming. In this study, we developed a novel, fully autonomous aerial scouting approach that preserves battery life by sampling sections of a field for sensing and predicting crop health for the whole field. Our approach uses reinforcement learning (RL) and convolutional neural networks (CNN) to accurately and autonomously sample the field. To develop and test the approach, we ran flight simulations on an aerial image dataset collected from an 80-acre corn field. The excess green vegetation Index was used as a proxy for crop health condition. Compared to the conventional UAS scouting approach, the proposed scouting approach sampled 40% of the field, predicted crop health with 89.8% accuracy, reduced labor cost by 4.8× and increased agricultural profits by 1.36×.

Fully autonomous aerial systems (FAAS) fly complex missions guided wholly by software. If users choose software, compute hardware and aircraft well, FAAS can complete missions faster and safer than unmanned aerial systems piloted by humans. On the other hand, poorly managed edge resources slow down missions, waste energy and inflate costs. This paper presents a model-driven approach to manage FAAS. We fly real FAAS missions, profile compute and aircraft resource usage and model expected demands. Naive profiling approaches use traces from previous flights to infer resource usage. However, edge resources can affect where FAAS fly and which data they sense. Usage profiles can diverge greatly across edge management policies. Instead of using traces, we characterize whole flight areas to accurately model resource usage for any flight path. We combine expected resource demands to model mission throughput, i.e., missions completed per fully charged battery. We validated our model by creating FAAS, measuring mission throughput across many system settings. Our FAAS benchmarks, released through our open source FAAS suite SoftwarePilot, execute realistic missions: autonomous photography, search and rescue, and agricultural scouting using well-known software. Our model predicted throughput with 4% error across mission, software and hardware settings. Competing approaches yielded 10--24% error. We used our SoftwarePilot benchmarks to study (1) GPU acceleration, scale up, and scale out, (2) onboard, edge and cloud computing, (3) energy and monetary budgets, and (4) software driven GPU management. We found that model-driven management can boost mission throughput by 10X and reduce costs by 87%.

Precision agriculture examines crop fields, gathers data, analyzes crop health and informs field management. This data driven approach can reduce fertilizer runoff, prevent crop disease and increase yield. Frequent data collection improves outcomes, but also increases operating costs. Fully autonomous aerial systems (FAAS) can capture detailed images of crop fields without human intervention. They can reduce operating costs significantly. However, FAAS software must embed agricultural expertise to decide where to fly, which images to capture and when to land. This paper explores fully autonomous precision agriculture where FAAS map crop fields frequently. We have designed hardware and software architecture. We use unmanned aerial systems, edge computing components and software driven by reinforcement learning and ensemble models. In early results, we have collected data from an Ohio cornfield. We use this data to simulate a FAAS modeling crop yield. Our results (1) show that our approach predicts yield well and (2) can quantify computational demand. Computational costs can be prohibitive. We discuss how research on adaptive systems can reduce costs and enable fully autonomous precision agriculture. We also provide our simulation tools and dataset as part of our open source FAAS middleware, SoftewarePilot.

Automation is utilized heavily in many domains to increase productivity. With new, more complex automation, like the self-driving car, humans will be required to forego direct task performance in favor of maintaining a supervisory role over automation systems. While the use of these systems generally results in greater performance than humans performing alone, humans are reluctant to adopt these superior systems due to a lack of trust. The United States Department of Defense is investigating trust in automation in order to influence the rate of adoption of automation technology. Studying trust in automation systems requires a mechanism for quantifying and measuring trust. This paper proposes a method for measuring human trust behaviors with regard to human-automation systems through response rates of compliance and reliance. Using behavioral data from a human-subjects experiment involving automated agents, we create a system dynamics model which relates trust to other system level variables. Using this trust model, engineers will be able to study trust in human-automation team scenarios in order to design automation systems with higher rates of adoption.