Equipment operation represents one of the most dangerous tasks on a construction sites and accidents related to such operation often result in death and property damage on the construction site and the surrounding area. Such accidents can also cause considerable delays and disruption, and negatively impact the efficiency of operations. This award will conduct research to improve the safety and efficiency of cranes by integrating advances in robotics, computer vision, and construction management. It will create tools for quick and easy planning of crane operations and incorporate them into a safe and efficient system that can monitor a crane's environment and provide control feedback to the crane and the operator. Resulting gains in safety and efficiency wil reduce fatal and non-fatal crane accidents. Partnerships with industry will also ensure that these advances have a positive impact on construction practice, and can be extended broadly to smart infrastructure, intelligent manufacturing, surveillance, traffic monitoring, and other application areas. The research will involve undergraduates and includes outreach to K-12 students. The work is driven by the hypothesis that the monitoring and control of cranes can be performed autonomously using robotics and computer vision algorithms, and that detailed and continuous monitoring and control feedback can lead to improved planning and simulation of equipment operations. It will particularly focus on developing methods for (a) planning construction operations while accounting for safety hazards through simulation; (b) estimating and providing analytics on the state of the equipment; (c) monitoring equipment surrounding the crane operating environment, including detection of safety hazards, and proximity analysis to dynamic resources including materials, equipment, and workers; (d) controlling crane stability in real-time; and (e) providing feedback to the user and equipment operators in a "transparent cockpit" using visual and haptic cues. It will address the underlying research challenges by improving the efficiency and reliability of planning through failure effects analysis and creating methods for contact state estimation and equilibrium analysis; improving monitoring through model-driven and real-time 3D reconstruction techniques, context-driven object recognition, and forecasting motion trajectories of objects; enhancing reliability of control through dynamic crane models, measures of instability, and algorithms for finding optimal controls; and, finally, improving efficiency of feedback loops through methods for providing visual and haptic cues.

The age of autonomous mobile systems is dawning -- from autonomous cars to household robots to aerial drones -- and they are expected to transform multiple industries and have significant impact on the US economy. Through wireless coordination, these systems create a whole that is greater than the sum of its parts. For example, vehicle "platoons" increase both highway throughput and fuel efficiency by traveling nearly bumper-to-bumper, using a wireless coupling to brake and accelerate simultaneously. Similarly, vehicles or drones can speed around blind corners using the sensing capabilities of the agents ahead of them. However, wireless communication is still considered too unreliable for safety-critical operations like these. This research is creating new techniques for safe wirelessly coordinated mobility, which is becoming increasingly important with the proliferation of autonomous mobile systems. The approach is to develop a framework for joint modeling and analysis of motion and communication in order to find provably safe coordination paths. This includes new models that can predict the effect of motion paths on the wireless channel, together with new formal methods that can use these models in a tractable manner to synthesize control strategies with provable guarantees. The key innovations include new methods to assess the validity of a Radio Frequency model, new methods for tractable probabilistic reasoning over complex models of the wireless channel and protocols, and new control strategies that achieve provable safety guarantees for states that would have been unsafe without wireless coordination. If successful, this research will allow mobile systems to realize the performance benefits of wireless coordination while preserving the ability to provide provable safety guarantees. The focus is not on improving the wireless channel reliability; instead, the aim is to provide safety guarantees on the entire mobile system by modeling and analyzing the channel's dynamic properties in a rapidly changing environment.

Human-robot teams engaged in transportation and data collection will often share a common physical workspace. This project will investigate fundamental challenges in human-cyberphysical-systems (h-CPS) for cooperative aerial payload transport. First, Unmanned Aerial Vehicles (UAVs) cooperatively lift and carry a payload through a cluttered environment under uncertain winds. The multi-UAV system (MUS) functions autonomously to allow human companions to focus attention on their environment while interacting with the MUS. We propose a novel interface where an operator pushes on the slung payload to guide the team and coordinates the mission through a networked tablet. A novel cooperative control strategy safely guides the MUS while physics-based algorithms distinguish human inputs from environmental disturbances. Flight tests will demonstrate and validate the h-CPS. The PI and mentored postdoctoral researcher will involve students from under-represented groups and K-12 students in safe MUS flight demonstrations. This project offers three research advances: MUS scalability and collision avoidance guarantees through continuum deformation cooperative control, safe MUS compensation for vehicle anomalies, and cognitively-tractable user interfaces. Particularly novel to this work is the h-CPS interface in which an operator pushes on the payload to guide the MUS team. We will apply linear momentum analysis to sense haptic cues and will validate our models in simulation and flight testing. Mission-level decision-making will be performed through system modeling as a Markov game in which game states are defined from human, environment, and aggregate MUS state. Our method abstracts MUS behaviors to reduce cognitive complexity and real-time network and computational overhead.