Visible to the public Co-Design of Networking and Decentralized Control to Enable Aerial Networking in an Uncertain Airspace

Abstract: Airborne networking utilizes direct flight-to-to-flight communication for flexible information sharing, safe maneuvering, and coordination of time-critical missions. It is challenging because of the high mobility, stringent safety requirements, and uncertain airspace environment. This project uses a co-design approach that exploits the mutual benefits of networking and decentralized mobility control in an uncertain heterogeneous environment. The approach departs from the usual perspective that views physical mobility as communication constraints, communication as constraints for decentralized mobility control, and uncertain environment as constraints for both. Instead, we proactively exploits the constraints, uncertainty, and new structures with information to enable high-performance designs. The features of the co-design such as scalability, fast response, trackability, and robustness to uncertainty advance the core CPS science on decision-making for large-scale networks under uncertainty.

During the period of 2017-2018, the following results are achieved:

  1. With respect to random mobility models with the equipment of safety constraints, we studied 1-D and 2-D Random Direction (RD) Random Mobility Models (RMM) equipped with the sense-and-stop protocol. This study provided us insights of the capacity limits of an airspace that has dense UAS operations. The analysis and simulation studies lead to interesting insights such as that the commonly used sense-and-stop protocol is not effective when the randomness of UAV mobility is high.
  2. With respect to the estimation of RMMs, we formulated the dynamics of the movement characteristics generated by the two types of random variables as a special Jump Markov System, and developed an estimation method based on the Expectation-Maximization principle. We applied the estimation method to the Smooth-Turn RMM developed for fixed-wing unmanned aerial vehicles.
  3. With respect to learning-based controls under high-dimensional uncertainties, we explore the capabilities of M-PCM and M-PCM-OFFD-based optimal control and adaptive control using the reinforcement learning approach. For three cases covering varying scenarios, we prove that the control solution optimal to the sampled uncertainty space produced by M-PCM or M-PCM-OFFD is also optimal to the original uncertainty space under simple assumptions on the forms of the cost functions and orders of uncertain parameters. We further integrated Q learning and M-PCM-OFFD to address a broad class of stochastic optimal control problems of finite state and control spaces, which is also new to the literature. The approach can quickly converge to the optimal solution under high-dimensional uncertainties.
  4. With respect to the effectiveness of layered structures and the equipment of additional memories to UAV networking, , we use distributed consensus as a canonical distributed computing task to study the effectiveness of the data transmission in digitized (quantized) channels for UAV networks. We show that layered structures are more effective than equivalent egalitarian structures in terms of the data transmission load required to reach consensus. In particular, we establish explicit relationships between simple structural characteristics and the performance of quantized consensus (e.g., consensus condition, consensus value, and transmission load to reach consensus) for broad classes of layered structures. We also provide analytical results on asymptotic and transient performance when additional local memories are used to further reduce the data transmission load to reach consensus.
  5. With respect to the communication and control co-design for the algorithm and testbed implementation of long-distance UAV networking, we completed two versions. The first version used a GPS-based control algorithm to automatically reject wind disturbance and align the directions of antennas in accordance with UAV movement. A received signal strength indicator-based decentralized initial scan algorithm was also designed and implemented to quickly establish initial connection between the UAVs. The second version realized the self-alignment of UAV-mounted directional antennas over a long distance through fusing the GPS and communication channel characteristic measured by received signal strength indicator, using unscented Kalman filter and fuzzy logic. The solution significantly enhances the performance of wireless communication channel in imperfect environment subject to the unavailability of GPS signals and unstable strength of wireless signals. We chose begalbone and ROS for the testbed implementation.

Explanation of Demonstration: We will demonstrate a UAV networking system that integrates communication and control components. The system includes an interface that allows the diagnosis of the performance of the remote system.

Creative Commons 2.5
Switch to experimental viewer