CPS: Breakthrough: Selective Listening - Control for Connected Autonomous Vehicles in Data-Rich Environments

Abstract: Two current trends promise to revolutionize the safety, reliability, and energy-efficiency of future automotive transportation: (i) wireless connectivity of vehicles to each other, to smart infrastructure, and to other mobile devices, and (ii) autonomy, ranging from driver assistance to full self-driving autonomy. Connected autonomous vehicles (CAVs) are cyber-physical systems with increasingly complex software algorithms in control of a physical vehicle moving in uncertain real-world environments. Planned connectivity regulations and recent advances in vehicular autonomy by leading manufacturers imply that CAVs will be ubiquitous in the near future. Roads will be data-rich environments where a large number of wireless devices attached to vehicles, infrastructure, personal electronics, and wearable gadgets will transmit multimodal data. In this scenario, two serious challenges arise for CAVs: (1) Onboard computational limitations may imply a de facto upper bound on the number of data sources that can be accommodated by the autonomy algorithms, and may introduce the problem of appropriately choosing a smaller subset of the available data, and (2) Given the finite amount of radio frequency spectrum and the rapidly growing number of wireless applications and end-users, spectrum scarcity arises, for which current communication protocols do not suffice. We observe that these two challenges are in fact intricately related, and that it is beneficial to address them together. To this end, the goal of this project is to investigate bidirectional interactions between the technologies of autonomy and of wireless connectivity in cyber-physical systems. Using CAVs as a case study in cyber-physical systems, we propose to investigate how estimation and control algorithms affect - and are affected by - software-defined radio communications in spectrum-scarce, data-rich environments. The technical premise of this project is an emphasis on data-rich environments, where too much data can overwhelm autonomy algorithms, e.g. real-time short-horizon trajectory planners for vehicles. The proposed approach of selecting the data sources that are likely to be the "most informative" is a new aspect compared to planning algorithms in the literature. Furthermore, this selection is dynamic, in that it evolves with the trajectory plan. This selective connectivity also helps the wireless spectrum sensing algorithm to converge faster by limiting the spatial regions to sweep for potential connections. The proposed trajectory planning algorithm is based on the so-called method of lifted graphs, which promises to bridge the gap between fast geometric path planning algorithms and slower control-theoretic techniques that incorporate vehicle dynamical constraints. Beyond CAVs, the proposed technical approach can be applied to other cyber-physical systems where several non-cooperative agents communicate over wireless channels. The proposed trajectory planning approach is sufficiently general to allow the formulations and solutions of different application-specific planning problems.