F-1 Roofline

Skyline is an interactive tool for the F-1 model to help computer architects understand various bottlenecks in a UAV system. Understanding the bottlenecks is crucial for building onboard compute systems. It considers the interactions of various components such as sensors, onboard compute, autonomy algorithms, and UAV physics to give insights into which of these limits the performance of the UAV. In addition, as computer architects design specialized hardware accelerators for these UAVs, it is essential to understand how their optimization effort translates to UAV performance. More importantly, it can quantify how much performance improvement is required to reach the peak performance possible for a given UAV.

Some of the questions that can be answered using the Skyline tool are:

• Given a UAV system and several onboard compute choices, how do we systematically select the compute system to maximize UAV performance (e.g., Safe high-speed autonomy)

• Given multiple autonomy algorithms (e.g., Sense-Plan-Act and E2E), how do we systematically characterize these algorithms?

• How do we characterize the impacts of adding Dual Modular Redundancies/Triple Modular Redundancies for UAV systems?

• Given multiple sensor choices (framerates, types, etc.), how do we systematically characterize the impact of these sensors?

• Given several UAV components, how do we perform holistic UAV system characterization and understand which one is better?