D3: A Dynamic Deadline-Driven Approach for Building Autonomous Vehicles

Adjust response times and handle missed deadlines to maximize application-wide accuracy.

D3 centralizes time management through the Deadline Policy, which sets an end-to-end deadline and decomposes it into per-operator deadlines. Operators proactively try to meet their deadlines. However, if deadlines are missed, D3 executes reactive measures and adjusts downstream deadlines.

Overview

• Autonomous Vehicles (AVs) must adjust their response times to react to changes in the driving environment, such as an obscured pedestrian emerging from behind a parked car.
• The environment further affects the processing time of the models and algorithms used by AVs. For example, predicting the future positions of vehicles and pedestrians at a busy intersection requires more processing power than navigating an empty highway.
• D3 provides abstractions to assign and manage deadlines across a pipeline, and enables AVs to handle missed deadlines and adjust to changes in the available processing time.
• With D3, we observe a 68% reduction in collisions compared to existing execution models across 50 km of challenging simulated driving scenarios.

Abstract

Autonomous vehicles (AVs) must drive across a variety of challenging environments that impose continuously-varying deadlines and runtime-accuracy tradeoffs on their software pipelines. A deadline-driven execution of such AV pipelines requires a new class of systems that enable the computation to maximize accuracy under dynamically-varying deadlines. Designing these systems presents interesting challenges that arise from combining ease-of-development of AV pipelines with deadline specification and enforcement mechanisms.

Our work addresses these challenges through D3 (Dynamic Deadline-Driven), a novel execution model that centralizes the deadline management, and allows applications to adjust their computation by modeling missed deadlines as exceptions. Further, we design and implement ERDOS, an open-source realization of D3 for AV pipelines that exposes fine-grained execution events to applications, and provides mechanisms to speculatively execute computation and enforce deadlines between an arbitrary set of events. Finally, we address the crucial lack of AV benchmarks through our state-of-the- art open-source AV pipeline, Pylot, that works seamlessly across simulators and real AVs. We evaluate the efficacy of D3 and ERDOS by driving Pylot across challenging driving scenarios spanning 50km, and observe a 68% reduction in collisions as compared to prior execution models

Cite

@inproceedings{gog2022d3,
  title={D3: a dynamic deadline-driven approach for building autonomous vehicles},
  author={Gog, Ionel and Kalra, Sukrit and Schafhalter, Peter and Gonzalez, Joseph E and Stoica, Ion},
  booktitle={Proceedings of the Seventeenth European Conference on Computer Systems},
  pages={453--471},
  year={2022}
}