Reproduce “Autonomous Driving With Perception Uncertainties: Deep-Ensemble Based Adaptive Cruise Control” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xiao Li, H. Eric Tseng, Anouck Girard, Ilya Kolmanovsky

An Adaptive Cruise Control (ACC) system that integrates deep learning-based perception uncertainty into its control loop has been developed. The system uses a Deep Ensemble to quantify distance headway uncertainty and Stochastic Model Predictive Control (MPC) to ensure probabilistic safety, demonstrating robust performance in both in-distribution and out-of-distribution driving conditions.

Key insights

  • Deep Ensembles, especially with heterogeneous architectures, effectively quantify both aleatoric and epistemic uncertainty in perception, which is critical for robustness in Out-Of-Distribution (OOD) scenarios.
  • Integrating quantified perception uncertainty directly into a Stochastic Model Predictive Control (MPC) with chance constraints enables the controller to make probabilistically safe and adaptive decisions.
  • Autonomous systems can adopt conservative, safer behaviors by acknowledging and reacting to high perceptual uncertainty (e.g., from OOD inputs), rather than failing catastrophically.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.