Spatial Uncertainty Sampling for End-to-End Control
This work addresses safety and reliability challenges for autonomous vehicles in ambiguous road conditions, though it is incremental as it builds on existing Bayesian deep learning approaches.
The paper tackles the problem of reliable uncertainty estimation in end-to-end neural networks for autonomous vehicle control by proposing a Bayesian neural network that exploits feature map correlation during training, achieving improved model fits and tighter uncertainty estimates compared to traditional dropout methods.
End-to-end trained neural networks (NNs) are a compelling approach to autonomous vehicle control because of their ability to learn complex tasks without manual engineering of rule-based decisions. However, challenging road conditions, ambiguous navigation situations, and safety considerations require reliable uncertainty estimation for the eventual adoption of full-scale autonomous vehicles. Bayesian deep learning approaches provide a way to estimate uncertainty by approximating the posterior distribution of weights given a set of training data. Dropout training in deep NNs approximates Bayesian inference in a deep Gaussian process and can thus be used to estimate model uncertainty. In this paper, we propose a Bayesian NN for end-to-end control that estimates uncertainty by exploiting feature map correlation during training. This approach achieves improved model fits, as well as tighter uncertainty estimates, than traditional element-wise dropout. We evaluate our algorithms on a challenging dataset collected over many different road types, times of day, and weather conditions, and demonstrate how uncertainties can be used in conjunction with a human controller in a parallel autonomous setting.