English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Deep learning provides a powerful tool for machine perception when theobservations resemble the training data. However, real-world robotic systemsmust react intelligently to their observations even in unexpectedcircumstances. This requires a system to reason about its own uncertainty givenunfamiliar, out-of-distribution observations. Approximate Bayesian approachesare commonly used to estimate uncertainty for neural network predictions, butcan struggle with out-of-distribution observations. Generative models can inprinciple detect out-of-distribution observations as those with a low estimateddensity. However, the mere presence of an out-of-distribution input does not byitself indicate an unsafe situation. In this paper, we present a method foruncertainty-aware robotic perception that combines generative modeling andmodel uncertainty to cope with uncertainty stemming from out-of-distributionstates. Our method estimates an uncertainty measure about the model'sprediction, taking into account an explicit (generative) model of theobservation distribution to handle out-of-distribution inputs. This isaccomplished by probabilistically projecting observations onto the trainingdistribution, such that out-of-distribution inputs map to uncertainin-distribution observations, which in turn produce uncertain task-relatedpredictions, but only if task-relevant parts of the image change. We evaluateour method on an action-conditioned collision prediction task with bothsimulated and real data, and demonstrate that our method of projectingout-of-distribution observations improves the performance of four standardBayesian and non-Bayesian neural network approaches, offering more favorabletrade-offs between the proportion of time a robot can remain autonomous and theproportion of impending crashes successfully avoided.

Robustness to Out-of-Distribution Inputs via Task-Aware Generative Uncertainty