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

Traversability prediction is a fundamental perception capability forautonomous navigation. Deep neural networks (DNNs) have been widely used topredict traversability during the last decade. The performance of DNNs issignificantly boosted by exploiting a large amount of data. However, thediversity of data in different domains imposes significant gaps in theprediction performance. In this work, we make efforts to reduce the gaps byproposing a novel pseudo-trilateral adversarial model that adopts acoarse-to-fine alignment (CALI) to perform unsupervised domain adaptation(UDA). Our aim is to transfer the perception model with high data efficiency,eliminate the prohibitively expensive data labeling, and improve thegeneralization capability during the adaptation from easy-to-access sourcedomains to various challenging target domains. Existing UDA methods usuallyadopt a bilateral zero-sum game structure. We prove that our CALI model -- apseudo-trilateral game structure is advantageous over existing bilateral gamestructures. This proposed work bridges theoretical analyses and algorithmdesigns, leading to an efficient UDA model with easy and stable training. Wefurther develop a variant of CALI -- Informed CALI (ICALI), which is inspiredby the recent success of mixup data augmentation techniques and mixesinformative regions based on the results of CALI. This mixture step provides anexplicit bridging between the two domains and exposes underperforming classesmore during training. We show the superiorities of our proposed models overmultiple baselines in several challenging domain adaptation setups. To furthervalidate the effectiveness of our proposed models, we then combine ourperception model with a visual planner to build a navigation system and showthe high reliability of our model in complex natural environments.

Pseudo-Trilateral Adversarial Training for Domain Adaptive Traversability Prediction