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

Today's robots often interface with data-driven perception and planningmodels with classical model-predictive controllers (MPC). Often, such learnedperception/planning models produce erroneous waypoint predictions onout-of-distribution (OoD) or even adversarial visual inputs, which increasecontrol costs. However, today's methods to train robust perception models arelargely task-agnostic - they augment a dataset using random imagetransformations or adversarial examples targeted at the vision model inisolation. As such, they often introduce pixel perturbations that areultimately benign for control. In contrast to prior work that synthesizesadversarial examples for single-step vision tasks, our key contribution is tosynthesize adversarial scenarios tailored to multi-step, model-based control.To do so, we use differentiable MPC methods to calculate the sensitivity of amodel-based controller to errors in state estimation. We show that re-trainingvision models on these adversarial datasets improves control performance on OoDtest scenarios by up to 36.2% compared to standard task-agnostic dataaugmentation. We demonstrate our method on examples of robotic navigation,manipulation in RoboSuite, and control of an autonomous air vehicle.

Synthesizing Adversarial Visual Scenarios for Model-Based Robotic Control