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

Physical simulations that accurately model reality are crucial for manyengineering disciplines such as mechanical engineering and robotic motionplanning. In recent years, learned Graph Network Simulators produced accuratemesh-based simulations while requiring only a fraction of the computationalcost of traditional simulators. Yet, the resulting predictors are confined tolearning from data generated by existing mesh-based simulators and thus cannotinclude real world sensory information such as point cloud data. As thesepredictors have to simulate complex physical systems from only an initialstate, they exhibit a high error accumulation for long-term predictions. Inthis work, we integrate sensory information to ground Graph Network Simulatorson real world observations. In particular, we predict the mesh state ofdeformable objects by utilizing point cloud data. The resulting model allowsfor accurate predictions over longer time horizons, even under uncertainties inthe simulation, such as unknown material properties. Since point clouds areusually not available for every time step, especially in online settings, weemploy an imputation-based model. The model can make use of such additionalinformation only when provided, and resorts to a standard Graph NetworkSimulator, otherwise. We experimentally validate our approach on a suite ofprediction tasks for mesh-based interactions between soft and rigid bodies. Ourmethod results in utilization of additional point cloud information toaccurately predict stable simulations where existing Graph Network Simulatorsfail.

Grounding Graph Network Simulators using Physical Sensor Observations