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

Training control policies in simulation is more appealing than on real robotsdirectly, as it allows for exploring diverse states in an efficient manner.Yet, robot simulators inevitably exhibit disparities from the real-world\rebut{dynamics}, yielding inaccuracies that manifest as the dynamicalsimulation-to-reality (sim-to-real) gap. Existing literature has proposed toclose this gap by actively modifying specific simulator parameters to align thesimulated data with real-world observations. However, the set of tunableparameters is usually manually selected to reduce the search space in acase-by-case manner, which is hard to scale up for complex systems and requiresextensive domain knowledge. To address the scalability issue and automate theparameter-tuning process, we introduce COMPASS, which aligns the simulator withthe real world by discovering the causal relationship between the environmentparameters and the sim-to-real gap. Concretely, our method learns adifferentiable mapping from the environment parameters to the differencesbetween simulated and real-world robot-object trajectories. This mapping isgoverned by a simultaneously learned causal graph to help prune the searchspace of parameters, provide better interpretability, and improvegeneralization on unseen parameters. We perform experiments to achieve bothsim-to-sim and sim-to-real transfer, and show that our method has significantimprovements in trajectory alignment and task success rate over strongbaselines in several challenging manipulation tasks.

What Went Wrong? Closing the Sim-to-Real Gap via Differentiable Causal Discovery