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

Simulated humanoids are an appealing research domain due to their physicalcapabilities. Nonetheless, they are also challenging to control, as a policymust drive an unstable, discontinuous, and high-dimensional physical system.One widely studied approach is to utilize motion capture (MoCap) data to teachthe humanoid agent low-level skills (e.g., standing, walking, and running) thatcan then be re-used to synthesize high-level behaviors. However, even withMoCap data, controlling simulated humanoids remains very hard, as MoCap dataoffers only kinematic information. Finding physical control inputs to realizethe demonstrated motions requires computationally intensive methods likereinforcement learning. Thus, despite the publicly available MoCap data, itsutility has been limited to institutions with large-scale compute. In thiswork, we dramatically lower the barrier for productive research on this topicby training and releasing high-quality agents that can track over three hoursof MoCap data for a simulated humanoid in the dm_control physics-basedenvironment. We release MoCapAct (Motion Capture with Actions), a dataset ofthese expert agents and their rollouts, which contain proprioceptiveobservations and actions. We demonstrate the utility of MoCapAct by using it totrain a single hierarchical policy capable of tracking the entire MoCap datasetwithin dm_control and show the learned low-level component can be re-used toefficiently learn downstream high-level tasks. Finally, we use MoCapAct totrain an autoregressive GPT model and show that it can control a simulatedhumanoid to perform natural motion completion given a motion prompt.  Videos of the results and links to the code and dataset are available athttps://microsoft.github.io/MoCapAct.

MoCapAct: A Multi-Task Dataset for Simulated Humanoid Control