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

Soft actor-critic (SAC) in reinforcement learning is expected to be one ofthe next-generation robot control schemes. Its ability to maximize policyentropy would make a robotic controller robust to noise and perturbation, whichis useful for real-world robot applications. However, the priority ofmaximizing the policy entropy is automatically tuned in the currentimplementation, the rule of which can be interpreted as one for equalityconstraint, binding the policy entropy into its specified lower bound. Thecurrent SAC is therefore no longer maximize the policy entropy, contrary to ourexpectation. To resolve this issue in SAC, this paper improves itsimplementation with a learnable state-dependent slack variable forappropriately handling the inequality constraint to maximize the policy entropyby reformulating it as the corresponding equality constraint. The introducedslack variable is optimized by a switching-type loss function that takes intoaccount the dual objectives of satisfying the equality constraint and checkingthe lower bound. In Mujoco and Pybullet simulators, the modified SACstatistically achieved the higher robustness for adversarial attacks thanbefore while regularizing the norm of action. A real-robot variable impedancetask was demonstrated for showing the applicability of the modified SAC toreal-world robot control. In particular, the modified SAC maintained adaptivebehaviors for physical human-robot interaction, which had no experience at allduring training. https://youtu.be/EH3xVtlVaJw

Soft Actor-Critic Algorithm with Truly-satisfied Inequality Constraint