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

Many realistic robotics tasks are best solved compositionally, throughcontrol architectures that sequentially invoke primitives and achieve errorcorrection through the use of loops and conditionals taking the system back toalternative earlier states. Recent end-to-end approaches to task learningattempt to directly learn a single controller that solves an entire task, butthis has been difficult for complex control tasks that would have otherwiserequired a diversity of local primitive moves, and the resulting solutions arealso not easy to inspect for plan monitoring purposes. In this work, we aim tobridge the gap between hand designed and learned controllers, by representingeach as an option in a hybrid hierarchical Reinforcement Learning framework -DynoPlan. We extend the options framework by adding a dynamics model and theuse of a nearness-to-goal heuristic, derived from demonstrations. Thistranslates the optimization of a hierarchical policy controller to a problem ofplanning with a model predictive controller. By unrolling the dynamics of eachoption and assessing the expected value of each future state, we can create asimple switching controller for choosing the optimal policy within aconstrained time horizon similarly to hill climbing heuristic search. Theindividual dynamics model allows each option to iterate and be activatedindependently of the specific underlying instantiation, thus allowing for a mixof motion planning and deep neural network based primitives. We can assess thesafety regions of the resulting hybrid controller by investigating theinitiation sets of the different options, and also by reasoning about thecompleteness and performance guarantees of the underpinning motion planners.

DynoPlan: Combining Motion Planning and Deep Neural Network based Controllers for Safe HRL