Matches in SemOpenAlex for { <https://semopenalex.org/work/W3174334221> ?p ?o ?g. }
Showing items 1 to 79 of
79
with 100 items per page.
- W3174334221 abstract "Underactuated robot designs are enticing due to their electromechanical simplicity; but, their operation is complex especially for compliant designs that may not have an explicit model. Deep reinforcement learning methods together with multi-joint dynamic simulation may help overcome this control bottleneck. However, such systems tend to have large continuous action spaces and exhibit sparse rewards, thus making control policy exploration variable and non-trivial. Using the observations: (1) that an underactuated system has coupled states, and (2) that a deep function approximator may generalize across compressed (and potentially nonintuitive) states, we consider a reduced order approach based around a generative autoencoder so as to automatically find a better representation to focus a control policy exploration process. To help evaluate this approach, we also introduce The Jitterbug Problem, which is a series of increasingly specific and challenging motion control tasks for a highly compliant legged toy robot with just a single motor locomoting across a field. We benchmarked this problem against offpolicy and on-policy deep reinforcement learning algorithms and find that an off-policy approach had better median rewards, but has higher variability. Through this study we find that reducing the model by taking advantage of the latent structure may exhibit similar (reward) performance, yet empirically has less training variance and slightly better learning rates." @default.
- W3174334221 created "2021-07-05" @default.
- W3174334221 creator A5010900748 @default.
- W3174334221 creator A5023263662 @default.
- W3174334221 creator A5091885274 @default.
- W3174334221 date "2021-01-27" @default.
- W3174334221 modified "2023-10-18" @default.
- W3174334221 title "A reduced-order approach to assist with reinforcement learning for underactuated robotics" @default.
- W3174334221 hasPublicationYear "2021" @default.
- W3174334221 type Work @default.
- W3174334221 sameAs 3174334221 @default.
- W3174334221 citedByCount "0" @default.
- W3174334221 crossrefType "proceedings-article" @default.
- W3174334221 hasAuthorship W3174334221A5010900748 @default.
- W3174334221 hasAuthorship W3174334221A5023263662 @default.
- W3174334221 hasAuthorship W3174334221A5091885274 @default.
- W3174334221 hasConcept C101738243 @default.
- W3174334221 hasConcept C108583219 @default.
- W3174334221 hasConcept C119857082 @default.
- W3174334221 hasConcept C149635348 @default.
- W3174334221 hasConcept C154945302 @default.
- W3174334221 hasConcept C17744445 @default.
- W3174334221 hasConcept C199539241 @default.
- W3174334221 hasConcept C202444582 @default.
- W3174334221 hasConcept C2776359362 @default.
- W3174334221 hasConcept C2780513914 @default.
- W3174334221 hasConcept C33923547 @default.
- W3174334221 hasConcept C34413123 @default.
- W3174334221 hasConcept C41008148 @default.
- W3174334221 hasConcept C88337583 @default.
- W3174334221 hasConcept C90509273 @default.
- W3174334221 hasConcept C94625758 @default.
- W3174334221 hasConcept C9652623 @default.
- W3174334221 hasConcept C97541855 @default.
- W3174334221 hasConceptScore W3174334221C101738243 @default.
- W3174334221 hasConceptScore W3174334221C108583219 @default.
- W3174334221 hasConceptScore W3174334221C119857082 @default.
- W3174334221 hasConceptScore W3174334221C149635348 @default.
- W3174334221 hasConceptScore W3174334221C154945302 @default.
- W3174334221 hasConceptScore W3174334221C17744445 @default.
- W3174334221 hasConceptScore W3174334221C199539241 @default.
- W3174334221 hasConceptScore W3174334221C202444582 @default.
- W3174334221 hasConceptScore W3174334221C2776359362 @default.
- W3174334221 hasConceptScore W3174334221C2780513914 @default.
- W3174334221 hasConceptScore W3174334221C33923547 @default.
- W3174334221 hasConceptScore W3174334221C34413123 @default.
- W3174334221 hasConceptScore W3174334221C41008148 @default.
- W3174334221 hasConceptScore W3174334221C88337583 @default.
- W3174334221 hasConceptScore W3174334221C90509273 @default.
- W3174334221 hasConceptScore W3174334221C94625758 @default.
- W3174334221 hasConceptScore W3174334221C9652623 @default.
- W3174334221 hasConceptScore W3174334221C97541855 @default.
- W3174334221 hasLocation W31743342211 @default.
- W3174334221 hasOpenAccess W3174334221 @default.
- W3174334221 hasPrimaryLocation W31743342211 @default.
- W3174334221 hasRelatedWork W2100370041 @default.
- W3174334221 hasRelatedWork W2610395436 @default.
- W3174334221 hasRelatedWork W2759034818 @default.
- W3174334221 hasRelatedWork W2782446262 @default.
- W3174334221 hasRelatedWork W2796496570 @default.
- W3174334221 hasRelatedWork W2803281228 @default.
- W3174334221 hasRelatedWork W2894383471 @default.
- W3174334221 hasRelatedWork W2944714385 @default.
- W3174334221 hasRelatedWork W2945500224 @default.
- W3174334221 hasRelatedWork W2966950754 @default.
- W3174334221 hasRelatedWork W2968021416 @default.
- W3174334221 hasRelatedWork W2978069508 @default.
- W3174334221 hasRelatedWork W2990123902 @default.
- W3174334221 hasRelatedWork W3005607450 @default.
- W3174334221 hasRelatedWork W3102797050 @default.
- W3174334221 hasRelatedWork W3109409708 @default.
- W3174334221 hasRelatedWork W3138650132 @default.
- W3174334221 hasRelatedWork W3149922422 @default.
- W3174334221 hasRelatedWork W3188853186 @default.
- W3174334221 hasRelatedWork W3202944960 @default.
- W3174334221 isParatext "false" @default.
- W3174334221 isRetracted "false" @default.
- W3174334221 magId "3174334221" @default.
- W3174334221 workType "article" @default.