Matches in SemOpenAlex for { <https://semopenalex.org/work/W3128176255> ?p ?o ?g. }
- W3128176255 endingPage "1235" @default.
- W3128176255 startingPage "1226" @default.
- W3128176255 abstract "Deploying reinforcement learning (RL) involves major concerns around safety. Engineering a reward signal that allows the agent to maximize its performance while remaining safe is not trivial. Safe RL studies how to mitigate such problems. For instance, we can decouple safety from reward using constrained Markov decision processes (CMDPs), where an independent signal models the safety aspects. In this setting, an RL agent can autonomously find tradeoffs between performance and safety. Unfortunately, most RL agents designed for CMDPs only guarantee safety after the learning phase, which might prevent their direct deployment. In this work, we investigate settings where a concise abstract model of the safety aspects is given, a reasonable assumption since a thorough understanding of safety-related matters is a prerequisite for deploying RL in typical applications. Factored CMDPs provide such compact models when a small subset of features describe the dynamics relevant for the safety constraints. We propose an RL algorithm that uses this abstract model to learn policies for CMDPs safely, that is without violating the constraints. During the training process, this algorithm can seamlessly switch from a conservative policy to a greedy policy without violating the safety constraints. We prove that this algorithm is safe under the given assumptions. Empirically, we show that even if safety and reward signals are contradictory, this algorithm always operates safely and, when they are aligned, this approach also improves the agent's performance." @default.
- W3128176255 created "2021-02-15" @default.
- W3128176255 creator A5012669709 @default.
- W3128176255 creator A5035570774 @default.
- W3128176255 creator A5075395956 @default.
- W3128176255 date "2021-01-01" @default.
- W3128176255 modified "2023-09-26" @default.
- W3128176255 title "AlwaysSafe: Reinforcement Learning without Safety Constraint Violations during Training" @default.
- W3128176255 cites W1518931405 @default.
- W3128176255 cites W1537180453 @default.
- W3128176255 cites W1650504995 @default.
- W3128176255 cites W1845972764 @default.
- W3128176255 cites W1850488217 @default.
- W3128176255 cites W1859820673 @default.
- W3128176255 cites W2020294948 @default.
- W3128176255 cites W2058735307 @default.
- W3128176255 cites W2071815241 @default.
- W3128176255 cites W2097931172 @default.
- W3128176255 cites W2103012681 @default.
- W3128176255 cites W2112899086 @default.
- W3128176255 cites W2119567691 @default.
- W3128176255 cites W2127323769 @default.
- W3128176255 cites W2128957716 @default.
- W3128176255 cites W2151620419 @default.
- W3128176255 cites W2222789563 @default.
- W3128176255 cites W2397240726 @default.
- W3128176255 cites W2468354762 @default.
- W3128176255 cites W2604677816 @default.
- W3128176255 cites W2618318883 @default.
- W3128176255 cites W2739747865 @default.
- W3128176255 cites W2758731390 @default.
- W3128176255 cites W2788084076 @default.
- W3128176255 cites W2804791273 @default.
- W3128176255 cites W2808155616 @default.
- W3128176255 cites W2904450917 @default.
- W3128176255 cites W2914474570 @default.
- W3128176255 cites W2945887696 @default.
- W3128176255 cites W2946284958 @default.
- W3128176255 cites W2963082979 @default.
- W3128176255 cites W2963575966 @default.
- W3128176255 cites W2964340170 @default.
- W3128176255 cites W2965997659 @default.
- W3128176255 cites W2971013032 @default.
- W3128176255 cites W2991935368 @default.
- W3128176255 cites W3001756029 @default.
- W3128176255 cites W3009922106 @default.
- W3128176255 cites W3034840734 @default.
- W3128176255 cites W3037396281 @default.
- W3128176255 cites W3038032959 @default.
- W3128176255 cites W3040161731 @default.
- W3128176255 cites W3040979139 @default.
- W3128176255 cites W3081016801 @default.
- W3128176255 cites W3092461179 @default.
- W3128176255 cites W3102039646 @default.
- W3128176255 cites W3129954366 @default.
- W3128176255 cites W3176452384 @default.
- W3128176255 cites W3176971532 @default.
- W3128176255 hasPublicationYear "2021" @default.
- W3128176255 type Work @default.
- W3128176255 sameAs 3128176255 @default.
- W3128176255 citedByCount "2" @default.
- W3128176255 countsByYear W31281762552020 @default.
- W3128176255 crossrefType "proceedings-article" @default.
- W3128176255 hasAuthorship W3128176255A5012669709 @default.
- W3128176255 hasAuthorship W3128176255A5035570774 @default.
- W3128176255 hasAuthorship W3128176255A5075395956 @default.
- W3128176255 hasConcept C105339364 @default.
- W3128176255 hasConcept C105795698 @default.
- W3128176255 hasConcept C106189395 @default.
- W3128176255 hasConcept C111919701 @default.
- W3128176255 hasConcept C112930515 @default.
- W3128176255 hasConcept C127413603 @default.
- W3128176255 hasConcept C154945302 @default.
- W3128176255 hasConcept C159886148 @default.
- W3128176255 hasConcept C2776036281 @default.
- W3128176255 hasConcept C33923547 @default.
- W3128176255 hasConcept C41008148 @default.
- W3128176255 hasConcept C71924100 @default.
- W3128176255 hasConcept C78519656 @default.
- W3128176255 hasConcept C97541855 @default.
- W3128176255 hasConcept C98045186 @default.
- W3128176255 hasConceptScore W3128176255C105339364 @default.
- W3128176255 hasConceptScore W3128176255C105795698 @default.
- W3128176255 hasConceptScore W3128176255C106189395 @default.
- W3128176255 hasConceptScore W3128176255C111919701 @default.
- W3128176255 hasConceptScore W3128176255C112930515 @default.
- W3128176255 hasConceptScore W3128176255C127413603 @default.
- W3128176255 hasConceptScore W3128176255C154945302 @default.
- W3128176255 hasConceptScore W3128176255C159886148 @default.
- W3128176255 hasConceptScore W3128176255C2776036281 @default.
- W3128176255 hasConceptScore W3128176255C33923547 @default.
- W3128176255 hasConceptScore W3128176255C41008148 @default.
- W3128176255 hasConceptScore W3128176255C71924100 @default.
- W3128176255 hasConceptScore W3128176255C78519656 @default.
- W3128176255 hasConceptScore W3128176255C97541855 @default.
- W3128176255 hasConceptScore W3128176255C98045186 @default.
- W3128176255 hasLocation W31281762551 @default.
- W3128176255 hasOpenAccess W3128176255 @default.