Safe exploration for reinforcement learning

By Alexander Hans, Daniel Schneegaß, Anton Maximilian Schäfer, Steffen Udluft. 2008

146 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Safe Model-based Reinforcement Learning with Stability Guarantees (2017, background): “In [20, 21], a priori known, safe global backup policies are used, while [22] learns to switch between several safe policies.”
Safe Exploration in Markov Decision Processes (2012, background): “The safe exploration for MDP methods proposed by (Geramifard et al., 2011; Hans et al., 2008) gauge safety based on the best best estimate of the transition measure but they ignore the level of uncertainty in this estimate.”
AI Safety Gridworlds (2017, background): “There are several possible formulations for safe exploration: being able to return to the starting state or some other safe state with high probability (Moldovan and Abbeel, 2012), achieving a minimum reward (Hans et al., 2008), or satisfying a given side constraint (Turchetta et al., 2016).”
Fairness for Unobserved Characteristics: Insights from Technological Impacts on Queer Communities (2021, background): “Furthermore, reinforcement learning systems for behavioural interventions will present risks to patients unless many open problems in the ﬁeld can be resolved, such as safe exploration (Hans et al. 2008) and reward speciﬁcation (Krakovna et al. 2020).”
Offline Reinforcement Learning from Human Feedback in Real-World Sequence-to-Sequence Tasks (2020, background): “While this exploration is limited by the input data, solutions for safe exploration might be attractive to transfer to NLP applications to actively guide exploration while not sacriﬁcing quality (Hans et al., 2008; Berkenkamp et al., 2017).”
Agent-Agnostic Human-in-the-Loop Reinforcement Learning (2017, methodology): “In Section 5, we apply action-pruning to prevent catastrophic outcomes during exploration, a problem explored by Lipton et al. [27], Garcia and Fernandez [14, 15], Hans et al. [18], Moldovan and Abbeel [31].”
Combating Deep Reinforcement Learning ’ s Sisyphean Curse with Reinforcement Learning (2017, background): “Hans et al. [2008] defines a fatality as any return below some threshold τ .”
A Review of Safe Reinforcement Learning: Methods, Theories, and Applications (2024, unknown): “The third type of safety deﬁnition: according to Hans et al. [5], humans need to label environmental states as “safe” or “unsafe,” and agents are considered “safe” if “they never reach unsafe states”.”
An exploration strategy for non-stationary opponents (2017, background): “In this cases, a hard-reset like we propose may not be the best approach and we leave that as an open question where ideas from safe-exploration [22,24] or social reward shaping [3] might be worth exploring.”
Safe Exploration in Finite Markov Decision Processes with Gaussian Processes (2016, methodology): “Another approach that makes use of a backup policy is shown by Hans et al. [9], where safety is defined in terms of a minimum reward, which is learned from data.”
Safe Exploration of State and Action Spaces in Reinforcement Learning (2012, background): “As a result, learner performance is heavily limited by the quality of the teacher’s demonstrations.”
Decomposition of Uncertainty in Bayesian Deep Learning for Efficient and Risk-sensitive Learning (2017, background): “In safe RL numerous other approaches exists for safe exploration (Joseph et al., 2013; Hans et al., 2008; Garcia & Fernández, 2012; Berkenkamp et al., 2017).”
Safe Exploration for Optimization with Gaussian Processes (2015, background): “The problem of safe exploration has been considered in control and reinforcement learning (Hans et al., 2008; Gillula & Tomlin, 2011; Garcia & Fernandez, 2012).”
Combating Reinforcement Learning's Sisyphean Curse with Intrinsic Fear (2016, background): “[10] deﬁnes a fatality as a return below some threshold τ .”
A comprehensive survey on safe reinforcement learning (2015, background): “In the level-based exploration proposed by Hans et al. (2008), the agent tries to explore all actions considered safe (i.e., it does not lead to a fatal transition) for each state gradually.”
Safe Exploration Algorithms for Reinforcement Learning Controllers (2018, methodology): “Among algorithms that directly avoid unsafe transitions, [30] relies on an a priori known safety function (acting as a go/no-go decision maker over feasible actions) and a ﬁxed backup policy valid in all workspace.”
Bayesian optimization with safety constraints: safe and automatic parameter tuning in robotics (2016, unknown): “Without a known model, safety in gradient-based policy search has been considered by disallowing exploration into areas of the state space that have not been visited before (Garcia & Fernández, 2012; Hans et al., 2008).”
Combating Deep Reinforcement Learning's Sisyphean Curse with Intrinsic Fear (2016, background): “Hans et al. [2008] deﬁnes a fatality as any return below some threshold τ .”