A comprehensive survey on safe reinforcement learning

By Javier García, Fernando Fernández. 2015

2,055 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Control and Systems Engineering, Industrial and Manufacturing Engineering, Reinforcement Learning, Safety, Robustness & Interpretability

How researchers see this paper

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

In Artificial Intelligence

In Control and Systems Engineering

In Industrial and Manufacturing Engineering

In Reinforcement Learning

In Safety, Robustness & Interpretability

Citations in the corpus

Scalable agent alignment via reward modeling: a research direction (2018, background): “First, for complex tasks there are always parts of the environment that are unknown and the agent needs to explore them safely (García & Fernández, 2015).”
Constrained Policy Optimization (2017, background): “Safety has long been a topic of interest in RL research, and a comprehensive overview of safety in RL was given by (Garc´ıa & Fern´andez, 2015).”
True Few-Shot Learning with Language Models (2021, methodology): “We evaluate the generalization loss of the algorithm chosen by CV (likewise for MDL): where Reliability is important for many practical or commercial applications where worst-case performance is important, such as image recognition [35, 36], dialogue systems [37, 38], and robotics [39, 40].”
Concrete Problems in AI Safety (2016, background): “• Risk-Sensitive Performance Criteria: A body of existing literature considers changing the optimization criteria from expected total reward to other objectives that are better at preventing rare, catastrophic events; see [53] for a thorough and up-to-date review of this literature.”
Deep Reinforcement Learning for Autonomous Driving: A Survey (2020, methodology): “ure. The SORL model was found to be not sensitive to reward function and can use different DRL algorithms like DDPG. Furthermore, a comprehensive survey on safe reinforcement learning can be found in [148] for interested readers. G. Open-source frameworks Reinforcement learning results are usually d”
Reinforcement Learning in Healthcare: A Survey (2019, background): “patience has been given a fatal medical treatment. Consequently, methods for safe exploration are of great real world interest in medical settings, in order to preclude unwanted, unsafe actions [32], [33], [325]. F. Credit Assignment Another important aspect of RL is the credit assignment problem th”
Verifiable Reinforcement Learning via Policy Extraction (2018, background): “Specific to reinforcement learning, there has been substantial interest in safe exploration [23, 36, 33]; see [14] for a survey.”
Safe Model-based Reinforcement Learning with Stability Guarantees (2017, background): “Related work Safety is an active research topic in RL and different definitions of safety exist [5, 6].”
EPOpt: Learning Robust Neural Network Policies Using Model Ensembles (2016, methodology): “This simple model adaptation rule allows us to illustrate the utility of EPOpt for robust policy search, as well as its integration with model adaptation to learn policies in cases where the target model could be very different from the initially assumed distribution.”
Batch Policy Learning under Constraints (2019, background): “A simple example is safe policy learning (Garcıa & Fernández, 2015).”
Efficient Exploration Through Bayesian Deep Q-Networks (2018, background): “One can use this distribution and come up with a safe RL criterion for the agent [29].”
Go-Explore: a New Approach for Hard-Exploration Problems (2019, methodology): “ibution of states that contain a particular agent (e.g. a game character or robot) or are near to catastrophic outcomes. For example, within AI safety [5] one open problem is that of safe exploration [83], wherein the process of training an effective real-world policy is constrained by avoiding cata”
AI Safety Gridworlds (2017, background): “How can we build agents that respect the safety constraints not only during normal operation, but also during the initial learning period? This problem is known as the safe exploration problem (Pecka and Svoboda, 2014; Garcı́a and Fernández, 2015).”
Trial without Error: Towards Safe Reinforcement Learning via Human Intervention (2017, background): “The existing literature contains a variety of deﬁnitions of “safe RL” [12].”
Sim-to-Real Transfer in Deep Reinforcement Learning for Robotics: a Survey (2020, background): “Safety is one of the main challenges towards achieving online training of complex agents in the real-world, from robot arms to self-driving cars [4].”
A review of cooperative multi-agent deep reinforcement learning (2019, positive): “See a comprehensive review in [40].”
Explainability in Deep Reinforcement Learning (2020, background): “From an XRL point of view, the explainability comes from the fact that the agent is able to express its actions as boolean formulas, which are easily readable by humans.”