Reachability-based safe learning with Gaussian processes

By Anayo K. Akametalu, Shahab Kaynama, Jaime Fernández Fisac, Melanie Nicole Zeilinger, Jeremy H. Gillula, Claire Jennifer Tomlin. 2014

313 citations recorded by the source corpus.

Subfields: Artificial Intelligence

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

Citations in the corpus

Verifiable Reinforcement Learning via Policy Extraction (2018, background): “For example, self-driving cars must robustly handle a variety of human behaviors [26], controllers for robotics typically need stability guarantees [2, 20, 8], and air trafﬁc control should provably satisfy safety properties including robustness [19].”
Safe Model-based Reinforcement Learning with Stability Guarantees (2017, methodology): “The method in [26] uses reachability analysis to construct safe regions in the state space.”
On Ensuring that Intelligent Machines Are Well-Behaved (2017, background): “…ensure (sometimes with high probability, rather than surely) that a system will never enter a predefined unsafe set of states [Tomlin, 1998; Oishi et al., 2001; Perkins and Barto, 2003; Mitchell et al., 2005; Hans et al., 2008; Arvelo and Martins, 2012; Akametalu et al., 2014; Zilberstein, 2015].”
Safe Exploration in Finite Markov Decision Processes with Gaussian Processes (2016, background): “In the control community, safety is mostly considered in terms of stability or constraint satisfaction of controlled systems.”
Safe Control under Uncertainty with Probabilistic Signal Temporal Logic (2016, background): “Designing controllers under reachability analysis and safe learning are well-studied methods that allow specifying safety and reachability properties, while learning the optimal strategy online [32, 33, 15, 3, 1].”
Safe Reinforcement Learning (2015, background): “However, if information about the transition function is known, then control theoretic approaches that leverage this knowledge may require less historical data to find good policies while enforcing various safety guarantees (Akametalu et al., 2014, Kretchmar et al., 2001, Perkins and Barto, 2003).”
Preventing undesirable behavior of intelligent machines (2019, background): “For example, control theoretic algorithms have been developed that ensure (sometimes with high probability, rather than surely) that a system will never enter a predefined unsafe set of states [57, 58, 59, 60, 61, 62, 63, 64].”
VALUATION : A N ADVERSARIAL APPROACH TO UNCOVER CATASTROPHIC FAILURES (2018, unknown): “These approaches typically rely on strong knowledge of the transition dynamics (Moldovan & Abbeel, 2012), or assumptions about the accuracy of a learned model (Berkenkamp et al., 2017; Akametalu et al., 2014), and we believe are complementary to adversarial testing, which can be viewed as a mechanis”
PAC Confidence Predictions for Deep Neural Network Classifiers (2020, background): “There has also been work on safe learning-based control (Akametalu et al., 2014; Fisac et al., 2019; Bastani, 2019; Li & Bastani, 2020; Wabersich & Zeilinger, 2018; Alshiekh et al., 2018); however, these approaches are not applicable to perception-based control.”
A Review of Safe Reinforcement Learning: Methods, Theories, and Applications (2024, unknown): “For instance, Akametalu et al. [38] develop a safe RL method based on reachability analysis, in which they use GP methods to measure the disturbances which may lead to unsafe states for agents, the maximal safe areas are computed iteratively in an unknown dynamics system.”
A Tour of Reinforcement Learning: The View from Continuous Control (2018, background): “How to use diverse sensor measurements in a safe and reliable manner remains an active and increasingly important research challenge [6, 8, 10].”
A Review of Safe Reinforcement Learning: Methods, Theory and Applications (2022, methodology): “[7] develop a safe RL method based on reachability analysis, in which they use GP methods to measure the disturbances which may lead to unsafe states for agents, the maximal safe areas are computed iteratively in an unknown dynamics system.”
Towards Verified Artificial Intelligence (2016, background): “Additionally, there has been considerable recent work on safe learning-based control (e.g., [10, 11, 2]).”
Safe controller optimization for quadrotors with Gaussian processes (2015, background): “Stability can be guaranteed by ensuring that either the controller is robustly stable for all possible models within the uncertainty specification [15] or the system never leaves a safe subset of the state space [16], [17], [18].”
Hamilton-Jacobi reachability: A brief overview and recent advances (2017, methodology): “In [56], [57], a general safety framework is proposed based on HJ reachability methods that can work in conjunction with an arbitrary learning algorithm.”
A General Safety Framework for Learning-Based Control in Uncertain Robotic Systems (2017, background): “This paper consolidates the preliminary work presented in [26]–[28], extending the theoretical results to provide a unified treatment of model learning and guarantee validation, and”
Safe Control under Uncertainty (2015, background): “More recently, safe learning approaches construct controllers that keep the system in the safe region, while the optimal strategy is learned online [15, 3, 1].”
Bayesian optimization with safety constraints: safe and automatic parameter tuning in robotics (2016, unknown): “…stable for all possible models within the uncertainty speciﬁcation (Berkenkamp and Schoellig 2015; Berkenkamp et al. 2017) or the system never leaves a safe subset of the state space (Ostafew et al. 2016; Aswani et al. 2013; Akametalu et al. 2014; Moldovan and Abbeel 2012; Turchetta et al. 2016).”
Reach-avoid problems with time-varying dynamics, targets and constraints (2014, background): “In many important application contexts, such as online safety analysis in dynamical systems [1], the substantial reduction in computational cost introduced by our technique can allow timely obtention of results that would otherwise entail an impractical computational effort.”