-
Risk-Aware Robotics: Tail Risk Measures in Planning, Control, and Verification
Authors:
Prithvi Akella,
Anushri Dixit,
Mohamadreza Ahmadi,
Lars Lindemann,
Margaret P. Chapman,
George J. Pappas,
Aaron D. Ames,
Joel W. Burdick
Abstract:
The need for a systematic approach to risk assessment has increased in recent years due to the ubiquity of autonomous systems that alter our day-to-day experiences and their need for safety, e.g., for self-driving vehicles, mobile service robots, and bipedal robots. These systems are expected to function safely in unpredictable environments and interact seamlessly with humans, whose behavior is no…
▽ More
The need for a systematic approach to risk assessment has increased in recent years due to the ubiquity of autonomous systems that alter our day-to-day experiences and their need for safety, e.g., for self-driving vehicles, mobile service robots, and bipedal robots. These systems are expected to function safely in unpredictable environments and interact seamlessly with humans, whose behavior is notably challenging to forecast. We present a survey of risk-aware methodologies for autonomous systems. We adopt a contemporary risk-aware approach to mitigate rare and detrimental outcomes by advocating the use of tail risk measures, a concept borrowed from financial literature. This survey will introduce these measures and explain their relevance in the context of robotic systems for planning, control, and verification applications.
△ Less
Submitted 27 March, 2024;
originally announced March 2024.
-
Risk-Aware Stability of Discrete-Time Systems
Authors:
Margaret P. Chapman,
Dionysios S. Kalogerias
Abstract:
We develop a generalized stability framework for stochastic discrete-time systems, where the generality pertains to the ways in which the distribution of the state energy can be characterized. We use tools from finance and operations research called risk functionals (i.e., risk measures) to facilitate diverse distributional characterizations. In contrast, classical stochastic stability notions cha…
▽ More
We develop a generalized stability framework for stochastic discrete-time systems, where the generality pertains to the ways in which the distribution of the state energy can be characterized. We use tools from finance and operations research called risk functionals (i.e., risk measures) to facilitate diverse distributional characterizations. In contrast, classical stochastic stability notions characterize the state energy on average or in probability, which can obscure the variability of stochastic system behavior. After drawing connections between various risk-aware stability concepts for nonlinear systems, we specialize to linear systems and derive sufficient conditions for the satisfaction of some risk-aware stability properties. These results pertain to real-valued coherent risk functionals and a mean-conditional-variance functional. The results reveal novel noise-to-state stability properties, which assess disturbances in ways that reflect the chosen measure of risk. We illustrate the theory through examples about robustness, parameter choices, and state-feedback controllers.
△ Less
Submitted 22 November, 2022;
originally announced November 2022.
-
Information-theoretic multi-time-scale partially observable systems with inspiration from leukemia treatment
Authors:
Margaret P. Chapman,
Emily Jensen,
Steven M. Chan,
Laurent Lessard
Abstract:
We study a partially observable nonlinear stochastic system with unknown parameters, where the given time scales of the states and measurements may be distinct. The proposed setting is inspired by disease management, particularly leukemia.
We study a partially observable nonlinear stochastic system with unknown parameters, where the given time scales of the states and measurements may be distinct. The proposed setting is inspired by disease management, particularly leukemia.
△ Less
Submitted 5 February, 2024; v1 submitted 26 April, 2022;
originally announced April 2022.
-
Toward Scalable Risk Analysis for Stochastic Systems Using Extreme Value Theory
Authors:
Evan Arsenault,
Yuheng Wang,
Margaret P. Chapman
Abstract:
We aim to analyze the behaviour of a finite-time stochastic system, whose model is not available, in the context of more rare and harmful outcomes. Standard estimators are not effective in making predictions about such outcomes due to their rarity. Instead, we use Extreme Value Theory (EVT), the theory of the long-term behaviour of normalized maxima of random variables. We quantify risk using the…
▽ More
We aim to analyze the behaviour of a finite-time stochastic system, whose model is not available, in the context of more rare and harmful outcomes. Standard estimators are not effective in making predictions about such outcomes due to their rarity. Instead, we use Extreme Value Theory (EVT), the theory of the long-term behaviour of normalized maxima of random variables. We quantify risk using the upper-semideviation $ρ(Y) = E(\max\{Y - μ,0\})$ of an integrable random variable $Y$ with mean $μ= E(Y)$. $ρ(Y)$ is the risk-aware part of the common mean-upper-semideviation functional $μ+ λρ(Y)$ with $λ\in [0,1]$. To assess more rare and harmful outcomes, we propose an EVT-based estimator for $ρ(Y)$ in a given fraction of the worst cases. We show that our estimator enjoys a closed-form representation in terms of the popular conditional value-at-risk functional. In experiments, we illustrate the extrapolation power of our estimator using a small number of i.i.d. samples ($<$50). Our approach is useful for estimating the risk of finite-time systems when models are inaccessible and data collection is expensive. The numerical complexity does not grow with the size of the state space.
△ Less
Submitted 21 June, 2022; v1 submitted 23 March, 2022;
originally announced March 2022.
-
A Non-linear Differentiable Model for Stormwater-based Irrigation of a Green Roof in Toronto
Authors:
Chia-Hui Yeh,
Margaret P. Chapman
Abstract:
Green infrastructure has potential to alleviate the environmental impact of rapidly growing cities. This potential has inspired laws in Toronto that require the inclusion of rooftops with large vegetation beds, called green roofs, into sufficiently sized construction projects. We study the problem of reusing stormwater to irrigate a green roof in Toronto, where potable water is the current irrigat…
▽ More
Green infrastructure has potential to alleviate the environmental impact of rapidly growing cities. This potential has inspired laws in Toronto that require the inclusion of rooftops with large vegetation beds, called green roofs, into sufficiently sized construction projects. We study the problem of reusing stormwater to irrigate a green roof in Toronto, where potable water is the current irrigation source. The vision is that widespread reuse of stormwater runoff for irrigation of green roofs and other purposes can reduce sewer overflow volumes without over-building (with the added benefit of conserving potable water). Towards this vision, our goal is to develop and evaluate two pump controllers for transporting stormwater to the green roof of interest in simulation. A key contribution is our development of a site-specific non-linear model for stormwater flow using smoothing techniques that permits linearization and a standard model predictive controller (MPC). We compare the efficacy of the MPC, which anticipates the weather, and an on/off controller, which is reactive rather than anticipative, for the site in simulation. With further study, we are hopeful that this research will advance control systems technology to improve the performance of green and stormwater infrastructure in growing urban areas.
△ Less
Submitted 26 October, 2021;
originally announced October 2021.
-
Risk-averse autonomous systems: A brief history and recent developments from the perspective of optimal control
Authors:
Yuheng Wang,
Margaret P. Chapman
Abstract:
We present an historical overview about the connections between the analysis of risk and the control of autonomous systems. We offer two main contributions. Our first contribution is to propose three overlap** paradigms to classify the vast body of literature: the worst-case, risk-neutral, and risk-averse paradigms. We consider an appropriate assessment for the risk of an autonomous system to de…
▽ More
We present an historical overview about the connections between the analysis of risk and the control of autonomous systems. We offer two main contributions. Our first contribution is to propose three overlap** paradigms to classify the vast body of literature: the worst-case, risk-neutral, and risk-averse paradigms. We consider an appropriate assessment for the risk of an autonomous system to depend on the application at hand. In contrast, it is typical to assess risk using an expectation, variance, or probability alone. Our second contribution is to unify the concepts of risk and autonomous systems. We achieve this by connecting approaches for quantifying and optimizing the risk that arises from a system's behaviour across academic fields. The survey is highly multidisciplinary. We include research from the communities of reinforcement learning, stochastic and robust control theory, operations research, and formal verification. We describe both model-based and model-free methods, with emphasis on the former. Lastly, we highlight fruitful areas for further research. A key direction is to blend risk-averse model-based and model-free methods to enhance the real-time adaptive capabilities of systems to improve human and environmental welfare.
△ Less
Submitted 11 July, 2022; v1 submitted 18 September, 2021;
originally announced September 2021.
-
CVaR-based Safety Analysis in the Infinite Time Horizon Setting
Authors:
Chuanning Wei,
Michael Fauss,
Margaret P. Chapman
Abstract:
We develop a risk-averse safety analysis method for stochastic systems on discrete infinite time horizons. Our method quantifies the notion of risk for a control system in terms of the severity of a harmful random outcome in a fraction of the worst cases. In contrast, classical methods quantify risk in terms of the probability of a harmful event. Our theoretical arguments are based on the analysis…
▽ More
We develop a risk-averse safety analysis method for stochastic systems on discrete infinite time horizons. Our method quantifies the notion of risk for a control system in terms of the severity of a harmful random outcome in a fraction of the worst cases. In contrast, classical methods quantify risk in terms of the probability of a harmful event. Our theoretical arguments are based on the analysis of a value iteration algorithm on an augmented state space. We provide conditions to guarantee the existence of an optimal policy on this space. We illustrate the method numerically using an example from the domain of stormwater management.
△ Less
Submitted 10 March, 2022; v1 submitted 15 August, 2021;
originally announced August 2021.
-
On Exponential Utility and Conditional Value-at-Risk as Risk-Averse Performance Criteria
Authors:
Kevin M. Smith,
Margaret P. Chapman
Abstract:
The standard approach to risk-averse control is to use the Exponential Utility (EU) functional, which has been studied for several decades. Like other risk-averse utility functionals, EU encodes risk aversion through an increasing convex map** $\varphi$ of objective costs to subjective costs. An objective cost is a realization $y$ of a random variable $Y$. In contrast, a subjective cost is a rea…
▽ More
The standard approach to risk-averse control is to use the Exponential Utility (EU) functional, which has been studied for several decades. Like other risk-averse utility functionals, EU encodes risk aversion through an increasing convex map** $\varphi$ of objective costs to subjective costs. An objective cost is a realization $y$ of a random variable $Y$. In contrast, a subjective cost is a realization $\varphi(y)$ of a random variable $\varphi(Y)$ that has been transformed to measure preferences about the outcomes. For EU, the transformation is $\varphi(y) = \exp(\frac{-θ}{2}y)$, and under certain conditions, the quantity $\varphi^{-1}(E(\varphi(Y)))$ can be approximated by a linear combination of the mean and variance of $Y$. More recently, there has been growing interest in risk-averse control using the Conditional Value-at-Risk (CVaR) functional. In contrast to the EU functional, the CVaR of a random variable $Y$ concerns a fraction of its possible realizations. If $Y$ is a continuous random variable with finite $E(|Y|)$, then the CVaR of $Y$ at level $α$ is the expectation of $Y$ in the $α\cdot 100 \%$ worst cases. Here, we study the applications of risk-averse functionals to controller synthesis and safety analysis through the development of numerical examples, with emphasis on EU and CVaR. Our contribution is to examine the decision-theoretic, mathematical, and computational trade-offs that arise when using EU and CVaR for optimal control and safety analysis. We are hopeful that this work will advance the interpretability and elucidate the potential benefits of risk-averse control technology.
△ Less
Submitted 5 May, 2023; v1 submitted 3 August, 2021;
originally announced August 2021.
-
Classical Risk-Averse Control for a Finite-Horizon Borel Model
Authors:
Margaret P. Chapman,
Kevin M. Smith
Abstract:
We study a risk-averse optimal control problem for a finite-horizon Borel model, where a cumulative cost is assessed via exponential utility. The setting permits non-linear dynamics, non-quadratic costs, and continuous state and control spaces but is less general than the problem of optimizing an expected utility. Our contribution is to show the existence of an optimal risk-averse controller witho…
▽ More
We study a risk-averse optimal control problem for a finite-horizon Borel model, where a cumulative cost is assessed via exponential utility. The setting permits non-linear dynamics, non-quadratic costs, and continuous state and control spaces but is less general than the problem of optimizing an expected utility. Our contribution is to show the existence of an optimal risk-averse controller without using state space augmentation and therefore offer a simpler solution method from first principles compared to what is currently available in the literature.
△ Less
Submitted 25 June, 2022; v1 submitted 29 July, 2021;
originally announced July 2021.
-
On Optimizing the Conditional Value-at-Risk of a Maximum Cost for Risk-Averse Safety Analysis
Authors:
Margaret P. Chapman,
Michael Fauss,
Kevin M. Smith
Abstract:
The popularity of Conditional Value-at-Risk (CVaR), a risk functional from finance, has been growing in the control systems community due to its intuitive interpretation and axiomatic foundation. We consider a nonstandard optimal control problem in which the goal is to minimize the CVaR of a maximum random cost subject to a Borel-space Markov decision process. The objective represents the maximum…
▽ More
The popularity of Conditional Value-at-Risk (CVaR), a risk functional from finance, has been growing in the control systems community due to its intuitive interpretation and axiomatic foundation. We consider a nonstandard optimal control problem in which the goal is to minimize the CVaR of a maximum random cost subject to a Borel-space Markov decision process. The objective represents the maximum departure from a desired operating region averaged over a given fraction of the worst cases. This problem provides a safety criterion for a stochastic system that is informed by both the probability and severity of the potential consequences of the system's behavior. In contrast, existing safety analysis frameworks apply stage-wise risk constraints or assess the probability of constraint violation without quantifying the potential severity of the violation. To the best of our knowledge, the problem of interest has not been solved. To solve the problem, we propose and study a family of stochastic dynamic programs on an augmented state space. We prove that the optimal CVaR of a maximum random cost enjoys an equivalent representation in terms of the solutions to these dynamic programs under appropriate assumptions. For each dynamic program, we show the existence of an optimal policy that depends on the dynamics of an augmented state under the assumptions. In a numerical example, we illustrate how our safety analysis framework is useful for assessing the severity of combined sewer overflows under precipitation uncertainty.
△ Less
Submitted 19 June, 2022; v1 submitted 1 June, 2021;
originally announced June 2021.
-
Toward a Scalable Upper Bound for a CVaR-LQ Problem
Authors:
Margaret P. Chapman,
Laurent Lessard
Abstract:
We study a linear-quadratic, optimal control problem on a discrete, finite time horizon with distributional ambiguity, in which the cost is assessed via Conditional Value-at-Risk (CVaR). We take steps toward deriving a scalable dynamic programming approach to upper-bound the optimal value function for this problem. This dynamic program yields a novel, tunable risk-averse control policy, which we c…
▽ More
We study a linear-quadratic, optimal control problem on a discrete, finite time horizon with distributional ambiguity, in which the cost is assessed via Conditional Value-at-Risk (CVaR). We take steps toward deriving a scalable dynamic programming approach to upper-bound the optimal value function for this problem. This dynamic program yields a novel, tunable risk-averse control policy, which we compare to existing state-of-the-art methods.
△ Less
Submitted 25 June, 2022; v1 submitted 2 March, 2021;
originally announced March 2021.
-
Risk-sensitive safety analysis using Conditional Value-at-Risk
Authors:
Margaret P. Chapman,
Riccardo Bonalli,
Kevin M. Smith,
Insoon Yang,
Marco Pavone,
Claire J. Tomlin
Abstract:
This paper develops a safety analysis method for stochastic systems that is sensitive to the possibility and severity of rare harmful outcomes. We define risk-sensitive safe sets as sub-level sets of the solution to a non-standard optimal control problem, where a random maximum cost is assessed via Conditional Value-at-Risk (CVaR). The objective function represents the maximum extent of constraint…
▽ More
This paper develops a safety analysis method for stochastic systems that is sensitive to the possibility and severity of rare harmful outcomes. We define risk-sensitive safe sets as sub-level sets of the solution to a non-standard optimal control problem, where a random maximum cost is assessed via Conditional Value-at-Risk (CVaR). The objective function represents the maximum extent of constraint violation of the state trajectory, averaged over a given percentage of worst cases. This problem is well-motivated but difficult to solve tractably because the temporal decomposition for CVaR is history-dependent. Our primary theoretical contribution is to derive computationally tractable under-approximations to risk-sensitive safe sets. Our method provides a novel, theoretically guaranteed, parameter-dependent upper bound to the CVaR of a maximum cost without the need to augment the state space. For a fixed parameter value, the solution to only one Markov decision process problem is required to obtain the under-approximations for any family of risk-sensitivity levels. In addition, we propose a second definition for risk-sensitive safe sets and provide a tractable method for their estimation without using a parameter-dependent upper bound. The second definition is expressed in terms of a new coherent risk functional, which is inspired by CVaR. We demonstrate our primary theoretical contribution via numerical examples.
△ Less
Submitted 25 June, 2022; v1 submitted 28 January, 2021;
originally announced January 2021.
-
Risk-sensitive safety specifications for stochastic systems using Conditional Value-at-Risk
Authors:
Margaret P. Chapman,
Jonathan P. Lacotte,
Kevin M. Smith,
Insoon Yang,
Yuxi Han,
Marco Pavone,
Claire J. Tomlin
Abstract:
This paper proposes a safety analysis method that facilitates a tunable balance between the worst-case and risk-neutral perspectives. First, we define a risk-sensitive safe set to specify the degree of safety attained by a stochastic system. This set is defined as a sublevel set of the solution to an optimal control problem that is expressed using the Conditional Value-at-Risk (CVaR) measure. This…
▽ More
This paper proposes a safety analysis method that facilitates a tunable balance between the worst-case and risk-neutral perspectives. First, we define a risk-sensitive safe set to specify the degree of safety attained by a stochastic system. This set is defined as a sublevel set of the solution to an optimal control problem that is expressed using the Conditional Value-at-Risk (CVaR) measure. This problem does not satisfy Bellman's Principle, thus our next contribution is to show how risk-sensitive safe sets can be under-approximated by the solution to a CVaR-Markov Decision Process. We adopt an existing value iteration algorithm to find an approximate solution to the reduced problem for a class of linear systems. Then, we develop a realistic numerical example of a stormwater system to show that this approach can be applied to non-linear systems. Finally, we compare the CVaR criterion to the exponential disutility criterion. The latter allocates control effort evenly across the cost distribution to reduce variance, while the CVaR criterion focuses control effort on a given worst-case quantile--where it matters most for safety.
△ Less
Submitted 27 July, 2020; v1 submitted 20 September, 2019;
originally announced September 2019.
-
A Risk-Sensitive Finite-Time Reachability Approach for Safety of Stochastic Dynamic Systems
Authors:
Margaret P. Chapman,
Jonathan Lacotte,
Aviv Tamar,
Donggun Lee,
Kevin M. Smith,
Victoria Cheng,
Jaime F. Fisac,
Susmit Jha,
Marco Pavone,
Claire J. Tomlin
Abstract:
A classic reachability problem for safety of dynamic systems is to compute the set of initial states from which the state trajectory is guaranteed to stay inside a given constraint set over a given time horizon. In this paper, we leverage existing theory of reachability analysis and risk measures to devise a risk-sensitive reachability approach for safety of stochastic dynamic systems under non-ad…
▽ More
A classic reachability problem for safety of dynamic systems is to compute the set of initial states from which the state trajectory is guaranteed to stay inside a given constraint set over a given time horizon. In this paper, we leverage existing theory of reachability analysis and risk measures to devise a risk-sensitive reachability approach for safety of stochastic dynamic systems under non-adversarial disturbances over a finite time horizon. Specifically, we first introduce the notion of a risk-sensitive safe set as a set of initial states from which the risk of large constraint violations can be reduced to a required level via a control policy, where risk is quantified using the Conditional Value-at-Risk (CVaR) measure. Second, we show how the computation of a risk-sensitive safe set can be reduced to the solution to a Markov Decision Process (MDP), where cost is assessed according to CVaR. Third, leveraging this reduction, we devise a tractable algorithm to approximate a risk-sensitive safe set, and provide theoretical arguments about its correctness. Finally, we present a realistic example inspired from stormwater catchment design to demonstrate the utility of risk-sensitive reachability analysis. In particular, our approach allows a practitioner to tune the level of risk sensitivity from worst-case (which is typical for Hamilton-Jacobi reachability analysis) to risk-neutral (which is the case for stochastic reachability analysis).
△ Less
Submitted 30 April, 2019; v1 submitted 28 February, 2019;
originally announced February 2019.
