Off-policy Evaluation (OPE) methods are crucial tools for evaluating policies in high-stakes domains such as healthcare, where direct deployment is often infeasible, unethical, or expensive. When deployment environments are expected to undergo changes (that is, dataset shifts), it is important for OPE methods to perform robust evaluation of the policies amidst such changes. Existing approaches consider robustness against a large class of shifts that can arbitrarily change any observable property of the environment. This often results in highly pessimistic estimates of the utilities, thereby invalidating policies that might have been useful in deployment. In this work, we address the aforementioned problem by investigating how domain knowledge can help provide more realistic estimates of the utilities of policies. We leverage human inputs on which aspects of the environments may plausibly change, and adapt the OPE methods to only consider shifts on these aspects. Specifically, we propose a novel framework, Robust OPE (ROPE), which considers shifts on a subset of covariates in the data based on user inputs, and estimates worst-case utility under these shifts. We then develop computationally efficient algorithms for OPE that are robust to the aforementioned shifts for contextual bandits and Markov decision processes. We also theoretically analyze the sample complexity of these algorithms. Extensive experimentation with synthetic and real world datasets from the healthcare domain demonstrates that our approach not only captures realistic dataset shifts accurately, but also results in less pessimistic policy evaluations.

New research published in AI Magazine explores how advanced AI could hack reward systems to dangerous effect. Researchers at the University of Oxford and Australian National University analyzed the behavior of future advanced reinforcement learning (RL) agents, which take actions, observe rewards, learn how their rewards depend on their actions, and pick actions to maximize expected future rewards. As RL agents get more advanced, they are better able to recognize and execute action plans that cause more expected reward, even in contexts where reward is only received after impressive feats. Lead author Michael K. Cohen says, "Our key insight was that advanced RL agents will have to question how their rewards depend on their actions." Answers to that question are called world-models.

This paper addresses the problem of developing an algorithm for autonomous ship landing of vertical take-off and landing (VTOL) capable unmanned aerial vehicles (UAVs), using only a monocular camera in the UAV for tracking and localization. Ship landing is a challenging task due to the small landing space, six degrees of freedom ship deck motion, limited visual references for localization, and adversarial environmental conditions such as wind gusts. We first develop a computer vision algorithm which estimates the relative position of the UAV with respect to a horizon reference bar on the landing platform using the image stream from a monocular vision camera on the UAV. Our approach is motivated by the actual ship landing procedure followed by the Navy helicopter pilots in tracking the horizon reference bar as a visual cue. We then develop a robust reinforcement learning (RL) algorithm for controlling the UAV towards the landing platform even in the presence of adversarial environmental conditions such as wind gusts. We demonstrate the superior performance of our algorithm compared to a benchmark nonlinear PID control approach, both in the simulation experiments using the Gazebo environment and in the real-world setting using a Parrot ANAFI quad-rotor and sub-scale ship platform undergoing 6 degrees of freedom (DOF) deck motion.

Distributional reinforcement learning~(RL) is a class of state-of-the-art algorithms that estimate the whole distribution of the total return rather than only its expectation. Despite the remarkable performance of distributional RL, a theoretical understanding of its advantages over expectation-based RL remains elusive. In this paper, we attribute the superiority of distributional RL to its regularization effect in terms of the value distribution information regardless of its expectation. Firstly, by leverage of a variant of the gross error model in robust statistics, we decompose the value distribution into its expectation and the remaining distribution part. As such, the extra benefit of distributional RL compared with expectation-based RL is mainly interpreted as the impact of a \textit{risk-sensitive entropy regularization} within the Neural Fitted Z-Iteration framework. Meanwhile, we establish a bridge between the risk-sensitive entropy regularization of distributional RL and the vanilla entropy in maximum entropy RL, focusing specifically on actor-critic algorithms. It reveals that distributional RL induces a corrected reward function and thus promotes a risk-sensitive exploration against the intrinsic uncertainty of the environment. Finally, extensive experiments corroborate the role of the regularization effect of distributional RL and uncover mutual impacts of different entropy regularization. Our research paves a way towards better interpreting the efficacy of distributional RL algorithms, especially through the lens of regularization.

Flocking control is a challenging problem, where multiple agents, such as drones or vehicles, need to reach a target position while maintaining the flock and avoiding collisions with obstacles and collisions among agents in the environment. Multi-agent reinforcement learning has achieved promising performance in flocking control. However, methods based on traditional reinforcement learning require a considerable number of interactions between agents and the environment. This paper proposes a sub-optimal policy aided multi-agent reinforcement learning algorithm (SPA-MARL) to boost sample efficiency. SPA-MARL directly leverages a prior policy that can be manually designed or solved with a non-learning method to aid agents in learning, where the performance of the policy can be sub-optimal. SPA-MARL recognizes the difference in performance between the sub-optimal policy and itself, and then imitates the sub-optimal policy if the sub-optimal policy is better. We leverage SPA-MARL to solve the flocking control problem. A traditional control method based on artificial potential fields is used to generate a sub-optimal policy. Experiments demonstrate that SPA-MARL can speed up the training process and outperform both the MARL baseline and the used sub-optimal policy.

Flocking control is a significant problem in multi-agent systems such as multi-agent unmanned aerial vehicles and multi-agent autonomous underwater vehicles, which enhances the cooperativity and safety of agents. In contrast to traditional methods, multi-agent reinforcement learning (MARL) solves the problem of flocking control more flexibly. However, methods based on MARL suffer from sample inefficiency, since they require a huge number of experiences to be collected from interactions between agents and the environment. We propose a novel method Pretraining with Demonstrations for MARL (PwD-MARL), which can utilize non-expert demonstrations collected in advance with traditional methods to pretrain agents. During the process of pretraining, agents learn policies from demonstrations by MARL and behavior cloning simultaneously, and are prevented from overfitting demonstrations. By pretraining with non-expert demonstrations, PwD-MARL improves sample efficiency in the process of online MARL with a warm start. Experiments show that PwD-MARL improves sample efficiency and policy performance in the problem of flocking control, even with bad or few demonstrations.

Though deep reinforcement learning (DRL) has obtained substantial success, it may encounter catastrophic failures due to the intrinsic uncertainty of both transition and observation. Most of the existing methods for safe reinforcement learning can only handle transition disturbance or observation disturbance since these two kinds of disturbance affect different parts of the agent; besides, the popular worst-case return may lead to overly pessimistic policies. To address these issues, we first theoretically prove that the performance degradation under transition disturbance and observation disturbance depends on a novel metric of Value Function Range (VFR), which corresponds to the gap in the value function between the best state and the worst state. Based on the analysis, we adopt conditional value-at-risk (CVaR) as an assessment of risk and propose a novel reinforcement learning algorithm of CVaR-Proximal-Policy-Optimization (CPPO) which formalizes the risk-sensitive constrained optimization problem by keeping its CVaR under a given threshold. Experimental results show that CPPO achieves a higher cumulative reward and is more robust against both observation and transition disturbances on a series of continuous control tasks in MuJoCo.

Reinforcement learning is one subfield of machine learning. It involves acting appropriately to maximize reward in a particular circumstance. It is used by various programs and machines to determine the optimal course of action to pursue in a given case. Reinforcement learning has no right or wrong solution; instead, the reinforcement agent decides what to do to finish the task. This differs from supervised learning, where the training data includes the solution key, and the model is trained with that answer.

A trustworthy reinforcement learning algorithm should be competent in solving challenging real-world problems, including {robustly} handling uncertainties, satisfying {safety} constraints to avoid catastrophic failures, and {generalizing} to unseen scenarios during deployments. This study aims to overview these main perspectives of trustworthy reinforcement learning considering its intrinsic vulnerabilities on robustness, safety, and generalizability. In particular, we give rigorous formulations, categorize corresponding methodologies, and discuss benchmarks for each perspective. Moreover, we provide an outlook section to spur promising future directions with a brief discussion on extrinsic vulnerabilities considering human feedback. We hope this survey could bring together separate threads of studies together in a unified framework and promote the trustworthiness of reinforcement learning.

With the wide spread of sensors and smart devices in recent years, the data generation speed of the Internet of Things (IoT) systems has increased dramatically. In IoT systems, massive volumes of data must be processed, transformed, and analyzed on a frequent basis to enable various IoT services and functionalities. Machine Learning (ML) approaches have shown their capacity for IoT data analytics. However, applying ML models to IoT data analytics tasks still faces many difficulties and challenges, specifically, effective model selection, design/tuning, and updating, which have brought massive demand for experienced data scientists. Additionally, the dynamic nature of IoT data may introduce concept drift issues, causing model performance degradation. To reduce human efforts, Automated Machine Learning (AutoML) has become a popular field that aims to automatically select, construct, tune, and update machine learning models to achieve the best performance on specified tasks. In this paper, we conduct a review of existing methods in the model selection, tuning, and updating procedures in the area of AutoML in order to identify and summarize the optimal solutions for every step of applying ML algorithms to IoT data analytics. To justify our findings and help industrial users and researchers better implement AutoML approaches, a case study of applying AutoML to IoT anomaly detection problems is conducted in this work. Lastly, we discuss and classify the challenges and research directions for this domain.