Safe reinforcement learning (RL) that solves constraint-satisfactory policies provides a promising way to the broader safety-critical applications of RL in real-world problems such as robotics. Among all safe RL approaches, model-based methods reduce training time violations further due to their high sample efficiency. However, lacking safety robustness against the model uncertainties remains an issue in safe model-based RL, especially in training time safety. In this paper, we propose a distributional reachability certificate (DRC) and its Bellman equation to address model uncertainties and characterize robust persistently safe states. Furthermore, we build a safe RL framework to resolve constraints required by the DRC and its corresponding shield policy. We also devise a line search method to maintain safety and reach higher returns simultaneously while leveraging the shield policy. Comprehensive experiments on classical benchmarks such as constrained tracking and navigation indicate that the proposed algorithm achieves comparable returns with much fewer constraint violations during training.

Social networks are frequently polluted by rumors, which can be detected by advanced models such as graph neural networks. However, the models are vulnerable to attacks and understanding the vulnerabilities is critical to rumor detection in practice. To discover subtle vulnerabilities, we design a powerful attacking algorithm to camouflage rumors in social networks based on reinforcement learning that can interact with and attack any black-box detectors. The environment has exponentially large state spaces, high-order graph dependencies, and delayed noisy rewards, making the state-of-the-art end-to-end approaches difficult to learn features as large learning costs and expressive limitation of graph deep models. Instead, we design domain-specific features to avoid learning features and produce interpretable attack policies. To further speed up policy optimization, we devise: (i) a credit assignment method that decomposes delayed rewards to atomic attacking actions proportional to the their camouflage effects on target rumors; (ii) a time-dependent control variate to reduce reward variance due to large graphs and many attacking steps, supported by the reward variance analysis and a Bayesian analysis of the prediction distribution. On three real world datasets of rumor detection tasks, we demonstrate: (i) the effectiveness of the learned attacking policy compared to rule-based attacks and current end-to-end approaches; (ii) the usefulness of the proposed credit assignment strategy and variance reduction components; (iii) the interpretability of the policy when generating strong attacks via the case study.

Interactive reinforcement learning can effectively facilitate the agent training via human feedback. However, such methods often require the human teacher to know what is the correct action that the agent should take. In other words, if the human teacher is not always reliable, then it will not be consistently able to guide the agent through its training. In this paper, we propose a more effective interactive reinforcement learning system by introducing multiple trainers, namely Multi-Trainer Interactive Reinforcement Learning (MTIRL), which could aggregate the binary feedback from multiple non-perfect trainers into a more reliable reward for an agent training in a reward-sparse environment. In particular, our trainer feedback aggregation experiments show that our aggregation method has the best accuracy when compared with the majority voting, the weighted voting, and the Bayesian method. Finally, we conduct a grid-world experiment to show that the policy trained by the MTIRL with the review model is closer to the optimal policy than that without a review model.

Multi-agent reinforcement learning has drawn increasing attention in practice, e.g., robotics and automatic driving, as it can explore optimal policies using samples generated by interacting with the environment. However, high reward uncertainty still remains a problem when we want to train a satisfactory model, because obtaining high-quality reward feedback is usually expensive and even infeasible. To handle this issue, previous methods mainly focus on passive reward correction. At the same time, recent active reward estimation methods have proven to be a recipe for reducing the effect of reward uncertainty. In this paper, we propose a novel Distributional Reward Estimation framework for effective Multi-Agent Reinforcement Learning (DRE-MARL). Our main idea is to design the multi-action-branch reward estimation and policy-weighted reward aggregation for stabilized training. Specifically, we design the multi-action-branch reward estimation to model reward distributions on all action branches. Then we utilize reward aggregation to obtain stable updating signals during training. Our intuition is that consideration of all possible consequences of actions could be useful for learning policies. The superiority of the DRE-MARL is demonstrated using benchmark multi-agent scenarios, compared with the SOTA baselines in terms of both effectiveness and robustness.

5G and beyond mobile networks will support heterogeneous use cases at an unprecedented scale, thus demanding automated control and optimization of network functionalities customized to the needs of individual users. Such fine-grained control of the Radio Access Network (RAN) is not possible with the current cellular architecture. To fill this gap, the Open RAN paradigm and its specification introduce an open architecture with abstractions that enable closed-loop control and provide data-driven, and intelligent optimization of the RAN at the user level. This is obtained through custom RAN control applications (i.e., xApps) deployed on near-real-time RAN Intelligent Controller (near-RT RIC) at the edge of the network. Despite these premises, as of today the research community lacks a sandbox to build data-driven xApps, and create large-scale datasets for effective AI training. In this paper, we address this by introducing ns-O-RAN, a software framework that integrates a real-world, production-grade near-RT RIC with a 3GPP-based simulated environment on ns-3, enabling the development of xApps and automated large-scale data collection and testing of Deep Reinforcement Learning-driven control policies for the optimization at the user-level. In addition, we propose the first user-specific O-RAN Traffic Steering (TS) intelligent handover framework. It uses Random Ensemble Mixture, combined with a state-of-the-art Convolutional Neural Network architecture, to optimally assign a serving base station to each user in the network. Our TS xApp, trained with more than 40 million data points collected by ns-O-RAN, runs on the near-RT RIC and controls its base stations. We evaluate the performance on a large-scale deployment, showing that the xApp-based handover improves throughput and spectral efficiency by an average of 50% over traditional handover heuristics, with less mobility overhead.

We propose a novel concept of augmented reality (AR) human-drone interaction driven by RL-based swarm behavior to achieve intuitive and immersive control of a swarm formation of unmanned aerial vehicles. The DroneARchery system developed by us allows the user to quickly deploy a swarm of drones, generating flight paths simulating archery. The haptic interface LinkGlide delivers a tactile stimulus of the bowstring tension to the forearm to increase the precision of aiming. The swarm of released drones dynamically avoids collisions between each other, the drone following the user, and external obstacles with behavior control based on deep reinforcement learning. The developed concept was tested in the scenario with a human, where the user shoots from a virtual bow with a real drone to hit the target. The human operator observes the ballistic trajectory of the drone in an AR and achieves a realistic and highly recognizable experience of the bowstring tension through the haptic display. The experimental results revealed that the system improves trajectory prediction accuracy by 63.3% through applying AR technology and conveying haptic feedback of pulling force. DroneARchery users highlighted the naturalness (4.3 out of 5 point Likert scale) and increased confidence (4.7 out of 5) when controlling the drone. We have designed the tactile patterns to present four sliding distances (tension) and three applied force levels (stiffness) of the haptic display. Users demonstrated the ability to distinguish tactile patterns produced by the haptic display representing varying bowstring tension(average recognition rate is of 72.8%) and stiffness (average recognition rate is of 94.2%). The novelty of the research is the development of an AR-based approach for drone control that does not require special skills and training from the operator.

During initial iterations of training in most Reinforcement Learning (RL) algorithms, agents perform a significant number of random exploratory steps. In the real world, this can limit the practicality of these algorithms as it can lead to potentially dangerous behavior. Hence safe exploration is a critical issue in applying RL algorithms in the real world. This problem has been recently well studied under the Constrained Markov Decision Process (CMDP) Framework, where in addition to single-stage rewards, an agent receives single-stage costs or penalties as well depending on the state transitions. The prescribed cost functions are responsible for mapping undesirable behavior at any given time-step to a scalar value. The goal then is to find a feasible policy that maximizes reward returns while constraining the cost returns to be below a prescribed threshold during training as well as deployment. We propose an On-policy Model-based Safe Deep RL algorithm in which we learn the transition dynamics of the environment in an online manner as well as find a feasible optimal policy using the Lagrangian Relaxation-based Proximal Policy Optimization. We use an ensemble of neural networks with different initializations to tackle epistemic and aleatoric uncertainty issues faced during environment model learning. We compare our approach with relevant model-free and model-based approaches in Constrained RL using the challenging Safe Reinforcement Learning benchmark - the Open AI Safety Gym. We demonstrate that our algorithm is more sample efficient and results in lower cumulative hazard violations as compared to constrained model-free approaches. Further, our approach shows better reward performance than other constrained model-based approaches in the literature.

A/B tests) are a powerful tool for estimating treatment effects, to inform decisions making in business, healthcare and other applications. In many problems, the treatment has a lasting effect that evolves over time. A limitation with randomized experiments is that they do not easily extend to measure long-term effects, since running long experiments is time-consuming and expensive. In this paper, we take a reinforcement learning (RL) approach that estimates the average reward in a Markov process. Motivated by real-world scenarios where the observed state transition is nonstationary, we develop a new algorithm for a class of nonstationary problems, and demonstrate promising results in two synthetic datasets and one online store dataset.

This paper considers offline multi-agent reinforcement learning. We propose the strategy-wise concentration principle which directly builds a confidence interval for the joint strategy, in contrast to the point-wise concentration principle that builds a confidence interval for each point in the joint action space. For two-player zero-sum Markov games, by exploiting the convexity of the strategy-wise bonus, we propose a computationally efficient algorithm whose sample complexity enjoys a better dependency on the number of actions than the prior methods based on the point-wise bonus. Furthermore, for offline multi-agent general-sum Markov games, based on the strategy-wise bonus and a novel surrogate function, we give the first algorithm whose sample complexity only scales $\sum_{i=1}^mA_i$ where $A_i$ is the action size of the $i$-th player and $m$ is the number of players. In sharp contrast, the sample complexity of methods based on the point-wise bonus would scale with the size of the joint action space $\Pi_{i=1}^m A_i$ due to the curse of multiagents. Lastly, all of our algorithms can naturally take a pre-specified strategy class $\Pi$ as input and output a strategy that is close to the best strategy in $\Pi$. In this setting, the sample complexity only scales with $\log |\Pi|$ instead of $\sum_{i=1}^mA_i$.

Planning coverage path for multiple robots in a decentralized way enhances robustness to coverage tasks handling uncertain malfunctions. To achieve high efficiency in a distributed manner for each single robot, a comprehensive understanding of both the complicated environments and cooperative agents intent is crucial. Unfortunately, existing works commonly consider only part of these factors, resulting in imbalanced subareas or unnecessary overlaps. To tackle this issue, we introduce a Decentralized reinforcement learning framework with dual guidance to train each agent to solve the decentralized multiple coverage path planning problem straightly through the environment states. As distributed robots require others intentions to perform better coverage efficiency, we utilize two guidance methods, artificial potential fields and heuristic guidance, to include and integrate others intentions into observations for each robot. With our constructed framework, results have shown our agents successfully learn to determine their own subareas while achieving full coverage, balanced subareas and low overlap rates. We then implement spanning tree cover within those subareas to construct actual routes for each robot and complete given coverage tasks. Our performance is also compared with the state of the art decentralized method showing at most 10 percent lower overlap rates while performing high efficiency in similar environments.