The aim of this paper is to improve the understanding of the optimization landscape for policy optimization problems in reinforcement learning. Specifically, we show that the superlevel set of the objective function with respect to the policy parameter is always a connected set both in the tabular setting and under policies represented by a class of neural networks. In addition, we show that the optimization objective as a function of the policy parameter and reward satisfies a stronger "equiconnectedness" property. To our best knowledge, these are novel and previously unknown discoveries. We present an application of the connectedness of these superlevel sets to the derivation of minimax theorems for robust reinforcement learning. We show that any minimax optimization program which is convex on one side and is equiconnected on the other side observes the minimax equality (i.e. has a Nash equilibrium). We find that this exact structure is exhibited by an interesting robust reinforcement learning problem under an adversarial reward attack, and the validity of its minimax equality immediately follows. This is the first time such a result is established in the literature.

Behavioral health interventions, delivered through digital platforms, have the potential to significantly improve health outcomes, through education, motivation, reminders, and outreach. We study the problem of optimizing personalized interventions for patients to maximize a long-term outcome, where interventions are costly and capacity-constrained. We assume there exists a dataset collected from an initial pilot study that we can leverage. We present a new approach for this problem that we dub DecompPI, which approximates one step of policy iteration. Implementing DecompPI simply consists of a prediction task using the dataset, alleviating the need for online experimentation. DecompPI is a generic model-free algorithm that can be used irrespective of the underlying patient behavior model. We derive theoretical guarantees on a simple, special case of the model that is representative of our problem setting. We establish an approximation ratio for DecompPI with respect to the improvement beyond a null policy that does not allocate interventions. Specifically, when the initial policy used to collect the data is randomized, the approximation ratio of the improvement approaches 1/2 as the intervention capacity of the initial policy decreases. We show that this guarantee is robust to estimation errors. We conduct a rigorous empirical case study using real-world data from a mobile health platform for improving treatment adherence for tuberculosis. Using a validated simulation model, we demonstrate that DecompPI can provide the same efficacy as the status quo approach with approximately half the capacity of interventions. DecompPI is simple and easy to implement for organizations aiming to improve long-term behavior through targeted interventions, and this paper demonstrates its strong performance both theoretically and empirically.

Simulation parameter settings such as contact models and object geometry approximations are critical to training robust robotic policies capable of transferring from simulation to real-world deployment. Previous approaches typically handcraft distributions over such parameters (domain randomization), or identify parameters that best match the dynamics of the real environment (system identification). However, there is often an irreducible gap between simulation and reality: attempting to match the dynamics between simulation and reality across all states and tasks may be infeasible and may not lead to policies that perform well in reality for a specific task. Addressing this issue, we propose AdaptSim, a new task-driven adaptation framework for sim-to-real transfer that aims to optimize task performance in target (real) environments -- instead of matching dynamics between simulation and reality. First, we meta-learn an adaptation policy in simulation using reinforcement learning for adjusting the simulation parameter distribution based on the current policy's performance in a target environment. We then perform iterative real-world adaptation by inferring new simulation parameter distributions for policy training, using a small amount of real data. We perform experiments in three robotic tasks: (1) swing-up of linearized double pendulum, (2) dynamic table-top pushing of a bottle, and (3) dynamic scooping of food pieces with a spatula. Our extensive simulation and hardware experiments demonstrate AdaptSim achieving 1-3x asymptotic performance and $\sim$2x real data efficiency when adapting to different environments, compared to methods based on Sys-ID and directly training the task policy in target environments. Website: https://irom-lab.github.io/AdaptSim/

It remains a significant challenge to train generally capable agents with reinforcement learning (RL). A promising avenue for improving the robustness of RL agents is through the use of curricula. One such class of methods frames environment design as a game between a student and a teacher, using regret-based objectives to produce environment instantiations (or levels) at the frontier of the student agent's capabilities. These methods benefit from their generality, with theoretical guarantees at equilibrium, yet they often struggle to find effective levels in challenging design spaces. By contrast, evolutionary approaches seek to incrementally alter environment complexity, resulting in potentially open-ended learning, but often rely on domain-specific heuristics and vast amounts of computational resources. In this paper we propose to harness the power of evolution in a principled, regret-based curriculum. Our approach, which we call Adversarially Compounding Complexity by Editing Levels (ACCEL), seeks to constantly produce levels at the frontier of an agent's capabilities, resulting in curricula that start simple but become increasingly complex. ACCEL maintains the theoretical benefits of prior regret-based methods, while providing significant empirical gains in a diverse set of environments. An interactive version of the paper is available at accelagent.github.io.

It can largely benefit the reinforcement learning (RL) process of each agent if multiple geographically distributed agents perform their separate RL tasks cooperatively. Different from multi-agent reinforcement learning (MARL) where multiple agents are in a common environment and should learn to cooperate or compete with each other, in this case each agent has its separate environment and only communicates with others to share knowledge without any cooperative or competitive behaviour as a learning outcome. In fact, this scenario exists widely in real life whose concept can be utilised in many applications, but is not well understood yet and not well formulated. As the first effort, we propose group-agent system for RL as a formulation of this scenario and the third type of RL system with respect to single-agent and multi-agent systems. We then propose a distributed RL framework called DDAL (Decentralised Distributed Asynchronous Learning) designed for group-agent reinforcement learning (GARL). We show through experiments that DDAL achieved desirable performance with very stable training and has good scalability.

Robot sequential decision-making in the real world is a challenge because it requires the robots to simultaneously reason about the current world state and dynamics, while planning actions to accomplish complex tasks. On the one hand, declarative languages and reasoning algorithms well support representing and reasoning with commonsense knowledge. But these algorithms are not good at planning actions toward maximizing cumulative reward over a long, unspecified horizon. On the other hand, probabilistic planning frameworks, such as Markov decision processes (MDPs) and partially observable MDPs (POMDPs), well support planning to achieve long-term goals under uncertainty. But they are ill-equipped to represent or reason about knowledge that is not directly related to actions. In this article, we present a novel algorithm, called iCORPP, to simultaneously estimate the current world state, reason about world dynamics, and construct task-oriented controllers. In this process, robot decision-making problems are decomposed into two interdependent (smaller) subproblems that focus on reasoning to "understand the world" and planning to "achieve the goal" respectively. Contextual knowledge is represented in the reasoning component, which makes the planning component epistemic and enables active information gathering. The developed algorithm has been implemented and evaluated both in simulation and on real robots using everyday service tasks, such as indoor navigation, dialog management, and object delivery. Results show significant improvements in scalability, efficiency, and adaptiveness, compared to competitive baselines including handcrafted action policies.

We consider the optimal sample complexity theory of tabular reinforcement learning (RL) for maximizing the infinite horizon discounted reward in a Markov decision process (MDP). Optimal worst-case complexity results have been developed for tabular RL problems in this setting, leading to a sample complexity dependence on $\gamma$ and $\epsilon$ of the form $\tilde \Theta((1-\gamma)^{-3}\epsilon^{-2})$, where $\gamma$ denotes the discount factor and $\epsilon$ is the solution error tolerance. However, in many applications of interest, the optimal policy (or all policies) induces mixing. We establish that in such settings, the optimal sample complexity dependence is $\tilde \Theta(t_{\text{mix}}(1-\gamma)^{-2}\epsilon^{-2})$, where $t_{\text{mix}}$ is the total variation mixing time. Our analysis is grounded in regeneration-type ideas, which we believe are of independent interest, as they can be used to study RL problems for general state space MDPs.

In real-world scenarios, the application of reinforcement learning is significantly challenged by complex non-stationarity. Most existing methods attempt to model changes in the environment explicitly, often requiring impractical prior knowledge. In this paper, we propose a new perspective, positing that non-stationarity can propagate and accumulate through complex causal relationships during state transitions, thereby compounding its sophistication and affecting policy learning. We believe that this challenge can be more effectively addressed by tracing the causal origin of non-stationarity. To this end, we introduce the Causal-Origin REPresentation (COREP) algorithm. COREP primarily employs a guided updating mechanism to learn a stable graph representation for states termed as causal-origin representation. By leveraging this representation, the learned policy exhibits impressive resilience to non-stationarity. We supplement our approach with a theoretical analysis grounded in the causal interpretation for non-stationary reinforcement learning, advocating for the validity of the causal-origin representation. Experimental results further demonstrate the superior performance of COREP over existing methods in tackling non-stationarity.

Traditional quantum system control methods often face different constraints, and are easy to cause both leakage and stochastic control errors under the condition of limited resources. Reinforcement learning has been proved as an efficient way to complete the quantum system control task. To learn a satisfactory control strategy under the condition of limited resources, a quantum system control method based on enhanced reinforcement learning (QSC-ERL) is proposed. The states and actions in reinforcement learning are mapped to quantum states and control operations in quantum systems. By using new enhanced neural networks, reinforcement learning can quickly achieve the maximization of long-term cumulative rewards, and a quantum state can be evolved accurately from an initial state to a target state. According to the number of candidate unitary operations, the three-switch control is used for simulation experiments. Compared with other methods, the QSC-ERL achieves close to 1 fidelity learning control of quantum systems, and takes fewer episodes to quantum state evolution under the condition of limited resources.

Due to the large state space of the two-qubit system, and the adoption of ladder reward function in the existing quantum state preparation methods, the convergence speed is slow and it is difficult to prepare the desired target quantum state with high fidelity under limited conditions. To solve the above problems, a difference-driven reinforcement learning (RL) algorithm for quantum state preparation of two-qubit system is proposed by improving the reward function and action selection strategy. Firstly, a model is constructed for the problem of preparing quantum states of a two-qubit system, with restrictions on the type of quantum gates and the time for quantum state evolution. In the preparation process, a weighted differential dynamic reward function is designed to assist the algorithm quickly obtain the maximum expected cumulative reward. Then, an adaptive e-greedy action selection strategy is adopted to achieve a balance between exploration and utilization to a certain extent, thereby improving the fidelity of the final quantum state. The simulation results show that the proposed algorithm can prepare quantum state with high fidelity under limited conditions. Compared with other algorithms, it has different degrees of improvement in convergence speed and fidelity of the final quantum state.