Federated Reinforcement Learning (FRL) has garnered increasing attention recently. However, due to the intrinsic spatio-temporal non-stationarity of data distributions, the current approaches typically suffer from high interaction and communication costs. In this paper, we introduce a new FRL algorithm, named $\texttt{MFPO}$, that utilizes momentum, importance sampling, and additional server-side adjustment to control the shift of stochastic policy gradients and enhance the efficiency of data utilization. We prove that by proper selection of momentum parameters and interaction frequency, $\texttt{MFPO}$ can achieve $\tilde{\mathcal{O}}(H N^{-1}\epsilon^{-3/2})$ and $\tilde{\mathcal{O}}(\epsilon^{-1})$ interaction and communication complexities ($N$ represents the number of agents), where the interaction complexity achieves linear speedup with the number of agents, and the communication complexity aligns the best achievable of existing first-order FL algorithms. Extensive experiments corroborate the substantial performance gains of $\texttt{MFPO}$ over existing methods on a suite of complex and high-dimensional benchmarks.

Intelligent agents must be generalists, capable of quickly adapting to various tasks. In reinforcement learning (RL), model-based RL learns a dynamics model of the world, in principle enabling transfer to arbitrary reward functions through planning. However, autoregressive model rollouts suffer from compounding error, making model-based RL ineffective for long-horizon problems. Successor features offer an alternative by modeling a policy's long-term state occupancy, reducing policy evaluation under new tasks to linear reward regression. Yet, policy improvement with successor features can be challenging. This work proposes a novel class of models, i.e., generalized occupancy models (GOMs), that learn a distribution of successor features from a stationary dataset, along with a policy that acts to realize different successor features. These models can quickly select the optimal action for arbitrary new tasks. By directly modeling long-term outcomes in the dataset, GOMs avoid compounding error while enabling rapid transfer across reward functions. We present a practical instantiation of GOMs using diffusion models and show their efficacy as a new class of transferable models, both theoretically and empirically across various simulated robotics problems.

Offline reinforcement learning (RL) aims to learn optimal policies from previously collected datasets. Recently, due to their powerful representational capabilities, diffusion models have shown significant potential as policy models for offline RL issues. However, previous offline RL algorithms based on diffusion policies generally adopt weighted regression to improve the policy. This approach optimizes the policy only using the collected actions and is sensitive to Q-values, which limits the potential for further performance enhancement. To this end, we propose a novel preferred-action-optimized diffusion policy for offline RL. In particular, an expressive conditional diffusion model is utilized to represent the diverse distribution of a behavior policy. Meanwhile, based on the diffusion model, preferred actions within the same behavior distribution are automatically generated through the critic function. Moreover, an anti-noise preference optimization is designed to achieve policy improvement by using the preferred actions, which can adapt to noise-preferred actions for stable training. Extensive experiments demonstrate that the proposed method provides competitive or superior performance compared to previous state-of-the-art offline RL methods, particularly in sparse reward tasks such as Kitchen and AntMaze. Additionally, we empirically prove the effectiveness of anti-noise preference optimization.

Reinforcement learning has traditionally focused on learning state-dependent policies to solve optimal control problems in a closed-loop fashion. In this work, we introduce the paradigm of open-loop reinforcement learning where a fixed action sequence is learned instead. We present three new algorithms: one robust model-based method and two sample-efficient model-free methods. Rather than basing our algorithms on Bellman's equation from dynamic programming, our work builds on Pontryagin's principle from the theory of open-loop optimal control. We provide convergence guarantees and evaluate all methods empirically on a pendulum swing-up task, as well as on two high-dimensional MuJoCo tasks, demonstrating remarkable performance compared to existing baselines.

We show that multi-agent reinforcement learning (MARL) with full parameter sharing outperforms independent and partially shared architectures in the competitive perfect-information homogenous game of Chinese Checkers. To run our experiments, we develop a new MARL environment: variable-size, six-player Chinese Checkers. This custom environment was developed in PettingZoo and supports all traditional rules of the game including chaining jumps. This is, to the best of our knowledge, the first implementation of Chinese Checkers that remains faithful to the true game. Chinese Checkers is difficult to learn due to its large branching factor and potentially infinite horizons. We borrow the concept of branching actions (submoves) from complex action spaces in other RL domains, where a submove may not end a player's turn immediately. This drastically reduces the dimensionality of the action space. Our observation space is inspired by AlphaGo with many binary game boards stacked in a 3D array to encode information. The PettingZoo environment, training and evaluation logic, and analysis scripts can be found on \href{https://github.com/noahadhikari/pettingzoo-chinese-checkers}{Github}.

Empowering safe exploration of reinforcement learning (RL) agents during training is a critical impediment towards deploying RL agents in many real-world scenarios. Training RL agents in unknown, black-box environments poses an even greater safety risk when prior knowledge of the domain/task is unavailable. We introduce ADVICE (Adaptive Shielding with a Contrastive Autoencoder), a novel post-shielding technique that distinguishes safe and unsafe features of state-action pairs during training, thus protecting the RL agent from executing actions that yield potentially hazardous outcomes. Our comprehensive experimental evaluation against state-of-the-art safe RL exploration techniques demonstrates how ADVICE can significantly reduce safety violations during training while maintaining a competitive outcome reward.

The purpose of offline multi-task reinforcement learning (MTRL) is to develop a unified policy applicable to diverse tasks without the need for online environmental interaction. Recent advancements approach this through sequence modeling, leveraging the Transformer architecture's scalability and the benefits of parameter sharing to exploit task similarities. However, variations in task content and complexity pose significant challenges in policy formulation, necessitating judicious parameter sharing and management of conflicting gradients for optimal policy performance. In this work, we introduce the Harmony Multi-Task Decision Transformer (HarmoDT), a novel solution designed to identify an optimal harmony subspace of parameters for each task. We approach this as a bi-level optimization problem, employing a meta-learning framework that leverages gradient-based techniques. The upper level of this framework is dedicated to learning a task-specific mask that delineates the harmony subspace, while the inner level focuses on updating parameters to enhance the overall performance of the unified policy. Empirical evaluations on a series of benchmarks demonstrate the superiority of HarmoDT, verifying the effectiveness of our approach.

Policy gradient methods have enabled deep reinforcement learning (RL) to approach challenging continuous control problems, even when the underlying systems involve highly nonlinear dynamics that generate complex non-smooth optimization landscapes. We develop a rigorous framework for understanding how policy gradient methods mollify non-smooth optimization landscapes to enable effective policy search, as well as the downside of it: while making the objective function smoother and easier to optimize, the stochastic objective deviates further from the original problem. We demonstrate the equivalence between policy gradient methods and solving backward heat equations. Following the ill-posedness of backward heat equations from PDE theory, we present a fundamental challenge to the use of policy gradient under stochasticity. Moreover, we make the connection between this limitation and the uncertainty principle in harmonic analysis to understand the effects of exploration with stochastic policies in RL. We also provide experimental results to illustrate both the positive and negative aspects of mollification effects in practice.

Learning from multi-step off-policy data collected by a set of policies is a core problem of reinforcement learning (RL). Approaches based on importance sampling (IS) often suffer from large variances due to products of IS ratios. Typical IS-free methods, such as $n$-step Q-learning, look ahead for $n$ time steps along the trajectory of actions (where $n$ is called the lookahead depth) and utilize off-policy data directly without any additional adjustment. They work well for proper choices of $n$. We show, however, that such IS-free methods underestimate the optimal value function (VF), especially for large $n$, restricting their capacity to efficiently utilize information from distant future time steps. To overcome this problem, we introduce a novel, IS-free, multi-step off-policy method that avoids the underestimation issue and converges to the optimal VF. At its core lies a simple but non-trivial \emph{highway gate}, which controls the information flow from the distant future by comparing it to a threshold. The highway gate guarantees convergence to the optimal VF for arbitrary $n$ and arbitrary behavioral policies. It gives rise to a novel family of off-policy RL algorithms that safely learn even when $n$ is very large, facilitating rapid credit assignment from the far future to the past. On tasks with greatly delayed rewards, including video games where the reward is given only at the end of the game, our new methods outperform many existing multi-step off-policy algorithms.

Machine learning (ML) methods have experienced significant growth in the past decade, yet their practical application in high-impact real-world domains has been hindered by their opacity. When ML methods are responsible for making critical decisions, stakeholders often require insights into how to alter these decisions. Counterfactual explanations (CFEs) have emerged as a solution, offering interpretations of opaque ML models and providing a pathway to transition from one decision to another. However, most existing CFE methods require access to the model's training dataset, few methods can handle multivariate time-series, and none can handle multivariate time-series without training datasets. These limitations can be formidable in many scenarios. In this paper, we present CFWoT, a novel reinforcement-learning-based CFE method that generates CFEs when training datasets are unavailable. CFWoT is model-agnostic and suitable for both static and multivariate time-series datasets with continuous and discrete features. Users have the flexibility to specify non-actionable, immutable, and preferred features, as well as causal constraints which CFWoT guarantees will be respected. We demonstrate the performance of CFWoT against four baselines on several datasets and find that, despite not having access to a training dataset, CFWoT finds CFEs that make significantly fewer and significantly smaller changes to the input time-series. These properties make CFEs more actionable, as the magnitude of change required to alter an outcome is vastly reduced.