We study the problem of online model selection in reinforcement learning, where the selector has access to a class of reinforcement learning agents and learns to adaptively select the agent with the right configuration. Our goal is to establish the improved efficiency and performance gains achieved by integrating online model selection methods into reinforcement learning training procedures. We examine the theoretical characterizations that are effective for identifying the right configuration in practice, and address three practical criteria from a theoretical perspective: 1) Efficient resource allocation, 2) Adaptation under non-stationary dynamics, and 3) Training stability across different seeds. Our theoretical results are accompanied by empirical evidence from various model selection tasks in reinforcement learning, including neural architecture selection, step-size selection, and self model selection.

The rapid diffusion of autonomous pricing algorithms has reshaped competitive dynamics in digital marketplaces, raising important economic and policy questions about their potential for collusive behavior. A substantial body of research demonstrates that reinforcement-learning (RL) agents can autonomously coordinate on supracompetitive outcomes even in the absence of explicit communication. Foundational contributions--including the work in [1]--show that algorithmic agents may systematically learn tacitly collusive strategies across multiple market structures, with Q-learning in particular generating prices above competitive levels in Logit, Hotelling, and linear demand environments. These concerns are reinforced by seminal work such as [2], which demonstrates that simple Q-learning agents reliably sustain collusion through structured punishment and reward cycles in repeated pricing games, as well as by [3], who document how algorithmic systems may generate sudden price spikes in response to high-impact, low-probability events (HILP), unintentionally coordinating on elevated prices. The study of [4] establishes a robust empirical and computational foundation demonstrating that pricing algorithms may autonomously learn to collude. A complementary line of research focuses specifically on Q-learning's capacity to learn collusive equilibria, as documented in papers [2], [5], and [6]. These findings are consistent with the theoretical properties of Q-learning established by [7], who show that the algorithm incrementally learns long-run discounted value-maximizing strategies in sequential decision problems. More recent studies further reveal that deep reinforcement-learning (deep RL) algorithms--including DDQN and SAC--may also display collusive tendencies. For instance, [8] documents that modern RL systems can coordinate on higher-than-competitive prices under a variety of market configurations.

-- For the task with complicated manipulation in unstructured environments, traditional hand-coded methods are ineffective, while reinforcement learning can provide more general and useful policy. Although the reinforcement learning is able to obtain impressive results, its stability and reliability is hard to guarantee, which would cause the potential safety threats. Besides, the transfer from simulation to real-world also will lead in unpredictable situations. T o enhance the safety and reliability of robots, we introduce the force and haptic perception into reinforcement learning. We demonstrate that the force-based reinforcement learning method can be more adaptive to environment, especially in sim-to-real transfer . Experimental results show in object pushing task, our strategy is safer and more efficient in both simulation and real world, thus it holds prospects for a wide variety of robotic applications.

General-purpose control demands agents that act across many tasks and embodiments, yet research on reinforcement learning (RL) for continuous control remains dominated by single-task or offline regimes, reinforcing a view that online RL does not scale. Inspired by the foundation model recipe (large-scale pretraining followed by light RL) we ask whether a single agent can be trained on hundreds of tasks with online interaction. To accelerate research in this direction, we introduce a new benchmark with 200 diverse tasks spanning many domains and embodiments, each with language instructions, demonstrations, and optionally image observations. We then present \emph{Newt}, a language-conditioned multitask world model that is first pretrained on demonstrations to acquire task-aware representations and action priors, and then jointly optimized with online interaction across all tasks. Experiments show that Newt yields better multitask performance and data-efficiency than a set of strong baselines, exhibits strong open-loop control, and enables rapid adaptation to unseen tasks. We release our environments, demonstrations, code for training and evaluation, as well as 200+ checkpoints.

Large transformer models, trained on diverse datasets, have demonstrated impressive few-shot performance on previously unseen tasks without requiring parameter updates. This capability has also been explored in Reinforcement Learning (RL), where agents interact with the environment to retrieve context and maximize cumulative rewards, showcasing strong adaptability in complex settings. However, in cooperative Multi-Agent Reinforcement Learning (MARL), where agents must coordinate toward a shared goal, decentralized policy deployment can lead to mismatches in task alignment and reward assignment, limiting the efficiency of policy adaptation. To address this challenge, we introduce Multi-agent In-context Coordination via Decentralized Memory Retrieval (MAICC), a novel approach designed to enhance coordination by fast adaptation. Our method involves training a centralized embedding model to capture fine-grained trajectory representations, followed by decentralized models that approximate the centralized one to obtain team-level task information. Based on the learned embeddings, relevant trajectories are retrieved as context, which, combined with the agents' current sub-trajectories, inform decision-making. During decentralized execution, we introduce a novel memory mechanism that effectively balances test-time online data with offline memory. Based on the constructed memory, we propose a hybrid utility score that incorporates both individual- and team-level returns, ensuring credit assignment across agents. Extensive experiments on cooperative MARL benchmarks, including Level-Based Foraging (LBF) and SMAC (v1/v2), show that MAICC enables faster adaptation to unseen tasks compared to existing methods. Code is available at https://github.com/LAMDA-RL/MAICC.

Constrained optimization provides a common framework for dealing with conflicting objectives in reinforcement learning (RL). In most of these settings, the objectives (and constraints) are expressed though the expected accumulated reward. However, this formulation neglects risky or even possibly catastrophic events at the tails of the reward distribution, and is often insufficient for high-stakes applications in which the risk involved in outliers is critical. In this work, we propose a framework for risk-aware constrained RL, which exhibits per-stage robustness properties jointly in reward values and time using optimized certainty equivalents (OCEs). Our framework ensures an exact equivalent to the original constrained problem within a parameterized strong Lagrangian duality framework under appropriate constraint qualifications, and yields a simple algorithmic recipe which can be wrapped around standard RL solvers, such as PPO. Lastly, we establish the convergence of the proposed algorithm under common assumptions, and verify the risk-aware properties of our approach through several numerical experiments.

Robustness to modeling errors and uncertainties remains a central challenge in reinforcement learning (RL). In this work, we address this challenge by leveraging diffusion models to train robust RL policies. Diffusion models have recently gained popularity in model-based RL due to their ability to generate full trajectories "all at once", mitigating the compounding errors typical of step-by-step transition models. Moreover, they can be conditioned to sample from specific distributions, making them highly flexible. We leverage conditional sampling to learn policies that are robust to uncertainty in environment dynamics. Building on the established connection between Conditional Value at Risk (CVaR) optimization and robust RL, we introduce Adversarial Diffusion for Robust Reinforcement Learning (AD-RRL). AD-RRL guides the diffusion process to generate worst-case trajectories during training, effectively optimizing the CVaR of the cumulative return. Empirical results across standard benchmarks show that AD-RRL achieves superior robustness and performance compared to existing robust RL methods.

Approaches for goal-conditioned reinforcement learning (GCRL) often use learned state representations to extract goal-reaching policies. Two frameworks for representation structure have yielded particularly effective GCRL algorithms: (1) *contrastive representations*, in which methods learn "successor features" with a contrastive objective that performs inference over future outcomes, and (2) *temporal distances*, which link the (quasimetric) distance in representation space to the transit time from states to goals. We propose an approach that unifies these two frameworks, using the structure of a quasimetric representation space (triangle inequality) with the right additional constraints to learn successor representations that enable optimal goal-reaching. Unlike past work, our approach is able to exploit a **quasimetric** distance parameterization to learn **optimal** goal-reaching distances, even with **suboptimal** data and in **stochastic** environments. This gives us the best of both worlds: we retain the stability and long-horizon capabilities of Monte Carlo contrastive RL methods, while getting the free stitching capabilities of quasimetric network parameterizations. On existing offline GCRL benchmarks, our representation learning objective improves performance on stitching tasks where methods based on contrastive learning struggle, and on noisy, high-dimensional environments where methods based on quasimetric networks struggle.

In this paper we propose two algorithms in the tabular setting and an algorithm for the function approximation setting for the Stochastic Shortest Path (SSP) problem. SSP problems form an important class of problems in Reinforcement Learning (RL), as other types of cost-criteria in RL can be formulated in the setting of SSP. We show asymptotic almost-sure convergence for all our algorithms. We observe superior performance of our tabular algorithms compared to other well-known convergent RL algorithms. We further observe reliable performance of our function approximation algorithm compared to other algorithms in the function approximation setting.

Humans and animals show remarkable learning efficiency, adapting to new environments with minimal experience. This capability is not well captured by standard reinforcement learning algorithms that rely on incremental value updates. Rapid adaptation likely depends on episodic memory -- the ability to retrieve specific past experiences to guide decisions in novel contexts. Transformers provide a useful setting for studying these questions because of their ability to learn rapidly in-context and because their key-value architecture resembles episodic memory systems in the brain. We train a transformer to in-context reinforcement learn in a distribution of planning tasks inspired by rodent behavior. We then characterize the learning algorithms that emerge in the model. We first find that representation learning is supported by in-context structure learning and cross-context alignment, where representations are aligned across environments with different sensory stimuli. We next demonstrate that the reinforcement learning strategies developed by the model are not interpretable as standard model-free or model-based planning. Instead, we show that in-context reinforcement learning is supported by caching intermediate computations within the model's memory tokens, which are then accessed at decision time. Overall, we find that memory may serve as a computational resource, storing both raw experience and cached computations to support flexible behavior. Furthermore, the representations developed in the model resemble computations associated with the hippocampal-entorhinal system in the brain, suggesting that our findings may be relevant for natural cognition. Taken together, our work offers a mechanistic hypothesis for the rapid adaptation that underlies in-context learning in artificial and natural settings.