We address the problem of personalized, interactive text-to-image (T2I) generation, designing a reinforcement learning (RL) agent which iteratively improves a set of generated images for a user through a sequence of prompt expansions. Using human raters, we create a novel dataset of sequential preferences, which we leverage, together with large-scale open-source (non-sequential) datasets. We construct user-preference and user-choice models using an EM strategy and identify varying user preference types. We then leverage a large multimodal language model (LMM) and a value-based RL approach to suggest a personalized and diverse slate of prompt expansions to the user. Our Personalized And Sequential Text-to-image Agent (PASTA) extends T2I models with personalized multi-turn capabilities, fostering collaborative co-creation and addressing uncertainty or underspecification in a user's intent. We evaluate PASTA using human raters, showing significant improvement compared to baseline methods. We also release our sequential rater dataset and simulated user-rater interactions to support future research in personalized, multi-turn T2I generation.

The rapid advancements in unmanned aerial vehicles (UAVs) have unlocked numerous applications, including environmental monitoring, disaster response, and agricultural surveying. Enhancing the collective behavior of multiple decentralized UAVs can significantly improve these applications through more efficient and coordinated operations. In this study, we explore a Recurrent PPO model for target localization in perceptually degraded environments like places without GNSS/GPS signals. We first developed a single-drone approach for target identification, followed by a decentralized two-drone model. Our approach can utilize two types of sensors on the UAVs, a detection sensor and a target signal sensor. The single-drone model achieved an accuracy of 93%, while the two-drone model achieved an accuracy of 86%, with the latter requiring fewer average steps to locate the target. This demonstrates the potential of our method in UAV swarms, offering efficient and effective localization of radiant targets in complex environmental conditions.

The performance of modern reinforcement learning algorithms critically relies on tuning ever-increasing numbers of hyperparameters. Often, small changes in a hyperparameter can lead to drastic changes in performance, and different environments require very different hyperparameter settings to achieve state-of-the-art performance reported in the literature. We currently lack a scalable and widely accepted approach to characterizing these complex interactions. This work proposes a new empirical methodology for studying, comparing, and quantifying the sensitivity of an algorithm's performance to hyperparameter tuning for a given set of environments. We then demonstrate the utility of this methodology by assessing the hyperparameter sensitivity of several commonly used normalization variants of PPO. The results suggest that several algorithmic performance improvements may, in fact, be a result of an increased reliance on hyperparameter tuning.

The incorporation of memory into agents is essential for numerous tasks within the domain of Reinforcement Learning (RL). In particular, memory is paramount for tasks that require the utilization of past information, adaptation to novel environments, and improved sample efficiency. However, the term "memory" encompasses a wide range of concepts, which, coupled with the lack of a unified methodology for validating an agent's memory, leads to erroneous judgments about agents' memory capabilities and prevents objective comparison with other memory-enhanced agents. This paper aims to streamline the concept of memory in RL by providing practical precise definitions of agent memory types, such as long-term versus short-term memory and declarative versus procedural memory, inspired by cognitive science. Using these definitions, we categorize different classes of agent memory, propose a robust experimental methodology for evaluating the memory capabilities of RL agents, and standardize evaluations. Furthermore, we empirically demonstrate the importance of adhering to the proposed methodology when evaluating different types of agent memory by conducting experiments with different RL agents and what its violation leads to. Reinforcement Learning (RL) effectively addresses various problems within the Markov Decision Process (MDP) framework, where agents make decisions based on immediately available information (Mnih et al., 2015; Badia et al., 2020). However, there are still challenges in applying RL to more complex tasks with partial observability. To successfully address such challenges, it is essential that an agent is able to efficiently store and process the history of its interactions with the environment (Ni et al., 2021). Sequence processing methods originally developed for natural language processing (NLP) can be effectively applied to these tasks because the history of interactions with the environment can be represented as a sequence (Hausknecht & Stone, 2015; Esslinger et al., 2022; Samsami et al., 2024). However, in many tasks, due to the complexity or noisiness of observations, the sparsity of events, the difficulty of designing the reward function, and the long duration of episodes, storing and retrieving important information becomes extremely challenging, and the need for memory mechanisms arises (Graves et al., 2016; Wayne et al., 2018; Goyal et al., 2022).

Recent advances in learning decision-making policies can largely be attributed to training expressive policy models, largely via imitation learning. While imitation learning discards non-expert data, reinforcement learning (RL) can still learn from suboptimal data. However, instantiating RL training of a new policy class often presents a different challenge: most deep RL machinery is co-developed with assumptions on the policy class and backbone, resulting in poor performance when the policy class changes. For instance, SAC utilizes a low-variance reparameterization policy gradient for Gaussian policies, but this is unstable for diffusion policies and intractable for autoregressive categorical policies. To address this issue, we develop an offline RL and online fine-tuning approach called policy-agnostic RL (PA-RL) that can effectively train multiple policy classes, with varying architectures and sizes. We build off the basic idea that a universal supervised learning loss can replace the policy improvement step in RL, as long as it is applied on "optimized" actions. To obtain these optimized actions, we first sample multiple actions from a base policy, and run global optimization (i.e., re-ranking multiple action samples using the Q-function) and local optimization (i.e., running gradient steps on an action sample) to maximize the critic on these candidates. PA-RL enables fine-tuning diffusion and transformer policies with either autoregressive tokens or continuous action outputs, at different sizes, entirely via actor-critic RL. Moreover, PA-RL improves the performance and sample-efficiency by up to 2 times compared to existing offline RL and online fine-tuning methods. We show the first result that successfully fine-tunes OpenVLA, a 7B generalist robot policy, autonomously with Cal-QL, an online RL fine-tuning algorithm, improving from 40% to 70% in the real world in 40 minutes.

We study learning in a dynamically evolving environment modeled as a Markov game between a learner and a strategic opponent that can adapt to the learner's strategies. While most existing works in Markov games focus on external regret as the learning objective, external regret becomes inadequate when the adversaries are adaptive. In this work, we focus on \emph{policy regret} -- a counterfactual notion that aims to compete with the return that would have been attained if the learner had followed the best fixed sequence of policy, in hindsight. We show that if the opponent has unbounded memory or if it is non-stationary, then sample-efficient learning is not possible. For memory-bounded and stationary, we show that learning is still statistically hard if the set of feasible strategies for the learner is exponentially large. To guarantee learnability, we introduce a new notion of \emph{consistent} adaptive adversaries, wherein, the adversary responds similarly to similar strategies of the learner. We provide algorithms that achieve $\sqrt{T}$ policy regret against memory-bounded, stationary, and consistent adversaries.

The outstanding capabilities of large language models (LLMs) render them a crucial component in various autonomous agent systems. While traditional methods depend on the inherent knowledge of LLMs without fine-tuning, more recent approaches have shifted toward the reinforcement learning strategy to further enhance agents' ability to solve complex interactive tasks with environments and tools. However, previous approaches are constrained by the sparse reward issue, where existing datasets solely provide a final scalar reward for each multi-step reasoning chain, potentially leading to ineffectiveness and inefficiency in policy learning. In this paper, we introduce StepAgent, which utilizes step-wise reward to optimize the agent's reinforcement learning process. Inheriting the spirit of novice-to-expert theory, we first compare the actions of the expert and the agent to automatically generate intermediate rewards for fine-grained optimization. Additionally, we propose implicit-reward and inverse reinforcement learning techniques to facilitate agent reflection and policy adjustment. Further theoretical analysis demonstrates that the action distribution of the agent can converge toward the expert action distribution over multiple training cycles. Experimental results across various datasets indicate that StepAgent outperforms existing baseline methods.

Emerging digital twin technology has the potential to revolutionize voltage control in power systems. However, the state-of-the-art digital twin method suffers from low computational and sampling efficiency, which hinders its applications. To address this issue, we propose a Gumbel-Consistency Digital Twin (GC-DT) method that enhances voltage control with improved computational and sampling efficiency. First, the proposed method incorporates a Gumbel-based strategy improvement that leverages the Gumbel-top trick to enhance non-repetitive sampling actions and reduce the reliance on Monte Carlo Tree Search simulations, thereby improving computational efficiency. Second, a consistency loss function aligns predicted hidden states with actual hidden states in the latent space, which increases both prediction accuracy and sampling efficiency. Experiments on IEEE 123-bus, 34-bus, and 13-bus systems demonstrate that the proposed GC-DT outperforms the state-of-the-art DT method in both computational and sampling efficiency.

While significant progress has been made in task-specific applications, current models struggle with deep reasoning, generality, and adaptation -- key components of System 2 reasoning that are crucial for achieving Artificial General Intelligence (AGI). Despite the promise of approaches such as program synthesis, language models, and transformers, these methods often fail to generalize beyond their training data and to adapt to novel tasks, limiting their ability to perform human-like reasoning. This paper explores the limitations of existing approaches in achieving advanced System 2 reasoning and highlights the importance of generality and adaptation for AGI. Moreover, we propose four key research directions to address these gaps: (1) learning human intentions from action sequences, (2) combining symbolic and neural models, (3) meta-learning for unfamiliar environments, and (4) reinforcement learning to reason multi-step. Through these directions, we aim to advance the ability to generalize and adapt, bringing computational models closer to the reasoning capabilities required for AGI.

To develop autonomous agents capable of executing complex, multi-step decision-making tasks as specified by humans in natural language, existing reinforcement learning approaches typically require expensive labeled datasets or access to real-time experimentation. Moreover, conventional methods often face difficulties in generalizing to unseen goals and states, thereby limiting their practical applicability. This paper presents TEDUO, a novel training pipeline for offline language-conditioned policy learning. TEDUO operates on easy-to-obtain, unlabeled datasets and is suited for the so-called in-the-wild evaluation, wherein the agent encounters previously unseen goals and states. To address the challenges posed by such data and evaluation settings, our method leverages the prior knowledge and instruction-following capabilities of large language models (LLMs) to enhance the fidelity of pre-collected offline data and enable flexible generalization to new goals and states. Empirical results demonstrate that the dual role of LLMs in our framework-as data enhancers and generalizers-facilitates both effective and data-efficient learning of generalizable language-conditioned policies.