Competitive Pokémon Singles (CPS) is a popular strategy game where players learn to exploit their opponent based on imperfect information in battles that can last more than one hundred stochastic turns. AI research in CPS has been led by heuristic tree search and online self-play, but the game may also create a platform to study adaptive policies trained offline on large datasets. We develop a pipeline to reconstruct the first-person perspective of an agent from logs saved from the third-person perspective of a spectator, thereby unlocking a dataset of real human battles spanning more than a decade that grows larger every day. This dataset enables a black-box approach where we train large sequence models to adapt to their opponent based solely on their input trajectory while selecting moves without explicit search of any kind. We study a progression from imitation learning to offline RL and offline fine-tuning on self-play data in the hardcore competitive setting of Pokémon's four oldest (and most partially observed) game generations. The resulting agents outperform a recent LLM Agent approach and a strong heuristic search engine. While playing anonymously in online battles against humans, our best agents climb to rankings inside the top 10% of active players. All agent checkpoints, training details, datasets, and baselines are available at https://metamon.tech.

-- This paper explores successor features for knowledge transfer in zero-sum, complete-information, and turn-based games. Prior research in single-agent systems has shown that successor features can provide a "jump start" for agents when facing new tasks with varying reward structures. However, knowledge transfer in games typically relies on value and equilibrium transfers, which heavily depends on the similarity between tasks. This reliance can lead to failures when the tasks differ significantly. T o address this issue, this paper presents an application of successor features to games and presents a novel algorithm called Game Generalized Policy Improvement (GGPI), designed to address Markov games in multi-agent reinforcement learning. The proposed algorithm enables the transfer of learning values and policies across games. An upper bound of the errors for transfer is derived as a function the similarity of the task. Through experiments with a turn-based pursuer-evader game, we demonstrate that the GGPI algorithm can generate high-reward interactions and one-shot policy transfer . When further tested in a wider set of initial conditions, the GGPI algorithm achieves higher success rates with improved path efficiency compared to those of the baseline algorithms.

The rapid advancement of multimodal large language models (MLLMs) has led to breakthroughs in various applications, yet their security remains a critical challenge. One pressing issue involves unsafe image-query pairs--jailbreak inputs specifically designed to bypass security constraints and elicit unintended responses from MLLMs. Compared to general multimodal data, such unsafe inputs are relatively sparse, which limits the diversity and richness of training samples available for developing robust defense models. Meanwhile, existing guardrail-type methods rely on external modules to enforce security constraints but fail to address intrinsic vulnerabilities within MLLMs. Traditional supervised fine-tuning (SFT), on the other hand, often over-refuses harmless inputs, compromising general performance. Given these challenges, we propose Secure Tug-of-War (SecTOW), an innovative iterative defense-attack training method to enhance the security of MLLMs. SecTOW consists of two modules: a defender and an auxiliary attacker, both trained iteratively using reinforcement learning (GRPO). During the iterative process, the attacker identifies security vulnerabilities in the defense model and expands jailbreak data. The expanded data are then used to train the defender, enabling it to address identified security vulnerabilities. We also design reward mechanisms used for GRPO to simplify the use of response labels, reducing dependence on complex generative labels and enabling the efficient use of synthetic data. Additionally, a quality monitoring mechanism is used to mitigate the defender's over-refusal of harmless inputs and ensure the diversity of the jailbreak data generated by the attacker. Experimental results on safety-specific and general benchmarks demonstrate that SecTOW significantly improves security while preserving general performance.

Navigation foundation models trained on massive webscale data enable agents to generalize across diverse environments and embodiments. However, these models trained solely on offline data, often lack the capacity to reason about the consequences of their actions or adapt through counterfactual understanding. They thus face significant limitations in the real-world urban navigation where interactive and safe behaviors, such as avoiding obstacles and moving pedestrians, are critical. To tackle these challenges, we introduce the Seeing-to-Experiencing framework to scale the capability of navigation foundation models with reinforcement learning. S2E combines the strengths of pre-training on videos and post-training through RL. It maintains the generalizability acquired from large-scale real-world videos while enhancing its interactivity through RL in simulation environments. Specifically, we introduce two innovations: an Anchor-Guided Distribution Matching strategy, which stabilizes learning and models diverse motion patterns through anchor-based supervision; and a Residual-Attention Module, which obtains reactive behaviors from simulation environments without erasing the model's pretrained knowledge. Moreover, we establish a comprehensive end-to-end evaluation benchmark, NavBench-GS, built on photorealistic 3DGS reconstructions of real-world scenes that incorporate physical interactions. It can systematically assess the generalizability and safety of navigation foundation models. Extensive experiments show that S2E mitigates the diminishing returns often seen when scaling with offline data alone. We perform a thorough analysis of the benefits of Reinforcement Learning compared to Supervised Fine-Tuning in the context of post-training for robot learning. Our findings emphasize the crucial role of integrating interactive online experiences to effectively scale foundation models in Robotics.

The state-of-the-art DGF techniques redefine and optimize the fitness metric to reach the target sites precisely and quickly. However, optimizations for fitness metrics are mainly based on heuristic algorithms, which usually rely on historical execution information and lack foresight on paths that have not been exercised yet. Thus, those hard-to-execute paths with complex constraints would hinder DGF from reaching the targets, making DGF less efficient. In this paper, we propose DeepGo, a predictive directed grey-box fuzzer that can combine historical and predicted information to steer DGF to reach the target site via an optimal path. We first propose the path transition model, which models DGF as a process of reaching the target site through specific path transition sequences. The new seed generated by mutation would cause the path transition, and the path corresponding to the high-reward path transition sequence indicates a high likelihood of reaching the target site through it. Then, to predict the path transitions and the corresponding rewards, we use deep neural networks to construct a Virtual Ensemble Environment (VEE), which gradually imitates the path transition model and predicts the rewards of path transitions that have not been taken yet. To determine the optimal path, we develop a Reinforcement Learning for Fuzzing (RLF) model to generate the transition sequences with the highest sequence rewards. The RLF model can combine historical and predicted path transitions to generate the optimal path transition sequences, along with the policy to guide the mutation strategy of fuzzing. Finally, to exercise the high-reward path transition sequence, we propose the concept of an action group, which comprehensively optimizes the critical steps of fuzzing to realize the optimal path to reach the target efficiently.

Large Language Models (LLMs) often produce plausible but poorly-calibrated answers, limiting their reliability on reasoning-intensive tasks. We present Reinforcement Learning from Self-Feedback (RLSF), a post-training stage that uses the model's own confidence as an intrinsic reward, mimicking how humans learn in the absence of external feedback. After a frozen LLM generates several chain-of-thought solutions, we define and compute the confidence of each final answer span and rank the traces accordingly. These synthetic preferences are then used to fine-tune the policy with standard preference optimization, similar to RLHF yet requiring no human labels, gold answers, or externally curated rewards. RLSF simultaneously (i) refines the model's probability estimates -- restoring well-behaved calibration -- and (ii) strengthens step-by-step reasoning, yielding improved performance on arithmetic reasoning and multiple-choice question answering. By turning a model's own uncertainty into useful self-feedback, RLSF affirms reinforcement learning on intrinsic model behaviour as a principled and data-efficient component of the LLM post-training pipeline and warrents further research in intrinsic rewards for LLM post-training.

Dynamic programming (DP) is a fundamental tool used across many engineering fields. The main goal of DP is to solve Bellman's optimality equations for a given Markov decision process (MDP). Standard methods like policy iteration exploit the fixed-point nature of these equations to solve them iteratively. However, these algorithms can be computationally expensive when the state-action space is large or when the problem involves long-term dependencies. Here we propose a new approach that unrolls and truncates policy iterations into a learnable parametric model dubbed BellNet, which we train to minimize the so-termed Bellman error from random value function initializations. Viewing the transition probability matrix of the MDP as the adjacency of a weighted directed graph, we draw insights from graph signal processing to interpret (and compactly re-parameterize) BellNet as a cascade of nonlinear graph filters. This fresh look facilitates a concise, transferable, and unifying representation of policy and value iteration, with an explicit handle on complexity during inference. Preliminary experiments conducted in a grid-like environment demonstrate that BellNet can effectively approximate optimal policies in a fraction of the iterations required by classical methods.

In this paper we investigate the notion of legibility in sequential decision-making in the context of teams and teamwork. There have been works that extend the notion of legibility to sequential decision making, for deterministic and for stochastic scenarios. However, these works focus on one agent interacting with one human, foregoing the benefits of having legible decision making in teams of agents or in team configurations with humans. In this work we propose an extension of legible decision-making to multi-agent settings that improves the performance of agents working in collaboration. We showcase the performance of legible decision making in team scenarios using our proposed extension in multi-agent benchmark scenarios. We show that a team with a legible agent is able to outperform a team composed solely of agents with standard optimal behaviour.

In recent years, mutual information optimal control has been proposed as an extension of maximum entropy optimal control. Both approaches introduce regularization terms to render the policy stochastic, and it is important to theoretically clarify the relationship between the temperature parameter (i.e., the coefficient of the regularization term) and the stochasticity of the policy. Unlike in maximum entropy optimal control, this relationship remains unexplored in mutual information optimal control. In this paper, we investigate this relationship for a mutual information optimal control problem (MIOCP) of discrete-time linear systems. After extending the result of a previous study of the MIOCP, we establish the existence of an optimal policy of the MIOCP, and then derive the respective conditions on the temperature parameter under which the optimal policy becomes stochastic and deterministic. Furthermore, we also derive the respective conditions on the temperature parameter under which the policy obtained by an alternating optimization algorithm becomes stochastic and deterministic. The validity of the theoretical results is demonstrated through numerical experiments.

-- This work focuses on multi-floor indoor exploration, which remains an open area of research. Compared to traditional methods, recent learning-based explorers have demonstrated significant potential due to their robust environmental learning and modeling capabilities, but most are restricted to 2D environments. In this paper, we proposed a learning-integrated topological explorer, LITE, for multi-floor indoor environments. As we incrementally build floor-stair topology in exploration using YOLO11-based instance segmentation model, the agent can transition between floors through a finite state machine. Additionally, we implement an attention-based 2D exploration policy that utilizes an attention mechanism to capture spatial dependencies between different regions, thereby determining the next global goal for more efficient exploration. Extensive comparison and ablation studies conducted on the HM3D and MP3D datasets demonstrate that our proposed 2D exploration policy significantly outperforms all baseline explorers in terms of exploration efficiency. Furthermore, experiments in several 3D multi-floor environments indicate that our framework is compatible with various 2D exploration methods, facilitating effective multi-floor indoor exploration. I. INTRODUCTION Autonomous exploration is a fundamental problem in the development of embodied intelligence and plays a crucial role in uncertain scenarios such as search and rescue [1], scene reconstruction [2], and extraterrestrial planetary exploration [3].