Large language models (LLMs) excel at logical and algorithmic reasoning, yet their emotional intelligence (EQ) still lags far behind their cognitive prowess. While reinforcement learning from verifiable rewards (RLVR) has advanced in other domains, its application to dialogue-especially for emotional intelligence-remains underexplored. In this work, we introduce RLVER, the first end-to-end reinforcement learning framework that leverages verifiable emotion rewards from simulated users to cultivate higher-order empathetic abilities in LLMs. Within this framework, self-consistent affective simulated users engage in dialogue rollouts and produce deterministic emotion scores during conversations, serving as reward signals to guide the LLM's learning. Fine-tuning publicly available Qwen2.5-7B-Instruct model with PPO boosts its Sentient-Benchmark score from 13.3 to 79.2 while largely preserving mathematical and coding competence. Extensive experiments reveal that: (i) RLVER consistently improves multiple dialogue capabilities; (ii) Thinking and non-thinking models show distinct trends--thinking models excel in empathy and insight, while non-thinking models favor action; (iii) GRPO often yields stable gains, while PPO can push certain capabilities to a higher ceiling; (iv) More challenging environments are not always better-moderate ones can yield stronger outcomes. Our results show that RLVER is a practical route toward emotionally intelligent and broadly capable language agents.
Recent reinforcement learning approaches, such as outcome-supervised GRPO, have advanced Chain-of-Thought reasoning in large language models (LLMs), yet their adaptation to multimodal LLMs (MLLMs) is unexplored. To address the lack of rigorous evaluation for MLLM post-training methods, we introduce SEED-Bench-R1, a benchmark with complex real-world videos requiring balanced perception and reasoning. It offers a large training set and evaluates generalization across three escalating challenges: in-distribution, cross-environment, and cross-environment-task scenarios. Using SEED-Bench-R1, we find that standard GRPO, while improving answer accuracy, often reduces logical coherence between reasoning steps and answers, with only a 57.9% consistency rate. This stems from reward signals focusing solely on final answers, encouraging shortcuts, and strict KL penalties limiting exploration.To address this, we propose GRPO-CARE, a consistency-aware RL framework optimizing both answer correctness and reasoning coherence without explicit supervision. GRPO-CARE introduces a two-tiered reward: (1) a base reward for answer correctness, and (2) an adaptive consistency bonus, computed by comparing the model's reasoning-to-answer likelihood (via a slowly-evolving reference model) against group peers.This dual mechanism amplifies rewards for reasoning paths that are both correct and logically consistent. Replacing KL penalties with this adaptive bonus, GRPO-CARE outperforms standard GRPO on SEED-Bench-R1, achieving a 6.7% performance gain on the hardest evaluation level and a 24.5% improvement in consistency. It also shows strong transferability, improving model performance across diverse video understanding benchmarks. Our work contributes a systematically designed benchmark and a generalizable post-training framework, advancing the development of more interpretable and robust MLLMs.
Inspired by the impressive reasoning capabilities demonstrated by reinforcement learning approaches like DeepSeek-R1, recent emerging research has begun exploring the use of reinforcement learning (RL) to enhance vision-language models (VLMs) for multimodal reasoning tasks. However, most existing multimodal reinforcement learning approaches remain limited to spatial reasoning within single-image contexts, yet still struggle to generalize to more complex and real-world scenarios involving multi-image positional reasoning, where understanding the relationships across images is crucial. To address this challenge, we propose a general reinforcement learning approach PeRL tailored for interleaved multimodal tasks, and a multi-stage strategy designed to enhance the exploration-exploitation trade-off, thereby improving learning efficiency and task performance. Specifically, we introduce permutation of image sequences to simulate varied positional relationships to explore more spatial and positional diversity. Furthermore, we design a rollout filtering mechanism for resampling to focus on trajectories that contribute most to learning optimal behaviors to exploit learned policies effectively. We evaluate our model on 5 widely-used multi-image benchmarks and 3 single-image benchmarks. Our experiments confirm that PeRL trained model consistently surpasses R1-related and interleaved VLM baselines by a large margin, achieving state-of-the-art performance on multi-image benchmarks, while preserving comparable performance on single-image tasks.
We present Step-wise Policy for Rare-tool Knowledge (SPaRK), a novel reinforcement learning framework that teaches large language models to explore diverse tool usage patterns beyond conventional high-temperature sampling. Building on recent advances in step-wise reinforcement learning, we introduce a dual-objective reward system that simultaneously optimizes for answer quality and tool diversity, training a Llama-3.1 8B model through offline PPO on synthetically generated trajectories from the MMLU-Pro dataset. Our approach uniquely employs a rarity-first exploitation strategy where a GPT-4o judge scores candidate actions across eight distinct tools plus chain-of-thought reasoning, with the policy favoring less-frequently used but still viable tools to encourage systematic exploration. Empirical results demonstrate that SPaRK achieves competitive performance across 14 MMLU-Pro categories while exhibiting significantly higher entropy in tool selection compared to both baseline and supervised fine-tuning approaches, suggesting that algorithmic exploration through explicit tool diversity can enhance reasoning capabilities without sacrificing accuracy.
Achieving fast and stable off-policy learning in deep reinforcement learning (RL) is challenging. Most existing methods rely on semi-gradient temporal-difference (TD) methods for their simplicity and efficiency, but are consequently susceptible to divergence. While more principled approaches like Gradient TD (GTD) methods have strong convergence guarantees, they have rarely been used in deep RL. Recent work introduced the Generalized Projected Bellman Error ($\GPBE$), enabling GTD methods to work efficiently with nonlinear function approximation. However, this work is only limited to one-step methods, which are slow at credit assignment and require a large number of samples. In this paper, we extend the $\GPBE$ objective to support multistep credit assignment based on the $\lambda$-return and derive three gradient-based methods that optimize this new objective. We provide both a forward-view formulation compatible with experience replay and a backward-view formulation compatible with streaming algorithms. Finally, we evaluate the proposed algorithms and show that they outperform both PPO and StreamQ in MuJoCo and MinAtar environments, respectively. Code available at https://github.com/esraaelelimy/gtd\_algos
Kernel development in deep learning requires optimizing computational units across hardware while balancing memory management, parallelism, and hardware-specific optimizations through extensive empirical tuning. Although domain-specific languages like Triton simplify GPU programming by abstracting low-level details, developers must still manually tune critical parameters such as tile sizes and memory access patterns through iterative experimentation, creating substantial barriers to optimal performance and wider adoption. In this work, we introduce AutoTriton, the first model dedicated to Triton programming powered by reinforcement learning (RL). AutoTriton performs supervised fine-tuning (SFT) to be equipped with essential Triton programming expertise using a high-quality data gathering pipeline, and conducts RL with Group Relative Policy Optimization (GRPO) algorithm, combining a rule-based reward and an execution-based reward to further improve Triton programming ability, sequentially. Experiments across five evaluation channels of TritonBench and KernelBench illustrate that our 8B model AutoTriton achieves performance comparable to mainstream large models, including Claude-4-Sonnet and DeepSeek-R1-0528. Further experimental analysis demonstrates the crucial role of each module within AutoTriton, including the SFT stage, the RL stage, and the reward design strategy. These findings underscore the promise of RL for automatically generating high-performance kernels, and since high-performance kernels are core components of AI systems, this breakthrough establishes an important foundation for building more efficient AI systems. The model and code will be available at https://github.com/AI9Stars/AutoTriton.
Zero-shot reinforcement learning is necessary for extracting optimal policies in absence of concrete rewards for fast adaptation to future problem settings. Forward-backward representations (FB) have emerged as a promising method for learning optimal policies in absence of rewards via a factorization of the policy occupancy measure. However, up until now, FB and many similar zero-shot reinforcement learning algorithms have been decoupled from the exploration problem, generally relying on other exploration algorithms for data collection. We argue that FB representations should fundamentally be used for exploration in order to learn more efficiently. With this goal in mind, we design exploration policies that arise naturally from the FB representation that minimize the posterior variance of the FB representation, hence minimizing its epistemic uncertainty. We empirically demonstrate that such principled exploration strategies improve sample complexity of the FB algorithm considerably in comparison to other exploration methods. Code is publicly available at https://sites.google.com/view/fbee-url.
Delayed and sparse rewards present a fundamental obstacle for reinforcement-learning (RL) agents, which struggle to assign credit for actions whose benefits emerge many steps later. The sliding-tile game 2048 epitomizes this challenge: although frequent small score changes yield immediate feedback, they often mislead agents into locally optimal but globally suboptimal strategies. In this work, we introduce a unified, distributional multi-step RL framework designed to directly optimize long-horizon performance. Using the open source Gym-2048 environment we develop and compare four agent variants: standard DQN, PPO, QR-DQN (Quantile Regression DQN), and a novel Horizon-DQN (H-DQN) that integrates distributional learning, dueling architectures, noisy networks, prioritized replay, and more. Empirical evaluation reveals a clear hierarchy in effectiveness: max episode scores improve from 3.988K (DQN) to 5.756K (PPO), 8.66K (QR-DQN), and 18.21K (H-DQN), with H-DQN reaching the 2048 tile. Upon scaling H-DQN it reaches a max score 41.828K and a 4096 tile. These results demonstrate that distributional, multi-step targets substantially enhance performance in sparse-reward domains, and they suggest promising avenues for further gains through model-based planning and curriculum learning.
Large Language Models (LLMs) have achieved strong performance in domains like mathematics, factual QA, and code generation, yet their multilingual reasoning capabilities in these tasks remain underdeveloped. Especially for low-resource languages such as Swahili or Thai, LLMs can often misinterpret prompts or default to reasoning in English. This implicit bias toward high-resource languages undermines factual accuracy, interpretability, and trust. Current multilingual benchmarks focus only on final answers, overlooking whether models actually reason in the target language. To address this gap, we introduce GeoFact-X, a geography-based multilingual factual reasoning benchmark with annotated reasoning traces in five languages: English, Hindi, Japanese, Swahili, and Thai. We further propose BRIDGE, a novel training method that guides supervised fine-tuning and test-time reinforcement learning with a language-consistency reward to align reasoning with the input language. Finally, we develop an automatic evaluation protocol using LLM-as-a-judge to assess answer correctness and the quality and language consistency of reasoning traces, enabling nuanced and scalable analysis beyond surface-level metrics. Our results show that BRIDGE significantly enhances multilingual reasoning fidelity, demonstrating that reasoning-aware multilingual reinforcement learning is crucial for robust cross-lingual generalization. https://jd730.github.io/projects/GeoFact-X_BRIDGE
Translating natural language mathematical statements into formal, executable code is a fundamental challenge in automated theorem proving. While prior work has focused on generation and compilation success, little attention has been paid to the critic phase-the evaluation of whether generated formalizations truly capture the semantic intent of the original problem. In this paper, we introduce CriticLean, a novel critic-guided reinforcement learning framework that elevates the role of the critic from a passive validator to an active learning component. Specifically, first, we propose the CriticLeanGPT, trained via supervised fine-tuning and reinforcement learning, to rigorously assess the semantic fidelity of Lean 4 formalizations. Then, we introduce CriticLeanBench, a benchmark designed to measure models' ability to distinguish semantically correct from incorrect formalizations, and demonstrate that our trained CriticLeanGPT models can significantly outperform strong open- and closed-source baselines. Building on the CriticLean framework, we construct FineLeanCorpus, a dataset comprising over 285K problems that exhibits rich domain diversity, broad difficulty coverage, and high correctness based on human evaluation. Overall, our findings highlight that optimizing the critic phase is essential for producing reliable formalizations, and we hope our CriticLean will provide valuable insights for future advances in formal mathematical reasoning.
The diversity of tasks and dynamic nature of reinforcement learning (RL) require RL agents to be able to learn sequentially and continuously, a learning paradigm known as continuous reinforcement learning. This survey reviews how continual learning transforms RL agents into dynamic continual learners. This enables RL agents to acquire and retain useful and reusable knowledge seamlessly. The paper delves into fundamental aspects of continual reinforcement learning, exploring key concepts, significant challenges, and novel methodologies. Special emphasis is placed on recent advancements in continual reinforcement learning within robotics, along with a succinct overview of evaluation environments utilized in prominent research, facilitating accessibility for newcomers to the field. The review concludes with a discussion on limitations and promising future directions, providing valuable insights for researchers and practitioners alike.
Reinforcement Learning (RL) is an important machine learning paradigm for solving sequential decision-making problems. Recent years have witnessed remarkable progress in this field due to the rapid development of deep neural networks. However, the success of RL currently relies on extensive training data and computational resources. In addition, RL's limited ability to generalize across tasks restricts its applicability in dynamic and real-world environments. With the arisen of Continual Learning (CL), Continual Reinforcement Learning (CRL) has emerged as a promising research direction to address these limitations by enabling agents to learn continuously, adapt to new tasks, and retain previously acquired knowledge. In this survey, we provide a comprehensive examination of CRL, focusing on its core concepts, challenges, and methodologies. Firstly, we conduct a detailed review of existing works, organizing and analyzing their metrics, tasks, benchmarks, and scenario settings. Secondly, we propose a new taxonomy of CRL methods, categorizing them into four types from the perspective of knowledge storage and/or transfer. Finally, our analysis highlights the unique challenges of CRL and provides practical insights into future directions.
Automated parsing of scanned documents into richly structured, machine-readable formats remains a critical bottleneck in Document AI, as traditional multi-stage pipelines suffer from error propagation and limited adaptability to diverse layouts. We introduce layoutRL, an end-to-end reinforcement learning framework that trains models to be explicitly layout-aware by optimizing a composite reward of normalized edit distance, paragraph count accuracy, and reading order preservation. Leveraging our newly released dataset, Infinity-Doc-55K, which combines 55K high-fidelity synthetic scanned document parsing data with expert-filtered real-world documents, we instantiate layoutRL in a vision-language-model-based parser called Infinity-Parser. Evaluated on English and Chinese benchmarks for OCR, table and formula extraction, and reading order detection, Infinity-Parser achieves new state-of-the-art performance in both accuracy and structural fidelity, outpacing specialist pipelines and general-purpose vision-language models. We will publicly release our code and dataset to accelerate progress in robust document understanding.
We introduce a real-time strategy game environment built on Generals.io, a game that hosts thousands of active players each week across multiple game formats. Our environment is fully compatible with Gymnasium and PettingZoo, capable of running thousands of frames per second on commodity hardware. Our reference agent -- trained with supervised pre-training and self-play -- hits the top 0.003\% of the 1v1 human leaderboard after just 36 hours on a single H100 GPU. To accelerate learning, we incorporate potential-based reward shaping and memory features. Our contributions -- a modular RTS benchmark and a competitive, state-of-the-art baseline agent -- provide an accessible yet challenging platform for advancing multi-agent reinforcement learning research.
This paper introduces TransDreamerV3, a reinforcement learning model that enhances the DreamerV3 architecture by integrating a transformer encoder. The model is designed to improve memory and decision-making capabilities in complex environments. We conducted experiments on Atari-Boxing, Atari-Freeway, Atari-Pong, and Crafter tasks, where TransDreamerV3 demonstrated improved performance over DreamerV3, particularly in the Atari-Freeway and Crafter tasks. While issues in the Minecraft task and limited training across all tasks were noted, TransDreamerV3 displays advancement in world model-based reinforcement learning, leveraging transformer architectures.
Transfer learning in Reinforcement Learning (RL) enables agents to leverage knowledge from source tasks to accelerate learning in target tasks. While prior work, such as the Attend, Adapt, and Transfer (A2T) framework, addresses negative transfer and selective transfer, other critical challenges remain underexplored. This paper introduces the Generalized Adaptive Transfer Network (GATN), a deep RL architecture designed to tackle task generalization across domains, robustness to environmental changes, and computational efficiency in transfer. GATN employs a domain-agnostic representation module, a robustness-aware policy adapter, and an efficient transfer scheduler to achieve these goals. We evaluate GATN on diverse benchmarks, including Atari 2600, MuJoCo, and a custom chatbot dialogue environment, demonstrating superior performance in cross-domain generalization, resilience to dynamic environments, and reduced computational overhead compared to baselines. Our findings suggest GATN is a versatile framework for real-world RL applications, such as adaptive chatbots and robotic control.
Training large language models (LLMs) to act as autonomous agents for multi-turn, long-horizon tasks remains significant challenges in scalability and training efficiency. To address this, we introduce L-Zero (L0), a scalable, end-to-end training pipeline for general-purpose agents. Featuring a low-cost, extensible, and sandboxed concurrent agent worker pool, L0 lowers the barrier for applying reinforcement learning in complex environments. We also introduce NB-Agent, the agent scaffold within L0, which operates in a "code-as-action" fashion via a Read-Eval-Print-Loop (REPL). We evaluate L0 on factuality question-answering benchmarks. Our experiments demonstrate that a base model can develop robust problem-solving skills using solely Reinforcement Learning with Verifiable Rewards (RLVR). On the Qwen2.5-7B-Instruct model, our method boosts accuracy on SimpleQA from 30 % to 80 % and on HotpotQA from 22 % to 41 %. We have open-sourced the entire L0 system, including our L0 series models, the NB-Agent, a complete training pipeline, and the corresponding training recipes on (https://github.com/cmriat/l0).
Deep Reinforcement Learning (DRL) has achieved remarkable success in complex sequential decision-making tasks, such as playing Atari 2600 games and mastering board games. A critical yet underexplored aspect of DRL is the temporal scale of action execution. We propose a novel paradigm that integrates contextual bandits with DRL to adaptively select action durations, enhancing policy flexibility and computational efficiency. Our approach augments a Deep Q-Network (DQN) with a contextual bandit module that learns to choose optimal action repetition rates based on state contexts. Experiments on Atari 2600 games demonstrate significant performance improvements over static duration baselines, highlighting the efficacy of adaptive temporal abstractions in DRL. This paradigm offers a scalable solution for real-time applications like gaming and robotics, where dynamic action durations are critical.
Large language models can exhibit improved mathematical reasoning capabilities following post-training with instruction tuning, reinforcement learning, or knowledge distillation. However, it remains unclear whether these improvements are driven by major changes in transformer layers or from minor adjustments that leave the relative layer importance structures of the base model largely unchanged. We investigate this question through systematic layer-wise ablation experiments, examining base, instruction-tuned, knowledge-distilled, and reinforcement learning variants on mathematical reasoning benchmarks. Our findings show that mathematical reasoning gives rise to a specific layer importance structure, and this structure persists across all post-training paradigms. Removal of such layers causes accuracy drops of up to 80%. In contrast, non-mathematical tasks like factual recall exhibit no critical layers. This distinction suggests that mathematical reasoning requires specialized layers that emerge during pre-training, while other non-reasoning tasks do not. From an information-theoretic perspective, we also observe that these critical layers are the same layers where major representational transformation occurs.
This work introduces a growable and modular neural network architecture that naturally avoids catastrophic forgetting and interference in continual reinforcement learning. The structure of each module allows the selective combination of previous policies along with its internal policy, accelerating the learning process on the current task. Unlike previous growing neural network approaches, we show that the number of parameters of the proposed approach grows linearly with respect to the number of tasks, and does not sacrifice plasticity to scale. Experiments conducted in benchmark continuous control and visual problems reveal that the proposed approach achieves greater knowledge transfer and performance than alternative methods.