The surge in reinforcement learning (RL) applications gave rise to diverse supporting technology, such as RL frameworks. However, the architectural patterns of these frameworks are inconsistent across implementations and there exists no reference architecture (RA) to form a common basis of comparison, evaluation, and integration. To address this gap, we propose an RA of RL frameworks. Through a grounded theory approach, we analyze 18 state-of-the-practice RL frameworks and, by that, we identify recurring architectural components and their relationships, and codify them in an RA. To demonstrate our RA, we reconstruct characteristic RL patterns. Finally, we identify architectural trends, e.g., commonly used components, and outline paths to improving RL frameworks.
Figures & tables
Fig. 1 : High-level architectural view of RL frameworks. (Component groups are underlined and correspond to the subsections of Sec. IV .)
Reinforcement learning studies how agents learn sequential decisions from rewards and interaction, balancing exploration, performance, safety, and adaptation.
Practitioners use reinforcement learning methods for language models through training frameworks, so a method that a framework includes is one configuration change away for its users. We measure how often this happens. We trace seven open-source frameworks for RL post-training through their commit histories and link them to 6,939 screened arXiv papers from January 2025 to June 2026. Of 1,598 papers that propose a general RL training method, 37 reached at least one framework. Across five quarterly cohorts of papers, the number included within 180 days stayed between 5 and 7 while the cohorts grew from 72 to 386 papers. Compared with papers from the same quarters, included papers more often trained policies of 30B parameters or more, at 41 against 10 percent, and more often changed a coefficient, clip, mask or weight in the existing loss, at 29 against 9 percent. Both gaps hold without papers from the companies that maintain the frameworks, while code release was equally common in both groups. The frameworks' issues and pull requests show that few methods are turned away in public. More than nine in ten of the general methods no framework included have no thread there that cites or names them, and many proposals that do appear stay open without a reply. Two further records, the frameworks' issues and the technical reports of 13 large open-weight models, show that users rarely report training failures in public and that the reports state 8 of the 13 widely included methods. We release the data as a tracker that links training failures to framework implementations and lists each framework's contribution rules and earlier proposals.
Reinforcement learning (RL) has achieved notable performance in high-dimensional sequential decision-making tasks, yet remains limited by low sample efficiency, sensitivity to noise, and weak generalization under partial observability. Most existing approaches address these issues primarily through optimization strategies, while the role of architectural priors in shaping representation learning and decision dynamics is less explored. Inspired by structural principles of the cerebellum, we propose a biologically grounded RL architecture that incorporate large expansion, sparse connectivity, sparse activation, and dendritic-level modulation. Experiments on noisy, high-dimensional RL benchmarks show that both the cerebellar architecture and dendritic modulation consistently improve sample efficiency, robustness, and generalization compared to conventional designs. Sensitivity analysis of architectural parameters suggests that cerebellum-inspired structures can offer optimized performance for RL with constrained model parameters. Overall, our work underscores the value of cerebellar structural priors as effective inductive biases for RL.
Sibo Zhang, Rui Jing, Liangfu Lv +2
Academy of Medical Engineering and Translational Medicine, Tianjin University, Tianjin, China.
Existing reinforcement learning (RL) largely focuses on optimizing learning algorithms within a pre-defined task formulation, while applying RL to new tasks still relies heavily on human expertise to design the learning pipeline: what an agent observes, how it is rewarded, which algorithm it uses, and how training is configured. Automating these choices is challenging because they are tightly coupled. We introduce FARL (Fully Automated Reinforcement Learning), a framework that automates RL design by jointly searching observation and reward programs, learning algorithms, and training settings. FARL uses training statistics and observed behavior to diagnose failures, propose changes, and test whether they improve learning. Its search preserves diverse candidates and the controls needed for these tests, using a staged budget to screen designs before investing in longer training. Experimental memory records each finding together with its conditions, so an unsuccessful component can be reconsidered in a different combination. Across expert-withheld and expert-available settings, FARL outperforms the evaluated automated baselines on challenging navigation, locomotion, and manipulation tasks. On Adroit Hand Relocate, FARL achieves 97.1% success at 5M training steps per policy, compared with 0% for Eureka and LiMEN and 1.0% for native Proximal Policy Optimization (PPO) under the same policy training budget.
Tianyu Jin, Liuyu Xiang, Chaowen Yang +3
Beijing University of Posts and Telecommunications, Beijing, China