Recent work has shown that supervised pre-training on learning histories of RL algorithms results in a model that captures the learning process and is able to improve in-context on novel tasks through interactions with an environment. Despite the progress in this area, there is still a gap in the existing literature, particularly in the in-context generalization to new action spaces. While existing methods show high performance on new tasks created by different reward distributions, their architectural design and training process are not suited for the introduction of new actions during evaluation. We aim to bridge this gap by developing an architecture and training methodology specifically for the task of generalizing to new action spaces. Inspired by Headless LLM, we remove the dependence on the number of actions by directly predicting the action embeddings. Furthermore, we use random embeddings to force the semantic inference of actions from context and to prepare for the new unseen embeddings during test time. Using multi-armed bandit environments with a variable number of arms, we show that our model achieves the performance of the data generation algorithm without requiring retraining for each new environment.
In-Context Reinforcement Learning is an emerging field with great potential for advancing Artificial Intelligence. Its core capability lies in generalizing to unseen tasks through interaction with the environment. To master these capabilities, an agent must be trained on specifically curated data that includes a policy improvement that an algorithm seeks to extract and then apply in context in the environment. However, for numerous tasks, training RL agents may be unfeasible, while obtaining human demonstrations can be relatively easy. Additionally, it is rare to be given the optimal policy, typically, only suboptimal demonstrations are available. We propose $AD^{\epsilon}$, a method that leverages demonstrations without policy improvement and enables multi-task in-context learning in the presence of a suboptimal demonstrator. This is achieved by artificially creating a history of incremental improvement, wherein noise is systematically introduced into the demonstrator's policy. Consequently, each successive transition illustrates a marginally better trajectory than the previous one. Our approach was tested on the Dark Room and Dark Key-to-Door environments, resulting in over a $\textbf{2}$x improvement compared to the best available policy in the data.
We present XLand-MiniGrid, a suite of tools and grid-world environments for meta-reinforcement learning research inspired by the diversity and depth of XLand and the simplicity and minimalism of MiniGrid. XLand-Minigrid is written in JAX, designed to be highly scalable, and can potentially run on GPU or TPU accelerators, democratizing large-scale experimentation with limited resources. To demonstrate the generality of our library, we have implemented some well-known single-task environments as well as new meta-learning environments capable of generating $10^8$ distinct tasks. We have empirically shown that the proposed environments can scale up to $2^{13}$ parallel instances on the GPU, reaching tens of millions of steps per second.
NetHack is known as the frontier of reinforcement learning research where learning-based methods still need to catch up to rule-based solutions. One of the promising directions for a breakthrough is using pre-collected datasets similar to recent developments in robotics, recommender systems, and more under the umbrella of offline reinforcement learning (ORL). Recently, a large-scale NetHack dataset was released; while it was a necessary step forward, it has yet to gain wide adoption in the ORL community. In this work, we argue that there are three major obstacles for adoption: tool-wise, implementation-wise, and benchmark-wise. To address them, we develop an open-source library that provides workflow fundamentals familiar to the ORL community: pre-defined D4RL-style tasks, uncluttered baseline implementations, and reliable evaluation tools with accompanying configs and logs synced to the cloud.
Recent years have witnessed significant advancements in offline reinforcement learning (RL), resulting in the development of numerous algorithms with varying degrees of complexity. While these algorithms have led to noteworthy improvements, many incorporate seemingly minor design choices that impact their effectiveness beyond core algorithmic advances. However, the effect of these design choices on established baselines remains understudied. In this work, we aim to bridge this gap by conducting a retrospective analysis of recent works in offline RL and propose ReBRAC, a minimalistic algorithm that integrates such design elements built on top of the TD3+BC method. We evaluate ReBRAC on 51 datasets with both proprioceptive and visual state spaces using D4RL and V-D4RL benchmarks, demonstrating its state-of-the-art performance among ensemble-free methods. To further illustrate the efficacy of these design choices, we perform a large-scale ablation study and hyperparameter sensitivity analysis on the scale of thousands of experiments.
Despite the success of Random Network Distillation (RND) in various domains, it was shown as not discriminative enough to be used as an uncertainty estimator for penalizing out-of-distribution actions in offline reinforcement learning. In this paper, we revisit these results and show that, with a naive choice of conditioning for the RND prior, it becomes infeasible for the actor to effectively minimize the anti-exploration bonus and discriminativity is not an issue. We show that this limitation can be avoided with conditioning based on Feature-wise Linear Modulation (FiLM), resulting in a simple and efficient ensemble-free algorithm based on Soft Actor-Critic. We evaluate it on the D4RL benchmark, showing that it is capable of achieving performance comparable to ensemble-based methods and outperforming ensemble-free approaches by a wide margin.
Offline reinforcement learning aims to train a policy on a pre-recorded and fixed dataset without any additional environment interactions. There are two major challenges in this setting: (1) extrapolation error caused by approximating the value of state-action pairs not well-covered by the training data and (2) distributional shift between behavior and inference policies. One way to tackle these problems is to induce conservatism - i.e., keeping the learned policies closer to the behavioral ones. To achieve this, we build upon recent works on learning policies in latent action spaces and use a special form of Normalizing Flows for constructing a generative model, which we use as a conservative action encoder. This Normalizing Flows action encoder is pre-trained in a supervised manner on the offline dataset, and then an additional policy model - controller in the latent space - is trained via reinforcement learning. This approach avoids querying actions outside of the training dataset and therefore does not require additional regularization for out-of-dataset actions. We evaluate our method on various locomotion and navigation tasks, demonstrating that our approach outperforms recently proposed algorithms with generative action models on a large portion of datasets.
Training large neural networks is known to be time-consuming, with the learning duration taking days or even weeks. To address this problem, large-batch optimization was introduced. This approach demonstrated that scaling mini-batch sizes with appropriate learning rate adjustments can speed up the training process by orders of magnitude. While long training time was not typically a major issue for model-free deep offline RL algorithms, recently introduced Q-ensemble methods achieving state-of-the-art performance made this issue more relevant, notably extending the training duration. In this work, we demonstrate how this class of methods can benefit from large-batch optimization, which is commonly overlooked by the deep offline RL community. We show that scaling the mini-batch size and naively adjusting the learning rate allows for (1) a reduced size of the Q-ensemble, (2) stronger penalization of out-of-distribution actions, and (3) improved convergence time, effectively shortening training duration by 3-4x times on average.
CORL is an open-source library that provides single-file implementations of Deep Offline Reinforcement Learning algorithms. It emphasizes a simple developing experience with a straightforward codebase and a modern analysis tracking tool. In CORL, we isolate methods implementation into distinct single files, making performance-relevant details easier to recognise. Additionally, an experiment tracking feature is available to help log metrics, hyperparameters, dependencies, and more to the cloud. Finally, we have ensured the reliability of the implementations by benchmarking a commonly employed D4RL benchmark. The source code can be found https://github.com/tinkoff-ai/CORL
Over the recent years, vast progress has been made in Offline Reinforcement Learning (Offline-RL) for various decision-making domains: from finance to robotics. However, comparing and reporting new Offline-RL algorithms has been noted as underdeveloped: (1) use of unlimited online evaluation budget for hyperparameter search (2) sidestepping offline policy selection (3) ad-hoc performance statistics reporting. In this work, we propose an evaluation technique addressing these issues, Expected Online Performance, that provides a performance estimate for a best-found policy given a fixed online evaluation budget. Using our approach, we can estimate the number of online evaluations required to surpass a given behavioral policy performance. Applying it to several Offline-RL baselines, we find that with a limited online evaluation budget, (1) Behavioral Cloning constitutes a strong baseline over various expert levels and data regimes, and (2) offline uniform policy selection is competitive with value-based approaches. We hope the proposed technique will make it into the toolsets of Offline-RL practitioners to help them arrive at informed conclusions when deploying RL in real-world systems.