Continual reinforcement learning (CRL) is the study of optimal strategies for maximizing rewards in sequential environments that change over time. This is particularly crucial in domains such as robotics, where the operational environment is inherently dynamic and subject to continual change. Nevertheless, research in this area has thus far concentrated on off-policy algorithms with replay buffers that are capable of amortizing the impact of distribution shifts. Such an approach is not feasible with on-policy reinforcement learning algorithms that learn solely from the data obtained from the current policy. In this paper, we examine the performance of proximal policy optimization (PPO), a prevalent on-policy reinforcement learning (RL) algorithm, in a classical CRL benchmark. Our findings suggest that the current methods are suboptimal in terms of average performance. Nevertheless, they demonstrate encouraging competitive outcomes with respect to forward transfer and forgetting metrics. This highlights the need for further research into continual on-policy reinforcement learning. The source code is available at https://github.com/Teddy298/continualworld-ppo.

Large Language Models (LLMs) and Vision Language Models (VLMs) possess extensive knowledge and exhibit promising reasoning abilities, however, they still struggle to perform well in complex, dynamic environments. Real-world tasks require handling intricate interactions, advanced spatial reasoning, long-term planning, and continuous exploration of new strategies—areas in which we lack effective methodologies for comprehensively evaluating these capabilities. To address this gap, we introduce BALROG, a novel benchmark designed to assess the agentic capabilities of LLMs and VLMs through a diverse set of challenging games. Our benchmark incorporates a range of existing reinforcement learning environments with varying levels of difficulty, including tasks that are solvable by non-expert humans in seconds to extremely challenging ones that may take years to master (e.g., the NetHack Learning Environment). We devise fine-grained metrics to measure performance and conduct an extensive evaluation of several popular open-source and closed-source LLMs and VLMs. Our findings indicate that while current models achieve partial success in the easier games, they struggle significantly with more challenging tasks. Notably, we observe severe deficiencies in vision-based decision-making, as several models perform worse when visual representations of the environments are provided. We release BALROG as an open and user-friendly benchmark to facilitate future research and development in the agentic community. Code and Leaderboard at balrogai.com

The real world is messy and unstructured. Uncovering critical information often requires active, goal-driven exploration. It remains to be seen whether Vision-Language Models (VLMs), which recently emerged as a popular zero-shot tool in many difficult tasks, can operate effectively in such conditions. In this paper, we answer this question by introducing FlySearch, a 3D, outdoor, photorealistic environment for searching and navigating to objects in complex scenes. We define three sets of scenarios with varying difficulty and observe that state-of-the-art VLMs cannot reliably solve even the simplest exploration tasks, with the gap to human performance increasing as the tasks get harder. We identify a set of central causes, ranging from vision hallucination, through context misunderstanding, to task planning failures, and we show that some of them can be addressed by finetuning. We publicly release the benchmark, scenarios, and the underlying codebase.

Many complex tasks can be decomposed into simpler, independent parts. Discovering such underlying compositional structure has the potential to enable compositional generalization. Despite progress, our most powerful systems struggle to compose flexibly. It therefore seems natural to make models more modular to help capture the compositional nature of many tasks. However, it is unclear under which circumstances modular systems can discover hidden compositional structure. To shed light on this question, we study a teacher-student setting with a modular teacher where we have full control over the composition of ground truth modules. This allows us to relate the problem of compositional generalization to that of identification of the underlying modules. In particular we study modularity in hypernetworks representing a general class of multiplicative interactions. We show theoretically that identification up to linear transformation purely from demonstrations is possible without having to learn an exponential number of module combinations. We further demonstrate empirically that under the theoretically identified conditions, meta-learning from finite data can discover modular policies that generalize compositionally in a number of complex environments.

Fine-tuning is a widespread technique that allows practitioners to transfer pre-trained capabilities, as recently showcased by the successful applications of foundation models. However, fine-tuning reinforcement learning (RL) models remains a challenge. This work conceptualizes one specific cause of poor transfer, accentuated in the RL setting by the interplay between actions and observations: forgetting of pre-trained capabilities. Namely, a model deteriorates on the state subspace of the downstream task not visited in the initial phase of fine-tuning, on which the model behaved well due to pre-training. This way, we lose the anticipated transfer benefits. We identify conditions when this problem occurs, showing that it is common and, in many cases, catastrophic. Through a detailed empirical analysis of the challenging NetHack and Montezuma’s Revenge environments, we show that standard knowledge retention techniques mitigate the problem and thus allow us to take full advantage of the pre-trained capabilities. In particular, in NetHack, we achieve a new state-of-the-art for neural models, improving the previous best score from $5$K to over $10$K points in the Human Monk scenario.

Although modern generative models achieve excellent quality in a variety of tasks, they often lack the essential ability to generate examples with requested properties, such as the age of the person in the photo or the weight of the generated molecule. To overcome these limitations we propose PluGeN (Plugin Generative Network), a simple yet effective generative technique that can be used as a plugin for pre-trained generative models. The idea behind our approach is to transform the entangled latent representation using a flow-based module into a multi-dimensional space where the values of each attribute are modeled as an independent one-dimensional distribution. In consequence, PluGeN can generate new samples with desired attributes as well as manipulate labeled attributes of existing examples. Due to the disentangling of the latent representation, we are even able to generate samples with rare or unseen combinations of attributes in the dataset, such as a young person with gray hair, men with make-up, or women with beards. In contrast to competitive approaches, PluGeN can be trained on partially labeled data. We combined PluGeN with GAN and VAE models and applied it to conditional generation and manipulation of images, chemical molecule modeling and 3D point clouds generation.

The problem of reducing processing time of large deep learning models is a fundamental challenge in many real-world applications. Early exit methods strive towards this goal by attaching additional Internal Classifiers (ICs) to intermediate layers of a neural network. ICs can quickly return predictions for easy examples and, as a result, reduce the average inference time of the whole model. However, if a particular IC does not decide to return an answer early, its predictions are discarded, with its computations effectively being wasted. To solve this issue, we introduce Zero Time Waste (ZTW), a novel approach in which each IC reuses predictions returned by its predecessors by (1) adding direct connections between ICs and (2) combining previous outputs in an ensemble-like manner. We conduct extensive experiments across various multiple modes, datasets, and architectures to demonstrate that ZTW achieves a significantly better accuracy vs. inference time trade-off than other early exit methods. On the ImageNet dataset, it obtains superior results over the best baseline method in 11 out of 16 cases, reaching up to 5 percentage points of improvement on low computational budgets.

Modern generative models achieve excellent quality in a variety of tasks including image or text generation and chemical molecule modeling. However, existing methods often lack the essential ability to generate examples with requested properties, such as the age of the person in the photo or the weight of the generated molecule. Incorporating such additional conditioning factors would require rebuilding the entire architecture and optimizing the parameters from scratch. Moreover, it is difficult to disentangle selected attributes so that to perform edits of only one attribute while leaving the others unchanged. To overcome these limitations we propose PluGeN (Plugin Generative Network), a simple yet effective generative technique that can be used as a plugin to pre-trained generative models. The idea behind our approach is to transform the entangled latent representation using a flow-based module into a multi-dimensional space where the values of each attribute are modeled as an independent one-dimensional distribution. In consequence, PluGeN can generate new samples with desired attributes as well as manipulate labeled attributes of existing examples. Due to the disentangling of the latent representation, we are even able to generate samples with rare or unseen combinations of attributes in the dataset, such as a young person with gray hair, men with make-up, or women with beards. We combined PluGeN with GAN and VAE models and applied it to conditional generation and manipulation of images and chemical molecule modeling. Experiments demonstrate that PluGeN preserves the quality of backbone models while adding the ability to control the values of labeled attributes. Implementation is available at https://github.com/gmum/plugen.

We introduce a new training paradigm that enforces interval constraints on neural network parameter space to control forgetting. Contemporary Continual Learning (CL) methods focus on training neural networks efficiently from a stream of data, while reducing the negative impact of catastrophic forgetting, yet they do not provide any firm guarantees that network performance will not deteriorate uncontrollably over time. In this work, we show how to put bounds on forgetting by reformulating continual learning of a model as a continual contraction of its parameter space. To that end, we propose Hyperrectangle Training, a new training methodology where each task is represented by a hyperrectangle in the parameter space, fully contained in the hyperrectangles of the previous tasks. This formulation reduces the NP-hard CL problem back to polynomial time while providing full resilience against forgetting. We validate our claim by developing InterContiNet (Interval Continual Learning) algorithm which leverages interval arithmetic to effectively model parameter regions as hyperrectangles. Through experimental results, we show that our approach performs well in a continual learning setup without storing data from previous tasks.

In this paper we present the first safe system for full control of self-driving vehicles trained from human demonstrations and deployed in challenging, real-world, urban environments. Current industry-standard solutions use rule-based systems for planning. Although they perform reasonably well in common scenarios, the engineering complexity renders this approach incompatible with human-level performance. On the other hand, the performance of machine-learned (ML) planning solutions can be improved by simply adding more exemplar data. However, ML methods cannot offer safety guarantees and sometimes behave unpredictably. To combat this, our approach uses a simple yet effective rule-based fallback layer that performs sanity checks on an ML planner's decisions (e.g. avoiding collision, assuring physical feasibility). This allows us to leverage ML to handle complex situations while still assuring the safety, reducing ML planner-only collisions by 95%. We train our ML planner on 300 hours of expert driving demonstrations using imitation learning and deploy it along with the fallback layer in downtown San Francisco, where it takes complete control of a real vehicle and navigates a wide variety of challenging urban driving scenarios.

The ability of continual learning systems to transfer knowledge from previously seen tasks in order to maximize performance on new tasks is a significant challenge for the field, limiting the applicability of continual learning solutions to realistic scenarios. Consequently, this study aims to broaden our understanding of transfer and its driving forces in the specific case of continual reinforcement learning. We adopt SAC as the underlying RL algorithm and Continual World as a suite of continuous control tasks. We systematically study how different components of SAC (the actor and the critic, exploration, and data) affect transfer efficacy, and we provide recommendations regarding various modeling options. The best set of choices, dubbed ClonEx-SAC, is evaluated on the recent Continual World benchmark. ClonEx-SAC achieves 87% final success rate compared to 80% of PackNet, the best method in the benchmark. Moreover, the transfer grows from 0.18 to 0.54 according to the metric provided by Continual World.

Experience replay is a simple and well-performing strategy for continual learning problems, often used as a basis for more advanced methods. However, the dynamics of experience replay are not yet well understood. To showcase this, we focus on a single component of this problem, namely choosing the batch size of the buffer samples. We find that small batches perform much better at stopping forgetting than larger batches, contrary to the intuitive assumption that it is better to recall more samples from the past to avoid forgetting. We show that this phenomenon does not disappear under learning rate tuning and we propose possible directions for further analysis.

The problem of reducing processing time of large deep learning models is a fundamental challenge in many real-world applications. Early exit methods strive towards this goal by attaching additional Internal Classifiers (ICs) to intermediate layers of a neural network. ICs can quickly return predictions for easy examples and, as a result, reduce the average inference time of the whole model. However, if a particular IC does not decide to return an answer early, its predictions are discarded, with its computations effectively being wasted. To solve this issue, we introduce Zero Time Waste (ZTW), a novel approach in which each IC reuses predictions returned by its predecessors by (1) adding direct connections between ICs and (2) combining previous outputs in an ensemble-like manner. We conduct extensive experiments across various datasets and architectures to demonstrate that ZTW achieves a significantly better accuracy vs. inference time trade-off than other recently proposed early exit methods.

Continual learning (CL) --- the ability to continuously learn, building on previously acquired knowledge --- is a natural requirement for long-lived autonomous reinforcement learning (RL) agents. While building such agents, one needs to balance opposing desiderata, such as constraints on capacity and compute, the ability to not catastrophically forget, and to exhibit positive transfer on new tasks. Understanding the right trade-off is conceptually and computationally challenging, which we argue has led the community to overly focus on catastrophic forgetting. In response to these issues, we advocate for the need to prioritize forward transfer and propose Continual World, a benchmark consisting of realistic and meaningfully diverse robotic tasks built on top of Meta-World as a testbed. Following an in-depth empirical evaluation of existing CL methods, we pinpoint their limitations and highlight unique algorithmic challenges in the RL setting. Our benchmark aims to provide a meaningful and computationally inexpensive challenge for the community and thus help better understand the performance of existing and future solutions. Information about the benchmark, including the open-source code, is available at https://sites.google.com/view/continualworld.

We propose a semi-supervised generative model, SeGMA, which learns a joint probability distribution of data and their classes and is implemented in a typical Wasserstein autoencoder framework. We choose a mixture of Gaussians as a target distribution in latent space, which provides a natural splitting of data into clusters. To connect Gaussian components with correct classes, we use a small amount of labeled data and a Gaussian classifier induced by the target distribution. SeGMA is optimized efficiently due to the use of the Cramer-Wold distance as a maximum mean discrepancy penalty, which yields a closed-form expression for a mixture of spherical Gaussian components and, thus, obviates the need of sampling. While SeGMA preserves all properties of its semi-supervised predecessors and achieves at least as good generative performance on standard benchmark data sets, it presents additional features: 1) interpolation between any pair of points in the latent space produces realistically looking samples; 2) combining the interpolation property with disentangling of class and style information, SeGMA is able to perform continuous style transfer from one class to another; and 3) it is possible to change the intensity of class characteristics in a data point by moving the latent representation of the data point away from specific Gaussian components.