Natural language provides a rich and intuitive interface to describe robot tasks. For instance, commands such as "navigate to the left corner" or "pick up the can" lend themselves as more powerful and flexible alternatives to specifying (x, y) coordinates or joint configurations. Using language descriptions to specify outcomes, particularly when interfacing with a team of robots, is thus a more natural choice, and one that does not require specially-trained operators. Recent work on commanding robots with natural language tend to utilize large pretrained transformers [1] known as LLMs [2, 3, 4] or Large Multimodal Models (LMMs) [5] for both language processing and control. Often, the transformer receives a task and observation, and produces either an action or a sequence of actions to complete the task [6, 7, 8, 9]. The latter case reduces to open-loop control, which cannot adapt to uncertainty, while the former is limited by the high latency of these models, typically measured in seconds or hundreds of milliseconds, precluding them from dynamic scenarios.

Most multi-agent reinforcement learning approaches adopt two types of policy optimization methods that either update policy simultaneously or sequentially. Simultaneously updating policies of all agents introduces non-stationarity problem. Although sequentially updating policies agent-by-agent in an appropriate order improves policy performance, it is prone to low efficiency due to sequential execution, resulting in longer model training and execution time. Intuitively, partitioning policies of all agents according to their interdependence and updating joint policy batch-by-batch can effectively balance performance and efficiency. However, how to determine the optimal batch partition of policies and batch updating order are challenging problems. Firstly, a sequential batched policy updating scheme, B2MAPO (Batch by Batch Multi-Agent Policy Optimization), is proposed with a theoretical guarantee of the monotonic incrementally tightened bound. Secondly, a universal modulized plug-and-play B2MAPO hierarchical framework, which satisfies CTDE principle, is designed to conveniently integrate any MARL models to fully exploit and merge their merits, including policy optimality and inference efficiency. Next, a DAG-based B2MAPO algorithm is devised, which is a carefully designed implementation of B2MAPO framework. Comprehensive experimental results conducted on StarCraftII Multi-agent Challenge and Google Football Research demonstrate the performance of DAG-based B2MAPO algorithm outperforms baseline methods. Meanwhile, compared with A2PO, our algorithm reduces the model training and execution time by 60.4% and 78.7%, respectively.

One important property of DIstribution Correction Estimation (DICE) methods is that the solution is the optimal stationary distribution ratio between the optimized and data collection policy. In this work, we show that DICE-based methods can be viewed as a transformation from the behavior distribution to the optimal policy distribution. Based on this, we propose a novel approach, Diffusion-DICE, that directly performs this transformation using diffusion models. We find that the optimal policy's score function can be decomposed into two terms: the behavior policy's score function and the gradient of a guidance term which depends on the optimal distribution ratio. The first term can be obtained from a diffusion model trained on the dataset and we propose an in-sample learning objective to learn the second term. Due to the multi-modality contained in the optimal policy distribution, the transformation in Diffusion-DICE may guide towards those local-optimal modes. We thus generate a few candidate actions and carefully select from them to approach global-optimum. Different from all other diffusion-based offline RL methods, the guide-then-select paradigm in Diffusion-DICE only uses in-sample actions for training and brings minimal error exploitation in the value function. We use a didatic toycase example to show how previous diffusion-based methods fail to generate optimal actions due to leveraging these errors and how Diffusion-DICE successfully avoids that. We then conduct extensive experiments on benchmark datasets to show the strong performance of Diffusion-DICE.

The integration of artificial intelligence across multiple domains has emphasized the importance of replicating human-like cognitive processes in AI. By incorporating emotional intelligence into AI agents, their emotional stability can be evaluated to enhance their resilience and dependability in critical decision-making tasks. In this work, we develop a methodology for modeling psychological disorders using Reinforcement Learning (RL) agents. We utilized Appraisal theory to train RL agents in a dynamic grid world environment with an Appraisal-Guided Proximal Policy Optimization (AG-PPO) algorithm. Additionally, we investigated numerous reward-shaping strategies to simulate psychological disorders and regulate the behavior of the agents. A comparison of various configurations of the modified PPO algorithm identified variants that simulate Anxiety disorder and Obsessive-Compulsive Disorder (OCD)-like behavior in agents. Furthermore, we compared standard PPO with AG-PPO and its configurations, highlighting the performance improvement in terms of generalization capabilities. Finally, we conducted an analysis of the agents' behavioral patterns in complex test environments to evaluate the associated symptoms corresponding to the psychological disorders. Overall, our work showcases the benefits of the appraisal-guided PPO algorithm over the standard PPO algorithm and the potential to simulate psychological disorders in a controlled artificial environment and evaluate them on RL agents.

Interpretability of deep reinforcement learning systems could assist operators with understanding how they interact with their environment. Vector quantization methods -- also called codebook methods -- discretize a neural network's latent space that is often suggested to yield emergent interpretability. We investigate whether vector quantization in fact provides interpretability in model-based reinforcement learning. Our experiments, conducted in the reinforcement learning environment Crafter, show that the codes of vector quantization models are inconsistent, have no guarantee of uniqueness, and have a limited impact on concept disentanglement, all of which are necessary traits for interpretability. We share insights on why vector quantization may be fundamentally insufficient for model interpretability.

As Deep Reinforcement Learning (Deep RL) research moves towards solving large-scale worlds, efficient environment simulations become crucial for rapid experimentation. However, most existing environments struggle to scale to high throughput, setting back meaningful progress. Interactions are typically computed on the CPU, limiting training speed and throughput, due to slower computation and communication overhead when distributing the task across multiple machines. Ultimately, Deep RL training is CPU-bound, and developing batched, fast, and scalable environments has become a frontier for progress. Among the most used Reinforcement Learning (RL) environments, MiniGrid is at the foundation of several studies on exploration, curriculum learning, representation learning, diversity, meta-learning, credit assignment, and language-conditioned RL, and still suffers from the limitations described above. In this work, we introduce NAVIX, a re-implementation of MiniGrid in JAX. NAVIX achieves over 200 000x speed improvements in batch mode, supporting up to 2048 agents in parallel on a single Nvidia A100 80 GB. This reduces experiment times from one week to 15 minutes, promoting faster design iterations and more scalable RL model development.

As service systems grow increasingly complex and dynamic, many interventions become localized, available and taking effect only in specific states. This paper investigates experiments with local treatments on a widely-used class of dynamic models, Markov Decision Processes (MDPs). Particularly, we focus on utilizing the local structure to improve the inference efficiency of the average treatment effect. We begin by demonstrating the efficiency of classical inference methods, including model-based estimation and temporal difference learning under a fixed policy, as well as classical A/B testing with general treatments. We then introduce a variance reduction technique that exploits the local treatment structure by sharing information for states unaffected by the treatment policy. Our new estimator effectively overcomes the variance lower bound for general treatments while matching the more stringent lower bound incorporating the local treatment structure. Furthermore, our estimator can optimally achieve a linear reduction with the number of test arms for a major part of the variance. Finally, we explore scenarios with perfect knowledge of the control arm and design estimators that further improve inference efficiency.

Federated learning combined with blockchain empowers secure data sharing in autonomous driving applications. Nevertheless, with the increasing granularity and complexity of vehicle-generated data, the lack of data quality audits raises concerns about multi-party mistrust in trajectory prediction tasks. In response, this paper proposes an asynchronous federated learning data sharing method based on an interpretable reputation quantization mechanism utilizing graph neural network tools. Data providers share data structures under differential privacy constraints to ensure security while reducing redundant data. We implement deep reinforcement learning to categorize vehicles by reputation level, which optimizes the aggregation efficiency of federated learning. Experimental results demonstrate that the proposed data sharing scheme not only reinforces the security of the trajectory prediction task but also enhances prediction accuracy.

Offline reinforcement learning has shown promise for solving tasks in safety-critical settings, such as clinical decision support. Its application, however, has been limited by the lack of interpretability and interactivity for clinicians. To address these challenges, we propose the medical decision transformer (MeDT), a novel and versatile framework based on the goal-conditioned reinforcement learning paradigm for sepsis treatment recommendation. MeDT uses the decision transformer architecture to learn a policy for drug dosage recommendation. During offline training, MeDT utilizes collected treatment trajectories to predict administered treatments for each time step, incorporating known treatment outcomes, target acuity scores, past treatment decisions, and current and past medical states. This analysis enables MeDT to capture complex dependencies among a patient's medical history, treatment decisions, outcomes, and short-term effects on stability. Our proposed conditioning uses acuity scores to address sparse reward issues and to facilitate clinician-model interactions, enhancing decision-making. Following training, MeDT can generate tailored treatment recommendations by conditioning on the desired positive outcome (survival) and user-specified short-term stability improvements. We carry out rigorous experiments on data from the MIMIC-III dataset and use off-policy evaluation to demonstrate that MeDT recommends interventions that outperform or are competitive with existing offline reinforcement learning methods while enabling a more interpretable, personalized and clinician-directed approach.

Reinforcement learning practitioners often avoid hierarchical policies, especially in image-based observation spaces. Typically, the single-task performance improvement over flat-policy counterparts does not justify the additional complexity associated with implementing a hierarchy. However, by introducing multiple decision-making levels, hierarchical policies can compose lower-level policies to more effectively generalize between tasks, highlighting the need for multi-task evaluations. We analyze the benefits of hierarchy through simulated multi-task robotic control experiments from pixels. Our results show that hierarchical policies trained with task conditioning can (1) increase performance on training tasks, (2) lead to improved reward and state-space generalizations in similar tasks, and (3) decrease the complexity of fine tuning required to solve novel tasks. Thus, we believe that hierarchical policies should be considered when building reinforcement learning architectures capable of generalizing between tasks.