There is no limit to how much a robot might explore and learn, but all of that knowledge needs to be searchable and actionable. Within language research, retrieval augmented generation (RAG) has become the workhouse of large-scale non-parametric knowledge, however existing techniques do not directly transfer to the embodied domain, which is multimodal, data is highly correlated, and perception requires abstraction. To address these challenges, we introduce Embodied-RAG, a framework that enhances the foundational model of an embodied agent with a non-parametric memory system capable of autonomously constructing hierarchical knowledge for both navigation and language generation. Embodied-RAG handles a full range of spatial and semantic resolutions across diverse environments and query types, whether for a specific object or a holistic description of ambiance. At its core, Embodied-RAG's memory is structured as a semantic forest, storing language descriptions at varying levels of detail. This hierarchical organization allows the system to efficiently generate context-sensitive outputs across different robotic platforms. We demonstrate that Embodied-RAG effectively bridges RAG to the robotics domain, successfully handling over 200 explanation and navigation queries across 19 environments, highlighting its promise for general-purpose non-parametric system for embodied agents.

Conventional wisdom suggests that policy gradient methods are better suited to complex action spaces than action-value methods. However, foundational studies have shown equivalences between these paradigms in small and finite action spaces (O'Donoghue et al., 2017; Schulman et al., 2017a). This raises the question of why their computational applicability and performance diverge as the complexity of the action space increases. We hypothesize that the apparent superiority of policy gradients in such settings stems not from intrinsic qualities of the paradigm, but from universal principles that can also be applied to action-value methods to serve similar functionality. We identify three such principles and provide a framework for incorporating them into action-value methods. To support our hypothesis, we instantiate this framework in what we term QMLE, for Q-learning with maximum likelihood estimation. Our results show that QMLE can be applied to complex action spaces with a controllable computational cost that is comparable to that of policy gradient methods, all without using policy gradients. Furthermore, QMLE demonstrates strong performance on the DeepMind Control Suite, even when compared to the state-of-the-art methods such as DMPO and D4PG.

We study EgalMAB, an egalitarian assignment problem in the context of stochastic multi-armed bandits. In EgalMAB, an agent is tasked with assigning a set of users to arms. At each time step, the agent must assign exactly one arm to each user such that no two users are assigned to the same arm. Subsequently, each user obtains a reward drawn from the unknown reward distribution associated with its assigned arm. The agent's objective is to maximize the minimum expected cumulative reward among all users over a fixed horizon. This problem has applications in areas such as fairness in job and resource allocations, among others. We design and analyze a UCB-based policy EgalUCB and establish upper bounds on the cumulative regret. In complement, we establish an almost-matching policy-independent impossibility result.

Coordinating heterogeneous teams of mobile robots for tasks such as search and rescue is highly challenging. This is due to the complexities of perception, decision making and planning in such environments, with agents' non-synchronous operation, constrained communication, and limited computational resources. This paper presents the Cooperative and Asynchronous Transformer-based Mission Planning (CATMiP) framework, which leverages multi-agent reinforcement learning (MARL) to effectively coordinate agents with heterogeneous sensing, motion, and actuation capabilities. The framework introduces a Class-based Macro-Action Decentralized Partially Observable Markov Decision Process (CMD-POMDP) model to handle asynchronous decision-making among different agent classes via macro-actions. It also extends the Multi-Agent Transformer (MAT) architecture to facilitate distributed, ad hoc communication among the agents. CATMiP easily adapts to mission complexities and communication constraints, and scales to varying environment sizes and team compositions. Simulations demonstrate its scalability and ability to achieve cooperative mission objectives with two classes of explorer and rescuer agents, even under severe communication constraints. The code is available at https://github.com/mylad13/CATMiP.

Summary ----------------- The paper presents a novel actor-critic algorithm, named MADDPG, for both cooperative and competitive multiagent problems. MADDPG relies on a number of key ideas: 1) The action value functions are learned in a'centralized' manner, meaning that it takes into account the actions of all other players. This allows to evaluate the effect of the joint policy on each agents long term reward. To remove the need of knowing other agents' actions, the authors suggest that each agent could learn an approximate model of their policies. At each episode during the learning process, each agent draws uniformaly a policy from its ensemble.

The paper tackles a multi-agent credit assignment problem, an egregious problem within multi-agent systems by extending existing methods on difference rewards for settings in which the population of the system is large. Though the results are relevant and lead to an improvement for large population systems, the contribution is nonetheless limited to a modification of existing techniques for a specific setting which seemingly requires the number of agents to be large and for the agents to observe a count of the agents within their neighbourhood. The results of the paper enable improved credit assignment in the presence of noise from other agents' actions, an improved baseline leading to reduced variance and, in turn, better estimates of the collective policy gradient (under homogeneity assumptions). The analysis of the paper applies to a specific setting in which the reward function has a term that is common to all agents and therefore is not decomposable. The extent to which this property is to be found in multi-agent systems, however, is not discussed.

Summary: "A Unified Game-Theoretic Approach to Multiagent Reinforcement Learning" presents a novel scalable algorithm that is shown to converge to better behaviours in partially-observable Multi-Agent Reinforcement Learning scenarios compared to previous methods. The paper begins with describing the problem, mainly that training reinforcement learning agents independently (i.e. each agent ignores the behaviours of the other agents and treats them as part of the environment) results in policies which can significantly overfit to only the agent behaviours observed during training time, failing to generalize when later set against new opponent behaviours. The paper then describes its solution, a generalization of the Double Oracle algorithm. The algorithm works using the following process: first, given a set of initial policies for each player, an empirical payoff tensor is created and from that a meta-strategy is learnt for each player which is the mixture over that initial policy set which achieves the highest value. Then each player i in the game is iterated, and a new policy is trained against policies sampled from the meta-strategies of the other agents not equal to i.

If we accept that distributed learning is interesting, then this article presents a nice treatment of distributed mirror descent in which feedback may be asynchronous and delayed. Indeed, we are presented with a provably convergent learning algorithm for continuous action sets (in classes of games) even when individual players' feedback are received with differing levels of delay; further more the regret at time T is controlled as a function of the total delay to time T. This is a strong result, achieved by using a suite of very current proof techniques - lambda-Fenchel couplings serving as primula-dual Bregman divergences and associated tools. I have some concerns, but overall I think this is a good paper. If the concept of variational stability implies that all Nash equilibria of a game are in a closed and convex set, to me this is a major restriction on the class of games for which the result is relevant.

Agents that "cooperate" with each other can take on many forms. The algorithmic architecture described herein addresses one of these forms, that of a team of computer agents that share many aspects of the same "brain.".

This paper proposes several alternative extensions of GAIL to multi-agent imitation learning settings. The paper includes strong, positive results on a wide range of environments against a suitable selection of baselines. However, insufficient details of the environments is provided to reproduce or fully appreciate the complexity of these environments. If accepted, I would request the authors add these to the appendix and would appreciate details (space permitting) to be discussed in the rebuttal - particularly the state representation. The more pressing point I would like to raise for discussion in the rebuttal is with regard to the MACK algorithm proposed for the generator. The authors make a justified argument for the novelty of the algorithm, but do not thoroughly justify why they used this algorithm instead of an established MARL algorithm (e.g.