One approach to meet the challenges of deep lifelong reinforcement learning (LRL) is careful management of the agent's learning experiences, to learn (without forgetting) and build internal meta-models (of the tasks, environments, agents, and world). Generative replay (GR) is a biologically inspired replay mechanism that augments learning experiences with self-labelled examples drawn from an internal generative model that is updated over time. We present a version of GR for LRL that satisfies two desiderata: (a) Introspective density modelling of the latent representations of policies learned using deep RL, and (b) Model-free end-to-end learning. In this paper, we study three deep learning architectures for model-free GR, starting from a na\"ive GR and adding ingredients to achieve (a) and (b). We evaluate our proposed algorithms on three different scenarios comprising tasks from the Starcraft-2 and Minigrid domains. We report several key findings showing the impact of the design choices on quantitative metrics that include transfer learning, generalization to unseen tasks, fast adaptation after task change, performance wrt task expert, and catastrophic forgetting. We observe that our GR prevents drift in the features-to-action mapping from the latent vector space of a deep RL agent. We also show improvements in established lifelong learning metrics. We find that a small random replay buffer significantly increases the stability of training. Overall, we find that "hidden replay" (a well-known architecture for class-incremental classification) is the most promising approach that pushes the state-of-the-art in GR for LRL and observe that the architecture of the sleep model might be more important for improving performance than the types of replay used. Our experiments required only 6% of training samples to achieve 80-90% of expert performance in most Starcraft-2 scenarios.

Terahertz (THz) wireless networks are expected to catalyze the beyond fifth generation (B5G) era. However, due to the directional nature and the line-of-sight demand of THz links, as well as the ultra-dense deployment of THz networks, a number of challenges that the medium access control (MAC) layer needs to face are created. In more detail, the need of rethinking user association and resource allocation strategies by incorporating artificial intelligence (AI) capable of providing "real-time" solutions in complex and frequently changing environments becomes evident. Moreover, to satisfy the ultra-reliability and low-latency demands of several B5G applications, novel mobility management approaches are required. Motivated by this, this article presents a holistic MAC layer approach that enables intelligent user association and resource allocation, as well as flexible and adaptive mobility management, while maximizing systems' reliability through blockage minimization. In more detail, a fast and centralized joint user association, radio resource allocation, and blockage avoidance by means of a novel metaheuristic-machine learning framework is documented, that maximizes the THz networks performance, while minimizing the association latency by approximately three orders of magnitude. To support, within the access point (AP) coverage area, mobility management and blockage avoidance, a deep reinforcement learning (DRL) approach for beam-selection is discussed. Finally, to support user mobility between coverage areas of neighbor APs, a proactive hand-over mechanism based on AI-assisted fast channel prediction is~reported.

Deep reinforcement learning (RL) algorithms suffer severe performance degradation when the interaction data is scarce, which limits their real-world application. Recently, visual representation learning has been shown to be effective and promising for boosting sample efficiency in RL. These methods usually rely on contrastive learning and data augmentation to train a transition model for state prediction, which is different from how the model is used in RL--performing value-based planning. Accordingly, the learned representation by these visual methods may be good for recognition but not optimal for estimating state value and solving the decision problem. To address this issue, we propose a novel method, called value-consistent representation learning (VCR), to learn representations that are directly related to decision-making. More specifically, VCR trains a model to predict the future state (also referred to as the ''imagined state'') based on the current one and a sequence of actions. Instead of aligning this imagined state with a real state returned by the environment, VCR applies a $Q$-value head on both states and obtains two distributions of action values. Then a distance is computed and minimized to force the imagined state to produce a similar action value prediction as that by the real state. We develop two implementations of the above idea for the discrete and continuous action spaces respectively. We conduct experiments on Atari 100K and DeepMind Control Suite benchmarks to validate their effectiveness for improving sample efficiency. It has been demonstrated that our methods achieve new state-of-the-art performance for search-free RL algorithms.

Such knowledge could be extremely useful to robots aiming to act upon high-level, temporally extended instructions expressed in natural language. However, a significant weakness of language models is that they lack real-world experience, which makes it difficult to leverage them for decision making within a given embodiment. For example, asking a language model to describe how to clean a spill might result in a reasonable narrative, but it may not be applicable to a particular agent, such as a robot, that needs to perform this task in a particular environment. We propose to provide real-world grounding by means of pretrained skills, which are used to constrain the model to propose natural language actions that are both feasible and contextually appropriate. The robot can act as the language model's "hands and eyes," while the language model supplies high-level semantic knowledge about the task. We show how low-level skills can be combined with large language models so that the language model provides high-level knowledge about the procedures for performing complex and temporally extended instructions, while value functions associated with these skills provide the grounding necessary to connect this knowledge to a particular physical environment. We evaluate our method on a number of real-world robotic tasks, where we show the need for real-world grounding and that this approach is capable of completing long-horizon, abstract, natural language instructions on a mobile manipulator. The project's website, the video, and open sourced code in a tabletop domain can be found at say-can.github.io. Figure 1: LLMs have not interacted with their environment and observed the outcome of their responses, and thus are not grounded in the world. SayCan grounds LLMs via value functions of pretrained skills, allowing them to execute real-world, abstract, long-horizon commands on robots.

Ad hoc teamwork is the research problem of designing agents that can collaborate with new teammates without prior coordination. This survey makes a two-fold contribution: First, it provides a structured description of the different facets of the ad hoc teamwork problem. Second, it discusses the progress that has been made in the field so far, and identifies the immediate and long-term open problems that need to be addressed in ad hoc teamwork.

This paper presents a Deep Reinforcement Learning based navigation approach in which we define the occupancy observations as heuristic evaluations of motion primitives, rather than using raw sensor data. Our method enables fast mapping of the occupancy data, generated by multi-sensor fusion, into trajectory values in 3D workspace. The computationally efficient trajectory evaluation allows dense sampling of the action space. We utilize our occupancy observations in different data structures to analyze their effects on both training process and navigation performance. We train and test our methodology on two different robots within challenging physics-based simulation environments including static and dynamic obstacles. We benchmark our occupancy representations with other conventional data structures from state-of-the-art methods. The trained navigation policies are also validated successfully with physical robots in dynamic environments. The results show that our method not only decreases the required training time but also improves the navigation performance as compared to other occupancy representations. The open-source implementation of our work and all related info are available at \url{https://github.com/RIVeR-Lab/tentabot}.

Dulac-Arnold et Games have facilitated the rapid development of RL algorithms al. [11] propose to choose actions in a small subset of the in recent years. Card games, as a classical type action space to speed up the action search process. This set of games, also pose many challenges to RL algorithms. The is chosen based on a proper action encoding method which direct applications of generic algorithms [1]-[4] in card games usually relies on prior knowledge. However, the prior human are problematic in many aspects because of the large-scale knowledge for our problem is hard to obtain due to the discrete action space [5]. Prior works have proposed RL diversity of the teams and the strategies. Chandak et al. [12] methods to approach a number of traditional card games, like propose an algorithm to learn action representations from the Texas Hold'em [6]-[8], Mahjong [9], DouDizhu [5], [10], etc. consequences of corresponding actions.

One of the common ways children learn is by mimicking adults. Imitation learning focuses on learning policies with suitable performance from demonstrations generated by an expert, with an unspecified performance measure, and unobserved reward signal. Popular methods for imitation learning start by either directly mimicking the behavior policy of an expert (behavior cloning) or by learning a reward function that prioritizes observed expert trajectories (inverse reinforcement learning). However, these methods rely on the assumption that covariates used by the expert to determine her/his actions are fully observed. In this paper, we relax this assumption and study imitation learning when sensory inputs of the learner and the expert differ. First, we provide a non-parametric, graphical criterion that is complete (both necessary and sufficient) for determining the feasibility of imitation from the combinations of demonstration data and qualitative assumptions about the underlying environment, represented in the form of a causal model. We then show that when such a criterion does not hold, imitation could still be feasible by exploiting quantitative knowledge of the expert trajectories. Finally, we develop an efficient procedure for learning the imitating policy from experts' trajectories.

The StarCraft II Multi-Agent Challenge (SMAC) was created to be a challenging benchmark problem for cooperative multi-agent reinforcement learning (MARL). SMAC focuses exclusively on the problem of StarCraft micromanagement and assumes that each unit is controlled individually by a learning agent that acts independently and only possesses local information; centralized training is assumed to occur with decentralized execution (CTDE). To perform well in SMAC, MARL algorithms must handle the dual problems of multi-agent credit assignment and joint action evaluation. This paper introduces a new architecture TransMix, a transformer-based joint action-value mixing network which we show to be efficient and scalable as compared to the other state-of-the-art cooperative MARL solutions. TransMix leverages the ability of transformers to learn a richer mixing function for combining the agents' individual value functions. It achieves comparable performance to previous work on easy SMAC scenarios and outperforms other techniques on hard scenarios, as well as scenarios that are corrupted with Gaussian noise to simulate fog of war.

Practising and honing skills forms a fundamental component of how humans learn, yet artificial agents are rarely specifically trained to perform them. Instead, they are usually trained end-to-end, with the hope being that useful skills will be implicitly learned in order to maximise discounted return of some extrinsic reward function. In this paper, we investigate how skills can be incorporated into the training of reinforcement learning (RL) agents in complex environments with large state-action spaces and sparse rewards. To this end, we created SkillHack, a benchmark of tasks and associated skills based on the game of NetHack. We evaluate a number of baselines on this benchmark, as well as our own novel skill-based method Hierarchical Kickstarting (HKS), which is shown to outperform all other evaluated methods. Our experiments show that learning with a prior knowledge of useful skills can significantly improve the performance of agents on complex problems. We ultimately argue that utilising predefined skills provides a useful inductive bias for RL problems, especially those with large state-action spaces and sparse rewards.