This paper analyzes reinforcement learning (RL) algorithms for Markov decision processes (MDPs) under the average-reward criterion. We focus on Q-learning algorithms based on relative value iteration (RVI), which are model-free stochastic analogues of the classical RVI method for average-reward MDPs. These algorithms have low per-iteration complexity, making them well-suited for large state space problems. We extend the almost-sure convergence analysis of RVI Q-learning algorithms developed by Abounadi, Bertsekas, and Borkar (2001) from unichain to weakly communicating MDPs. This extension is important both practically and theoretically: weakly communicating MDPs cover a much broader range of applications compared to unichain MDPs, and their optimality equations have a richer solution structure (with multiple degrees of freedom), introducing additional complexity in proving algorithmic convergence. We also characterize the sets to which RVI Q-learning algorithms converge, showing that they are compact, connected, potentially nonconvex, and comprised of solutions to the average-reward optimality equation, with exactly one less degree of freedom than the general solution set of this equation. Furthermore, we extend our analysis to two RVI-based hierarchical average-reward RL algorithms using the options framework, proving their almost-sure convergence and characterizing their sets of convergence under the assumption that the underlying semi-Markov decision process is weakly communicating.

What data or environments to use for training to improve downstream performance is a longstanding and very topical question in reinforcement learning. In particular, Unsupervised Environment Design (UED) methods have gained recent attention as their adaptive curricula enable agents to be robust to in- and out-of-distribution tasks. We ask to what extent these methods are themselves robust when applied to a novel setting, closely inspired by a real-world robotics problem. Surprisingly, we find that the state-of-the-art UED methods either do not improve upon the na\"{i}ve baseline of Domain Randomisation (DR), or require substantial hyperparameter tuning to do so. Our analysis shows that this is due to their underlying scoring functions failing to predict intuitive measures of ``learnability'', i.e., in finding the settings that the agent sometimes solves, but not always. Based on this, we instead directly train on levels with high learnability and find that this simple and intuitive approach outperforms UED methods and DR in several binary-outcome environments, including on our domain and the standard UED domain of Minigrid. We further introduce a new adversarial evaluation procedure for directly measuring robustness, closely mirroring the conditional value at risk (CVaR). We open-source all our code and present visualisations of final policies here: https://github.com/amacrutherford/sampling-for-learnability.

Learning a transition model via Maximum Likelihood Estimation (MLE) followed by planning inside the learned model is perhaps the most standard and simplest Model-based Reinforcement Learning (RL) framework. In this work, we show that such a simple Model-based RL scheme, when equipped with optimistic and pessimistic planning procedures, achieves strong regret and sample complexity bounds in online and offline RL settings. Particularly, we demonstrate that under the conditions where the trajectory-wise reward is normalized between zero and one and the transition is time-homogenous, it achieves horizon-free and second-order bounds. Horizon-free means that our bounds have no polynomial dependence on the horizon of the Markov Decision Process. A second-order bound is a type of instance-dependent bound that scales with respect to the variances of the returns of the policies which can be small when the system is nearly deterministic and (or) the optimal policy has small values. We highlight that our algorithms are simple, fairly standard, and indeed have been extensively studied in the RL literature: they learn a model via MLE, build a version space around the MLE solution, and perform optimistic or pessimistic planning depending on whether operating in the online or offline mode. These algorithms do not rely on additional specialized algorithmic designs such as learning variances and performing variance-weighted learning and thus can leverage rich function approximations that are significantly beyond linear or tabular structures. The simplicity of the algorithms also implies that our horizon-free and second-order regret analysis is actually standard and mainly follows the general framework of optimism/pessimism in the face of uncertainty.

Robotics can help address the growing worker shortage challenge of the manufacturing industry. As such, machine tending is a task collaborative robots can tackle that can also highly boost productivity. Nevertheless, existing robotics systems deployed in that sector rely on a fixed single-arm setup, whereas mobile robots can provide more flexibility and scalability. In this work, we introduce a multi-agent multi-machine tending learning framework by mobile robots based on Multi-agent Reinforcement Learning (MARL) techniques with the design of a suitable observation and reward. Moreover, an attention-based encoding mechanism is developed and integrated into Multi-agent Proximal Policy Optimization (MAPPO) algorithm to boost its performance for machine tending scenarios. Our model (AB-MAPPO) outperformed MAPPO in this new challenging scenario in terms of task success, safety, and resources utilization. Furthermore, we provided an extensive ablation study to support our various design decisions.

LLM-based autonomous agents often fail to execute complex web tasks that require dynamic interaction due to the inherent uncertainty and complexity of these environments. Existing LLM-based web agents typically rely on rigid, expert-designed policies specific to certain states and actions, which lack the flexibility and generalizability needed to adapt to unseen tasks. In contrast, humans excel by exploring unknowns, continuously adapting strategies, and resolving ambiguities through exploration. To emulate human-like adaptability, web agents need strategic exploration and complex decision-making. Monte Carlo Tree Search (MCTS) is well-suited for this, but classical MCTS struggles with vast action spaces, unpredictable state transitions, and incomplete information in web tasks. In light of this, we develop WebPilot, a multi-agent system with a dual optimization strategy that improves MCTS to better handle complex web environments. Specifically, the Global Optimization phase involves generating a high-level plan by breaking down tasks into manageable subtasks and continuously refining this plan, thereby focusing the search process and mitigating the challenges posed by vast action spaces in classical MCTS. Subsequently, the Local Optimization phase executes each subtask using a tailored MCTS designed for complex environments, effectively addressing uncertainties and managing incomplete information. Experimental results on WebArena and MiniWoB++ demonstrate the effectiveness of WebPilot. Notably, on WebArena, WebPilot achieves SOTA performance with GPT-4, achieving a 93% relative increase in success rate over the concurrent tree search-based method. WebPilot marks a significant advancement in general autonomous agent capabilities, paving the way for more advanced and reliable decision-making in practical environments.

Reasoning about real-life events is a unifying challenge in AI and NLP that has profound utility in a variety of domains, while fallacy in high-stake applications could be catastrophic. Able to work with diverse text in these domains, large language models (LLMs) have proven capable of answering questions and solving problems. However, I show that end-to-end LLMs still systematically fail to reason about complex events, and they lack interpretability due to their black-box nature. To address these issues, I propose three general approaches to use LLMs in conjunction with a structured representation of events. The first is a language-based representation involving relations of sub-events that can be learned by LLMs via fine-tuning. The second is a semi-symbolic representation involving states of entities that can be predicted and leveraged by LLMs via few-shot prompting. The third is a fully symbolic representation that can be predicted by LLMs trained with structured data and be executed by symbolic solvers. On a suite of event reasoning tasks spanning common-sense inference and planning, I show that each approach greatly outperforms end-to-end LLMs with more interpretability. These results suggest manners of synergy between LLMs and structured representations for event reasoning and beyond.

Datasets are essential for any machine learning task. Automatic Music Transcription (AMT) is one such task, where considerable amount of data is required depending on the way the solution is achieved. Considering the fact that a music dataset, complete with audio and its time-aligned transcriptions would require the effort of people with musical experience, it could be stated that the task becomes even more challenging. Musical experience is required in playing the musical instrument(s), and in annotating and verifying the transcriptions. We propose a method that would help in streamlining this process, making the task of obtaining a dataset from a particular instrument easy and efficient. We use predefined guitar exercises and hidden Markov model(HMM) based forced viterbi alignment to accomplish this. The guitar exercises are designed to be simple. Since the note sequence are already defined, HMM based forced viterbi alignment provides time-aligned transcriptions of these audio files. The onsets of the transcriptions are manually verified and the labels are accurate up to 10ms, averaging at 5ms. The contributions of the proposed work is two fold, i) a well streamlined and efficient method for generating datasets for any instrument, especially monophonic and, ii) an acoustic plectrum guitar dataset containing wave files and transcriptions in the form of label files. This method will aid as a preliminary step towards building concrete datasets for building AMT systems for different instruments.

Incorporating inductive biases is a promising approach for tackling challenging robot learning domains with sample-efficient solutions. This paper identifies partially observable domains where symmetries can be a useful inductive bias for efficient learning. Specifically, by encoding the equivariance regarding specific group symmetries into the neural networks, our actor-critic reinforcement learning agents can reuse solutions in the past for related scenarios. Consequently, our equivariant agents outperform non-equivariant approaches significantly in terms of sample efficiency and final performance, demonstrated through experiments on a range of robotic tasks in simulation and real hardware.

Centralized Training for Decentralized Execution, where agents are trained offline in a centralized fashion and execute online in a decentralized manner, has become a popular approach in Multi-Agent Reinforcement Learning (MARL). In particular, it has become popular to develop actor-critic methods that train decentralized actors with a centralized critic where the centralized critic is allowed access global information of the entire system, including the true system state. Such centralized critics are possible given offline information and are not used for online execution. While these methods perform well in a number of domains and have become a de facto standard in MARL, using a centralized critic in this context has yet to be sufficiently analyzed theoretically or empirically. In this paper, we therefore formally analyze centralized and decentralized critic approaches, and analyze the effect of using state-based critics in partially observable environments. We derive theories contrary to the common intuition: critic centralization is not strictly beneficial, and using state values can be harmful. We further prove that, in particular, state-based critics can introduce unexpected bias and variance compared to history-based critics. Finally, we demonstrate how the theory applies in practice by comparing different forms of critics on a wide range of common multi-agent benchmarks. The experiments show practical issues such as the difficulty of representation learning with partial observability, which highlights why the theoretical problems are often overlooked in the literature.

Decentralized stochastic control problems are intrinsically difficult to study because of the inapplicability of standard tools from centralized control such as dynamic programming and the resulting computational complexity. In this paper, we address some of these challenges for decentralized stochastic control with Borel spaces under three different but tightly related information structures under a unified theme: the one-step delayed information sharing pattern, the K-step periodic information sharing pattern, and the completely decentralized information structure where no sharing of information occurs. We will show that the one-step delayed and K-step periodic problems can be reduced to a centralized MDP, generalizing prior results which considered finite, linear, or static models, by addressing several measurability questions. The separated nature of policies under both information structures is then established. We then provide sufficient conditions for the transition kernels of both centralized reductions to be weak-Feller, which facilitates rigorous approximation and learning theoretic results. We will then show that for the completely decentralized control problem finite memory local policies are near optimal under a joint conditional mixing condition. This is achieved by obtaining a bound for finite memory policies which goes to zero as memory size increases. We will also provide a performance bound for the K-periodic problem, which results from replacing the full common information by a finite sliding window of information. The latter will depend on the condition of predictor stability in expected total variation, which we will establish. We finally show that under the periodic information sharing pattern, a quantized Q-learning algorithm converges asymptotically towards a near optimal solution. Each of the above, to our knowledge, is a new contribution to the literature.