Deep Reinforcement Learning (DRL) has achieved remarkable success in solving complex decision-making problems by combining the representation capabilities of deep learning with the decision-making power of reinforcement learning. However, learning in sparse reward environments remains challenging due to insufficient feedback to guide the optimization of agents, especially in real-life environments with high-dimensional states. To tackle this issue, experience replay is commonly introduced to enhance learning efficiency through past experiences. Nonetheless, current methods of experience replay, whether based on uniform or prioritized sampling, frequently struggle with suboptimal learning efficiency and insufficient utilization of samples. This paper proposes a novel approach, diversity-based experience replay (DBER), which leverages the deterministic point process to prioritize diverse samples in state realizations. We conducted extensive experiments on Robotic Manipulation tasks in MuJoCo, Atari games, and realistic in-door environments in Habitat. The results show that our method not only significantly improves learning efficiency but also demonstrates superior performance in sparse reward environments with high-dimensional states, providing a simple yet effective solution for this field.

There have been key advancements to building universal approximators for multi-goal collections of reinforcement learning value functions -- key elements in estimating long-term returns of states in a parameterized manner. We extend this to hierarchical reinforcement learning, using the options framework, by introducing hierarchical universal value function approximators (H-UVFAs). This allows us to leverage the added benefits of scaling, planning, and generalization expected in temporal abstraction settings. We develop supervised and reinforcement learning methods for learning embeddings of the states, goals, options, and actions in the two hierarchical value functions: $Q(s, g, o; \theta)$ and $Q(s, g, o, a; \theta)$. Finally we demonstrate generalization of the HUVFAs and show they outperform corresponding UVFAs.

Aligning Large Language Models (LLMs) to human preferences in content, style, and presentation is challenging, in part because preferences are varied, contextdependent, and sometimes inherently ambiguous. While successful, Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) are prone to the issue of proxy reward overoptimization. Analysis of the DPO loss reveals a critical need for regularization for mislabeled or ambiguous preference pairs to avoid reward hacking. In this work, we develop a pessimistic framework for DPO by introducing preference uncertainty penalization schemes, inspired by offline reinforcement learning. The penalization serves as a correction to the loss which attenuates the loss gradient for uncertain samples. Evaluation of the methods is performed with GPT2 Medium on the Anthropic-HH dataset using a model ensemble to obtain uncertainty estimates, and shows improved overall performance compared to vanilla DPO, as well as better completions on prompts from high-uncertainty chosen/rejected responses. Aligning LLMs to human preferences in content, style, and presentation has become a central challenge in improving and deploying LLMs, leading to the advent of Reinforcement Learning with Human Feedback (RLHF), now a prominent technique to fine-tune state-of-the-art LLMs (Casper et al., 2023). The standard RLHF pipeline involves human feedback collection, reward model training, and LLM policy optimization via reinforcement learning (RL). Despite its success, each stage presents challenges, from feedback interpretation and policy generalization to challenging RL implementation (Casper et al., 2023). Direct Preference Optimisation (DPO) (Rafailov et al., 2023) effectively bypasses the reward model by fine-tuning the policy to maximize the likelihood of the preference data under the Bradley-Terry model (A. DPO is easier to implement than RL algorithms, and benefits from computational efficiency and stability by avoiding potential inaccuracies and biases of a reward model (Xu et al., 2024; Casper et al., 2023).

In order to mitigate the sample complexity of real-world reinforcement learning, common practice is to first train a policy in a simulator where samples are cheap, and then deploy this policy in the real world, with the hope that it generalizes effectively. Such \emph{direct sim2real} transfer is not guaranteed to succeed, however, and in cases where it fails, it is unclear how to best utilize the simulator. In this work, we show that in many regimes, while direct sim2real transfer may fail, we can utilize the simulator to learn a set of \emph{exploratory} policies which enable efficient exploration in the real world. In particular, in the setting of low-rank MDPs, we show that coupling these exploratory policies with simple, practical approaches -- least-squares regression oracles and naive randomized exploration -- yields a polynomial sample complexity in the real world, an exponential improvement over direct sim2real transfer, or learning without access to a simulator. To the best of our knowledge, this is the first evidence that simulation transfer yields a provable gain in reinforcement learning in settings where direct sim2real transfer fails. We validate our theoretical results on several realistic robotic simulators and a real-world robotic sim2real task, demonstrating that transferring exploratory policies can yield substantial gains in practice as well.

The ``AI Olympics with RealAIGym'' competition challenges participants to stabilize chaotic underactuated dynamical systems with advanced control algorithms. In this paper, we present a novel solution submitted to IROS'24 competition, which builds upon Soft Actor-Critic (SAC), a popular model-free entropy-regularized Reinforcement Learning (RL) algorithm. We add a `context' vector to the state, which encodes the immediate history via a Convolutional Neural Network (CNN) to counteract the unmodeled effects on the real system. Our method achieves high performance scores and competitive robustness scores on both tracks of the competition: Pendubot and Acrobot.

Reinforcement Learning (RL) empowers agents to acquire various skills by learning from reward signals. Unfortunately, designing high-quality instance-level rewards often demands significant effort. An emerging alternative, RL with delayed reward, focuses on learning from rewards presented periodically, which can be obtained from human evaluators assessing the agent's performance over sequences of behaviors. However, traditional methods in this domain assume the existence of underlying Markovian rewards and that the observed delayed reward is simply the sum of instance-level rewards, both of which often do not align well with real-world scenarios. In this paper, we introduce the problem of RL from Composite Delayed Reward (RLCoDe), which generalizes traditional RL from delayed rewards by eliminating the strong assumption. We suggest that the delayed reward may arise from a more complex structure reflecting the overall contribution of the sequence. To address this problem, we present a framework for modeling composite delayed rewards, using a weighted sum of non-Markovian components to capture the different contributions of individual steps. Building on this framework, we propose Composite Delayed Reward Transformer (CoDeTr), which incorporates a specialized in-sequence attention mechanism to effectively model these contributions. We conduct experiments on challenging locomotion tasks where the agent receives delayed rewards computed from composite functions of observable step rewards. The experimental results indicate that CoDeTr consistently outperforms baseline methods across evaluated metrics. Additionally, we demonstrate that it effectively identifies the most significant time steps within the sequence and accurately predicts rewards that closely reflect the environment feedback.

Offline-to-Online Reinforcement Learning has emerged as a powerful paradigm, leveraging offline data for initialization and online fine-tuning to enhance both sample efficiency and performance. However, most existing research has focused on single-agent settings, with limited exploration of the multi-agent extension, i.e., Offline-to-Online Multi-Agent Reinforcement Learning (O2O MARL). In O2O MARL, two critical challenges become more prominent as the number of agents increases: (i) the risk of unlearning pre-trained Q-values due to distributional shifts during the transition from offline-to-online phases, and (ii) the difficulty of efficient exploration in the large joint state-action space. To tackle these challenges, we propose a novel O2O MARL framework called Offline Value Function Memory with Sequential Exploration (OVMSE). First, we introduce the Offline Value Function Memory (OVM) mechanism to compute target Q-values, preserving knowledge gained during offline training, ensuring smoother transitions, and enabling efficient fine-tuning. Second, we propose a decentralized Sequential Exploration (SE) strategy tailored for O2O MARL, which effectively utilizes the pre-trained offline policy for exploration, thereby significantly reducing the joint state-action space to be explored. Extensive experiments on the StarCraft Multi-Agent Challenge (SMAC) demonstrate that OVMSE significantly outperforms existing baselines, achieving superior sample efficiency and overall performance.

Data collection in imitation learning often requires significant, laborious human supervision, such as numerous demonstrations, and/or frequent environment resets for methods that incorporate reinforcement learning. In this work, we propose an alternative approach, MILES: a fully autonomous, self-supervised data collection paradigm, and we show that this enables efficient policy learning from just a single demonstration and a single environment reset. MILES autonomously learns a policy for returning to and then following the single demonstration, whilst being self-guided during data collection, eliminating the need for additional human interventions. We evaluated MILES across several real-world tasks, including tasks that require precise contact-rich manipulation such as locking a lock with a key. We found that, under the constraints of a single demonstration and no repeated environment resetting, MILES significantly outperforms state-of-the-art alternatives like imitation learning methods that leverage reinforcement learning. Videos of our experiments and code can be found on our webpage: www.robot-learning.uk/miles.

Updates of extensive Internet of Things (IoT) data are critical to the immersion of vehicular metaverse services. However, providing high-quality and sustainable data in unstable and resource-constrained vehicular networks remains a significant challenge. To address this problem, we put forth a novel immersion-aware model trading framework that incentivizes metaverse users (MUs) to contribute learning models trained by their latest local data for augmented reality (AR) services in the vehicular metaverse, while preserving their privacy through federated learning. To comprehensively evaluate the contribution of locally trained learning models provided by MUs to AR services, we design a new immersion metric that captures service immersion by considering the freshness and accuracy of learning models, as well as the amount and potential value of raw data used for training. We model the trading interactions between metaverse service providers (MSPs) and MUs as an equilibrium problem with equilibrium constraints (EPEC) to analyze and balance their costs and gains. Moreover, considering dynamic network conditions and privacy concerns, we formulate the reward decisions of MSPs as a multi-agent Markov decision process. Then, a fully distributed dynamic reward method based on deep reinforcement learning is presented, which operates without any private information about MUs and other MSPs. Experimental results demonstrate that the proposed framework can effectively provide higher-value models for object detection and classification in AR services on real AR-related vehicle datasets compared to benchmark schemes.

Reinforcement Learning (RL) has the potential to enable robots to learn from their own actions in the real world. Unfortunately, RL can be prohibitively expensive, in terms of on-robot runtime, due to inefficient exploration when learning from a sparse reward signal. Designing dense reward functions is labour-intensive and requires domain expertise. In our work, we propose GCR (Goal-Contrastive Rewards), a dense reward function learning method that can be trained on passive video demonstrations. By using videos without actions, our method is easier to scale, as we can use arbitrary videos. GCR combines two loss functions, an implicit value loss function that models how the reward increases when traversing a successful trajectory, and a goal-contrastive loss that discriminates between successful and failed trajectories. We perform experiments in simulated manipulation environments across RoboMimic and MimicGen tasks, as well as in the real world using a Franka arm and a Spot quadruped. We find that GCR leads to a more-sample efficient RL, enabling model-free RL to solve about twice as many tasks as our baseline reward learning methods. We also demonstrate positive cross-embodiment transfer from videos of people and of other robots performing a task. Appendix: \url{https://tinyurl.com/gcr-appendix-2}.