Extrinsic rewards can effectively guide reinforcement learning (RL) agents in specific tasks. However, extrinsic rewards frequently fall short in complex environments due to the significant human effort needed for their design and annotation. This limitation underscores the necessity for intrinsic rewards, which offer auxiliary and dense signals and can enable agents to learn in an unsupervised manner. Although various intrinsic reward formulations have been proposed, their implementation and optimization details are insufficiently explored and lack standardization, thereby hindering research progress. To address this gap, we introduce RLeXplore, a unified, highly modularized, and plug-and-play framework offering reliable implementations of eight state-of-the-art intrinsic reward algorithms. Furthermore, we conduct an in-depth study that identifies critical implementation details and establishes well-justified standard practices in intrinsically-motivated RL.

We consider off-policy evaluation (OPE) of deterministic target policies for reinforcement learning (RL) in environments with continuous action spaces. While it is common to use importance sampling for OPE, it suffers from high variance when the behavior policy deviates significantly from the target policy. In order to address this issue, some recent works on OPE proposed in-sample learning with importance resampling. Yet, these approaches are not applicable to deterministic target policies for continuous action spaces. To address this limitation, we propose to relax the deterministic target policy using a kernel and learn the kernel metrics that minimize the overall mean squared error of the estimated temporal difference update vector of an action value function, where the action value function is used for policy evaluation. We derive the bias and variance of the estimation error due to this relaxation and provide analytic solutions for the optimal kernel metric. In empirical studies using various test domains, we show that the OPE with in-sample learning using the kernel with optimized metric achieves significantly improved accuracy than other baselines.

Decision making demands intricate interplay between perception, memory, and reasoning to discern optimal policies. Conventional approaches to decision making face challenges related to low sample efficiency and poor generalization. In contrast, foundation models in language and vision have showcased rapid adaptation to diverse new tasks. Therefore, we advocate for the construction of foundation agents as a transformative shift in the learning paradigm of agents. This proposal is underpinned by the formulation of foundation agents with their fundamental characteristics and challenges motivated by the success of large language models (LLMs). Moreover, we specify the roadmap of foundation agents from large interactive data collection or generation, to self-supervised pretraining and adaptation, and knowledge and value alignment with LLMs. Lastly, we pinpoint critical research questions derived from the formulation and delineate trends for foundation agents supported by real-world use cases, addressing both technical and theoretical aspects to propel the field towards a more comprehensive and impactful future.

Federated reinforcement learning (FRL) is a distributed learning framework that combines the principles of reinforcement learning (RL) [1] and federated learning (FL) [2]. Focusing on sequential decision-making, FRL aims to learn an optimal policy through parallel explorations by multiple agents under the coordination of a central server. Often modeled as a Markov decision process (MDP), multiple agents independently interact with an initially unknown environment and collaboratively train their decision-making models with limited information exchange between the agents. This approach accelerates the learning process with low communication costs. Some model-based algorithms (e.g., [3]) and policy-based algorithms (e.g., [4]) have shown speedup with respect to the number of agents in terms of learning regret or convergence rate. Recent progress has been made in FRL algorithms based on model-free value-based approaches, which directly learn the value functions and the optimal policy without estimating the underlying model (e.g., [5]). However, most existing model-free federated algorithms do not actively update the exploration policies for local agents and fail to provide low regret. A comprehensive literature review is provided in Appendix A.

Reinforcement learning (RL) has garnered increasing recognition for its potential to optimise dynamic treatment regimes (DTRs) in personalised medicine, particularly for drug dosage prescriptions and medication recommendations. However, a significant challenge persists: the absence of a unified framework for simulating diverse healthcare scenarios and a comprehensive analysis to benchmark the effectiveness of RL algorithms within these contexts. To address this gap, we introduce \textit{DTR-Bench}, a benchmarking platform comprising four distinct simulation environments tailored to common DTR applications, including cancer chemotherapy, radiotherapy, glucose management in diabetes, and sepsis treatment. We evaluate various state-of-the-art RL algorithms across these settings, particularly highlighting their performance amidst real-world challenges such as pharmacokinetic/pharmacodynamic (PK/PD) variability, noise, and missing data. Our experiments reveal varying degrees of performance degradation among RL algorithms in the presence of noise and patient variability, with some algorithms failing to converge. Additionally, we observe that using temporal observation representations does not consistently lead to improved performance in DTR settings. Our findings underscore the necessity of developing robust, adaptive RL algorithms capable of effectively managing these complexities to enhance patient-specific healthcare. We have open-sourced our benchmark and code at https://github.com/GilesLuo/DTR-Bench.

Typically, traditional Imitation Learning (IL) methods first shape a reward or Q function and then use this shaped function within a reinforcement learning (RL) framework to optimize the empirical policy. However, if the shaped reward/Q function does not adequately represent the ground truth reward/Q function, updating the policy within a multi-step RL framework may result in cumulative bias, further impacting policy learning. Although utilizing behavior cloning (BC) to learn a policy by directly mimicking a few demonstrations in a single-step updating manner can avoid cumulative bias, BC tends to greedily imitate demonstrated actions, limiting its capacity to generalize to unseen state action pairs. To address these challenges, we propose ADR-BC, which aims to enhance behavior cloning through augmented density-based action support, optimizing the policy with this augmented support. Specifically, the objective of ADR-BC shares the similar physical meanings that matching expert distribution while diverging the sub-optimal distribution. Therefore, ADR-BC can achieve more robust expert distribution matching. Meanwhile, as a one-step behavior cloning framework, ADR-BC avoids the cumulative bias associated with multi-step RL frameworks. To validate the performance of ADR-BC, we conduct extensive experiments. Specifically, ADR-BC showcases a 10.5% improvement over the previous state-of-the-art (SOTA) generalized IL baseline, CEIL, across all tasks in the Gym-Mujoco domain. Additionally, it achieves an 89.5% improvement over Implicit Q Learning (IQL) using real rewards across all tasks in the Adroit and Kitchen domains. On the other hand, we conduct extensive ablations to further demonstrate the effectiveness of ADR-BC.

With the advent of AI technologies, humans and robots are increasingly teaming up to perform collaborative tasks. To enable smooth and effective collaboration, the topic of value alignment (operationalized herein as the degree of dynamic goal alignment within a task) between the robot and the human is gaining increasing research attention. Prior literature on value alignment makes an inherent assumption that aligning the values of the robot with that of the human benefits the team. This assumption, however, has not been empirically verified. Moreover, prior literature does not account for human's trust in the robot when analyzing human-robot value alignment. Thus, a research gap needs to be bridged by answering two questions: How does alignment of values affect trust? Is it always beneficial to align the robot's values with that of the human? We present a simulation study and a human-subject study to answer these questions. Results from the simulation study show that alignment of values is important for trust when the overall risk level of the task is high. We also present an adaptive strategy for the robot that uses Inverse Reinforcement Learning (IRL) to match the values of the robot with those of the human during interaction. Our simulations suggest that such an adaptive strategy is able to maintain trust across the full spectrum of human values. We also present results from an empirical study that validate these findings from simulation. Results indicate that real-time personalized value alignment is beneficial to trust and perceived performance by the human when the robot does not have a good prior on the human's values.

Implicit Q-learning (IQL) serves as a strong baseline for offline RL, which learns the value function using only dataset actions through quantile regression. However, it is unclear how to recover the implicit policy from the learned implicit Q-function and why IQL can utilize weighted regression for policy extraction. IDQL reinterprets IQL as an actor-critic method and gets weights of implicit policy, however, this weight only holds for the optimal value function. In this work, we introduce a different way to solve the implicit policy-finding problem (IPF) by formulating this problem as an optimization problem. Based on this optimization problem, we further propose two practical algorithms AlignIQL and AlignIQL-hard, which inherit the advantages of decoupling actor from critic in IQL and provide insights into why IQL can use weighted regression for policy extraction. Compared with IQL and IDQL, we find our method keeps the simplicity of IQL and solves the implicit policy-finding problem. Experimental results on D4RL datasets show that our method achieves competitive or superior results compared with other SOTA offline RL methods. Especially in complex sparse reward tasks like Antmaze and Adroit, our method outperforms IQL and IDQL by a significant margin.

The field of risk-constrained reinforcement learning (RCRL) has been developed to effectively reduce the likelihood of worst-case scenarios by explicitly handling risk-measure-based constraints. However, the nonlinearity of risk measures makes it challenging to achieve convergence and optimality. To overcome the difficulties posed by the nonlinearity, we propose a spectral risk measure-constrained RL algorithm, spectral-risk-constrained policy optimization (SRCPO), a bilevel optimization approach that utilizes the duality of spectral risk measures. In the bilevel optimization structure, the outer problem involves optimizing dual variables derived from the risk measures, while the inner problem involves finding an optimal policy given these dual variables. The proposed method, to the best of our knowledge, is the first to guarantee convergence to an optimum in the tabular setting. Furthermore, the proposed method has been evaluated on continuous control tasks and showed the best performance among other RCRL algorithms satisfying the constraints.

Preference-based reinforcement learning (PbRL) has shown impressive capabilities in training agents without reward engineering. However, a notable limitation of PbRL is its dependency on substantial human feedback. This dependency stems from the learning loop, which entails accurate reward learning compounded with value/policy learning, necessitating a considerable number of samples. To boost the learning loop, we propose SEER, an efficient PbRL method that integrates label smoothing and policy regularization techniques. Label smoothing reduces overfitting of the reward model by smoothing human preference labels. Additionally, we bootstrap a conservative estimate $\widehat{Q}$ using well-supported state-action pairs from the current replay memory to mitigate overestimation bias and utilize it for policy learning regularization. Our experimental results across a variety of complex tasks, both in online and offline settings, demonstrate that our approach improves feedback efficiency, outperforming state-of-the-art methods by a large margin. Ablation studies further reveal that SEER achieves a more accurate Q-function compared to prior work.