In visual-based Reinforcement Learning (RL), agents often struggle to generalize well to environmental variations in the state space that were not observed during training. The variations can arise in both task-irrelevant features, such as background noise, and task-relevant features, such as robot configurations, that are related to the optimal decisions. To achieve generalization in both situations, agents are required to accurately understand the impact of changed features on the decisions, i.e., establishing the true associations between changed features and decisions in the policy model. However, due to the inherent correlations among features in the state space, the associations between features and decisions become entangled, making it difficult for the policy to distinguish them. To this end, we propose Saliency-Guided Features Decorrelation (SGFD) to eliminate these correlations through sample reweighting. Concretely, SGFD consists of two core techniques: Random Fourier Functions (RFF) and the saliency map. RFF is utilized to estimate the complex non-linear correlations in high-dimensional images, while the saliency map is designed to identify the changed features. Under the guidance of the saliency map, SGFD employs sample reweighting to minimize the estimated correlations related to changed features, thereby achieving decorrelation in visual RL tasks. Our experimental results demonstrate that SGFD can generalize well on a wide range of test environments and significantly outperforms state-of-the-art methods in handling both task-irrelevant variations and task-relevant variations.

In this paper, we consider federated reinforcement learning for tabular episodic Markov Decision Processes (MDP) where, under the coordination of a central server, multiple agents collaboratively explore the environment and learn an optimal policy without sharing their raw data. While linear speedup in the number of agents has been achieved for some metrics, such as convergence rate and sample complexity, in similar settings, it is unclear whether it is possible to design a model-free algorithm to achieve linear regret speedup with low communication cost. We propose two federated Q-Learning algorithms termed as FedQ-Hoeffding and FedQ-Bernstein, respectively, and show that the corresponding total regrets achieve a linear speedup compared with their single-agent counterparts when the time horizon is sufficiently large, while the communication cost scales logarithmically in the total number of time steps $T$. Those results rely on an event-triggered synchronization mechanism between the agents and the server, a novel step size selection when the server aggregates the local estimates of the state-action values to form the global estimates, and a set of new concentration inequalities to bound the sum of non-martingale differences. This is the first work showing that linear regret speedup and logarithmic communication cost can be achieved by model-free algorithms in federated reinforcement learning.

Stochastic optimal control of dynamical systems is a crucial challenge in sequential decision-making. Recently, control-as-inference approaches have had considerable success, providing a viable risk-sensitive framework to address the exploration-exploitation dilemma. Nonetheless, a majority of these techniques only invoke the inference-control duality to derive a modified risk objective that is then addressed within a reinforcement learning framework. This paper introduces a novel perspective by framing risk-sensitive stochastic control as Markovian score climbing under samples drawn from a conditional particle filter. Our approach, while purely inference-centric, provides asymptotically unbiased estimates for gradient-based policy optimization with optimal importance weighting and no explicit value function learning. To validate our methodology, we apply it to the task of learning neural non-Gaussian feedback policies, showcasing its efficacy on numerical benchmarks of stochastic dynamical systems.

De novo drug design is a pivotal issue in pharmacology and a new area of focus in AI for science research. A central challenge in this field is to generate molecules with specific properties while also producing a wide range of diverse candidates. Although advanced technologies such as transformer models and reinforcement learning have been applied in drug design, their potential has not been fully realized. Therefore, we propose MolRL-MGPT, a reinforcement learning algorithm with multiple GPT agents for drug molecular generation. To promote molecular diversity, we encourage the agents to collaborate in searching for desirable molecules in diverse directions. Our algorithm has shown promising results on the GuacaMol benchmark and exhibits efficacy in designing inhibitors against SARS-CoV-2 protein targets. The codes are available at: https://github.com/HXYfighter/

In this work, we propose REBEL, an algorithm for sample efficient reward regularization based robotic reinforcement learning from human feedback (RRLHF). Reinforcement learning (RL) performance for continuous control robotics tasks is sensitive to the underlying reward function. In practice, the reward function often ends up misaligned with human intent, values, social norms, etc., leading to catastrophic failures in the real world. We leverage human preferences to learn regularized reward functions and eventually align the agents with the true intended behavior. We introduce a novel notion of reward regularization to the existing RRLHF framework, which is termed as agent preferences. So, we not only consider human feedback in terms of preferences, we also propose to take into account the preference of the underlying RL agent while learning the reward function. We show that this helps to improve the over-optimization associated with the design of reward functions in RL. We experimentally show that REBEL exhibits up to 70% improvement in sample efficiency to achieve a similar level of episodic reward returns as compared to the state-of-the-art methods such as PEBBLE and PEBBLE+SURF.

The RoboCup 3D Soccer Simulation League serves as a competitive platform for showcasing innovation in autonomous humanoid robot agents through simulated soccer matches. Our team, FC Portugal, developed a new codebase from scratch in Python after RoboCup 2021. The team's performance is based on a set of skills centered around novel unifying primitives and a custom, symmetry-extended version of the Proximal Policy Optimization algorithm. Our methods have been thoroughly tested in official RoboCup matches, where FC Portugal has won the last two main competitions, in 2022 and 2023. This paper presents our training framework, as well as a timeline of skills developed using our skill-set-primitives, which considerably improve the sample efficiency and stability of skills, and motivate seamless transitions. We start with a significantly fast sprint-kick developed in 2021 and progress to the most recent skill set, which includes a multi-purpose omnidirectional walk, a dribble with unprecedented ball control, a solid kick, and a push skill. The push tackles both low-level collision-prone scenarios and high-level strategies to increase ball possession. We address the resource-intensive nature of this task through an innovative multi-agent learning approach. Finally, we release the codebase of our team to the RoboCup community, enabling other teams to transition to Python more easily and providing new teams with a robust and modern foundation upon which they can build new features.

Generative Flow Networks (GFNs) have emerged as a powerful tool for sampling discrete objects from unnormalized distributions, offering a scalable alternative to Markov Chain Monte Carlo (MCMC) methods. While GFNs draw inspiration from maximum entropy reinforcement learning (RL), the connection between the two has largely been unclear and seemingly applicable only in specific cases. This paper addresses the connection by constructing an appropriate reward function, thereby establishing an exact relationship between GFNs and maximum entropy RL. This construction allows us to introduce maximum entropy GFNs, which, in contrast to GFNs with uniform backward policy, achieve the maximum entropy attainable by GFNs without constraints on the state space.

Preference-based Reinforcement Learning (PbRL) is an active area of research, and has made significant strides in single-agent actor and in observer human-in-the-loop scenarios. However, its application within the co-operative multi-agent RL frameworks, where humans actively participate and express preferences for agent behavior, remains largely uncharted. We consider a two-agent (Human-AI) cooperative setup where both the agents are rewarded according to human's reward function for the team. However, the agent does not have access to it, and instead, utilizes preference-based queries to elicit its objectives and human's preferences for the robot in the human-robot team. We introduce the notion of Human-Flexibility, i.e. whether the human partner is amenable to multiple team strategies, with a special case being Specified Orchestration where the human has a single team policy in mind (most constrained case). We propose a suite of domains to study PbRL for Human-AI cooperative setup which explicitly require forced cooperation. Adapting state-of-the-art single-agent PbRL algorithms to our two-agent setting, we conduct a comprehensive benchmarking study across our domain suite. Our findings highlight the challenges associated with high degree of Human-Flexibility and the limited access to the human's envisioned policy in PbRL for Human-AI cooperation. Notably, we observe that PbRL algorithms exhibit effective performance exclusively in the case of Specified Orchestration which can be seen as an upper bound PbRL performance for future research.

Learning rewards from expert videos offers an affordable and effective solution to specify the intended behaviors for reinforcement learning tasks. In this work, we propose Diffusion Reward, a novel framework that learns rewards from expert videos via conditional video diffusion models for solving complex visual RL problems. Our key insight is that lower generative diversity is observed when conditioned on expert trajectories. Diffusion Reward is accordingly formalized by the negative of conditional entropy that encourages productive exploration of expert-like behaviors. We show the efficacy of our method over 10 robotic manipulation tasks from MetaWorld and Adroit with visual input and sparse reward. Moreover, Diffusion Reward could even solve unseen tasks successfully and effectively, largely surpassing baseline methods. Project page and code: https://diffusion-reward.github.io/.

We develop a method of adapting the AlphaZero model to General Game Playing (GGP) that focuses on faster model generation and requires less knowledge to be extracted from the game rules. The dataset generation uses MCTS playing instead of self-play; only the value network is used, and attention layers replace the convolutional ones. This allows us to abandon any assumptions about the action space and board topology. We implement the method within the Regular Boardgames GGP system and show that we can build models outperforming the UCT baseline for most games efficiently.