Despite the recent advancements in offline RL, no unified algorithm could achieve superior performance across a broad range of tasks. Offline \textit{value function learning}, in particular, struggles with sparse-reward, long-horizon tasks due to the difficulty of solving credit assignment and extrapolation errors that accumulates as the horizon of the task grows.~On the other hand, models that can perform well in long-horizon tasks are designed specifically for goal-conditioned tasks, which commonly perform worse than value function learning methods on short-horizon, dense-reward scenarios. To bridge this gap, we propose a hierarchical planner designed for offline RL called PlanDQ. PlanDQ incorporates a diffusion-based planner at the high level, named D-Conductor, which guides the low-level policy through sub-goals. At the low level, we used a Q-learning based approach called the Q-Performer to accomplish these sub-goals. Our experimental results suggest that PlanDQ can achieve superior or competitive performance on D4RL continuous control benchmark tasks as well as AntMaze, Kitchen, and Calvin as long-horizon tasks.

In the course of the energy transition, the expansion of generation and consumption will change, and many of these technologies, such as PV systems, electric cars and heat pumps, will influence the power flow, especially in the distribution grids. Scalable methods that can make decisions for each grid connection are needed to enable congestion-free grid operation in the distribution grids. This paper presents a novel end-to-end approach to resolving congestion in distribution grids with deep reinforcement learning. Our architecture learns to curtail power and set appropriate reactive power to determine a non-congested and, thus, feasible grid state. State-of-the-art methods such as the optimal power flow (OPF) demand high computational costs and detailed measurements of every bus in a grid. In contrast, the presented method enables decisions under sparse information with just some buses observable in the grid. Distribution grids are generally not yet fully digitized and observable, so this method can be used for decision-making on the majority of low-voltage grids. On a real low-voltage grid the approach resolves 100\% of violations in the voltage band and 98.8\% of asset overloads. The results show that decisions can also be made on real grids that guarantee sufficient quality for congestion-free grid operation.

In deep Reinforcement Learning (RL), value functions are typically approximated using deep neural networks and trained via mean squared error regression objectives to fit the true value functions. Recent research has proposed an alternative approach, utilizing the cross-entropy classification objective, which has demonstrated improved performance and scalability of RL algorithms. However, existing study have not extensively benchmarked the effects of this replacement across various domains, as the primary objective was to demonstrate the efficacy of the concept across a broad spectrum of tasks, without delving into in-depth analysis. Our work seeks to empirically investigate the impact of such a replacement in an offline RL setup and analyze the effects of different aspects on performance. Through large-scale experiments conducted across a diverse range of tasks using different algorithms, we aim to gain deeper insights into the implications of this approach. Our results reveal that incorporating this change can lead to superior performance over state-of-the-art solutions for some algorithms in certain tasks, while maintaining comparable performance levels in other tasks, however for other algorithms this modification might lead to the dramatic performance drop. This findings are crucial for further application of classification approach in research and practical tasks.

Reinforcement learning (RL) has proven to be a powerful tool for training agents that excel in various games. However, the black-box nature of neural network models often hinders our ability to understand the reasoning behind the agent's actions. Recent research has attempted to address this issue by using the guidance of pretrained neural agents to encode logic-based policies, allowing for interpretable decisions. A drawback of such approaches is the requirement of large amounts of predefined background knowledge in the form of predicates, limiting its applicability and scalability. In this work, we propose a novel approach, Explanatory Predicate Invention for Learning in Games (EXPIL), that identifies and extracts predicates from a pretrained neural agent, later used in the logic-based agents, reducing the dependency on predefined background knowledge. Our experimental evaluation on various games demonstrate the effectiveness of EXPIL in achieving explainable behavior in logic agents while requiring less background knowledge.

Multi-task reinforcement learning (MTRL) has shown great promise in many real-world applications. Existing MTRL algorithms often aim to learn a policy that optimizes individual objective functions simultaneously with a given prior preference (or weights) on different tasks. However, these methods often suffer from the issue of \textit{gradient conflict} such that the tasks with larger gradients dominate the update direction, resulting in a performance degeneration on other tasks. In this paper, we develop a novel dynamic weighting multi-task actor-critic algorithm (MTAC) under two options of sub-procedures named as CA and FC in task weight updates. MTAC-CA aims to find a conflict-avoidant (CA) update direction that maximizes the minimum value improvement among tasks, and MTAC-FC targets at a much faster convergence rate. We provide a comprehensive finite-time convergence analysis for both algorithms. We show that MTAC-CA can find a $\epsilon+\epsilon_{\text{app}}$-accurate Pareto stationary policy using $\mathcal{O}({\epsilon^{-5}})$ samples, while ensuring a small $\epsilon+\sqrt{\epsilon_{\text{app}}}$-level CA distance (defined as the distance to the CA direction), where $\epsilon_{\text{app}}$ is the function approximation error. The analysis also shows that MTAC-FC improves the sample complexity to $\mathcal{O}(\epsilon^{-3})$, but with a constant-level CA distance. Our experiments on MT10 demonstrate the improved performance of our algorithms over existing MTRL methods with fixed preference.

This paper proposes a novel deep symbolic regression approach to enhance the robustness and interpretability of data-driven mathematical expression discovery. Despite the success of the state-of-the-art method, DSR, it is built on recurrent neural networks, purely guided by data fitness, and potentially meet tail barriers, which can zero out the policy gradient and cause inefficient model updates. To overcome these limitations, we use transformers in conjunction with breadth-first-search to improve the learning performance. We use Bayesian information criterion (BIC) as the reward function to explicitly account for the expression complexity and optimize the trade-off between interpretability and data fitness. We propose a modified risk-seeking policy that not only ensures the unbiasness of the gradient, but also removes the tail barriers, thus ensuring effective updates from top performers. Through a series of benchmarks and systematic experiments, we demonstrate the advantages of our approach.

Humanoid activities involving sequential contacts are crucial for complex robotic interactions and operations in the real world and are traditionally solved by model-based motion planning, which is time-consuming and often relies on simplified dynamics models. Although model-free reinforcement learning (RL) has become a powerful tool for versatile and robust whole-body humanoid control, it still requires tedious task-specific tuning and state machine design and suffers from long-horizon exploration issues in tasks involving contact sequences. In this work, we propose WoCoCo (Whole-Body Control with Sequential Contacts), a unified framework to learn whole-body humanoid control with sequential contacts by naturally decomposing the tasks into separate contact stages. Such decomposition facilitates simple and general policy learning pipelines through task-agnostic reward and sim-to-real designs, requiring only one or two task-related terms to be specified for each task. We demonstrated that end-to-end RL-based controllers trained with WoCoCo enable four challenging whole-body humanoid tasks involving diverse contact sequences in the real world without any motion priors: 1) versatile parkour jumping, 2) box loco-manipulation, 3) dynamic clap-and-tap dancing, and 4) cliffside climbing. We further show that WoCoCo is a general framework beyond humanoid by applying it in 22-DoF dinosaur robot loco-manipulation tasks.

Two-time-scale optimization is a framework introduced in Zeng et al. (2024) that abstracts a range of policy evaluation and policy optimization problems in reinforcement learning (RL). Akin to bi-level optimization under a particular type of stochastic oracle, the two-time-scale optimization framework has an upper level objective whose gradient evaluation depends on the solution of a lower level problem, which is to find the root of a strongly monotone operator. In this work, we propose a new method for solving two-time-scale optimization that achieves significantly faster convergence than the prior arts. The key idea of our approach is to leverage an averaging step to improve the estimates of the operators in both lower and upper levels before using them to update the decision variables. These additional averaging steps eliminate the direct coupling between the main variables, enabling the accelerated performance of our algorithm. We characterize the finite-time convergence rates of the proposed algorithm under various conditions of the underlying objective function, including strong convexity, convexity, Polyak-Lojasiewicz condition, and general non-convexity. These rates significantly improve over the best-known complexity of the standard two-time-scale stochastic approximation algorithm. When applied to RL, we show how the proposed algorithm specializes to novel online sample-based methods that surpass or match the performance of the existing state of the art. Finally, we support our theoretical results with numerical simulations in RL.

In this paper, we study the non-asymptotic sample complexity for the pure exploration problem in contextual bandits and tabular reinforcement learning (RL): identifying an epsilon-optimal policy from a set of policies with high probability. Existing work in bandits has shown that it is possible to identify the best policy by estimating only the difference between the behaviors of individual policies, which can be substantially cheaper than estimating the behavior of each policy directly. However, the best-known complexities in RL fail to take advantage of this and instead estimate the behavior of each policy directly. Does it suffice to estimate only the differences in the behaviors of policies in RL? We answer this question positively for contextual bandits but in the negative for tabular RL, showing a separation between contextual bandits and RL. However, inspired by this, we show that it almost suffices to estimate only the differences in RL: if we can estimate the behavior of a single reference policy, it suffices to only estimate how any other policy deviates from this reference policy. We develop an algorithm which instantiates this principle and obtains, to the best of our knowledge, the tightest known bound on the sample complexity of tabular RL.

Automation holds the potential to assist surgeons in robotic interventions, shifting their mental work load from visuomotor control to high level decision making. Reinforcement learning has shown promising results in learning complex visuomotor policies, especially in simulation environments where many samples can be collected at low cost. A core challenge is learning policies in simulation that can be deployed in the real world, thereby overcoming the sim-to-real gap. In this work, we bridge the visual sim-to-real gap with an image-based reinforcement learning pipeline based on pixel-level domain adaptation and demonstrate its effectiveness on an image-based task in deformable object manipulation. We choose a tissue retraction task because of its importance in clinical reality of precise cancer surgery. After training in simulation on domain-translated images, our policy requires no retraining to perform tissue retraction with a 50% success rate on the real robotic system using raw RGB images. Furthermore, our sim-to-real transfer method makes no assumptions on the task itself and requires no paired images. This work introduces the first successful application of visual sim-to-real transfer for robotic manipulation of deformable objects in the surgical field, which represents a notable step towards the clinical translation of cognitive surgical robotics.