Reinforcement learning in sparse-reward navigation environments with expensive and limited interactions is challenging and poses a need for effective exploration. Motivated by complex navigation tasks that require real-world training (when cheap simulators are not available), we consider an agent that faces an unknown distribution of environments and must decide on an exploration strategy. It may leverage a series of training environments to improve its policy before it is evaluated in a test environment drawn from the same environment distribution. Most existing approaches focus on fixed exploration strategies, while the few that view exploration as a meta-optimization problem tend to ignore the need for cost-efficient exploration. We propose a cost-aware Bayesian optimization approach that efficiently searches over a class of dynamic subgoal-based exploration strategies. The algorithm adjusts a variety of levers -- the locations of the subgoals, the length of each episode, and the number of replications per trial -- in order to overcome the challenges of sparse rewards, expensive interactions, and noise. An experimental evaluation demonstrates that the new approach outperforms existing baselines across a number of problem domains. We also provide a theoretical foundation and prove that the method asymptotically identifies a near-optimal subgoal design.

Learning high-quality Q-value functions plays a key role in the success of many modern off-policy deep reinforcement learning (RL) algorithms. Previous works focus on addressing the value overestimation issue, an outcome of adopting function approximators and off-policy learning. Deviating from the common viewpoint, we observe that Q-values are indeed underestimated in the latter stage of the RL training process, primarily related to the use of inferior actions from the current policy in Bellman updates as compared to the more optimal action samples in the replay buffer. We hypothesize that this long-neglected phenomenon potentially hinders policy learning and reduces sample efficiency. Our insight to address this issue is to incorporate sufficient exploitation of past successes while maintaining exploration optimism. We propose the Blended Exploitation and Exploration (BEE) operator, a simple yet effective approach that updates Q-value using both historical best-performing actions and the current policy. The instantiations of our method in both model-free and model-based settings outperform state-of-the-art methods in various continuous control tasks and achieve strong performance in failure-prone scenarios and real-world robot tasks.

Modern commercial Heating, Ventilation, and Air Conditioning (HVAC) devices form a complex and interconnected thermodynamic system with the building and outside weather conditions, and current setpoint control policies are not fully optimized for minimizing energy use and carbon emission. Given a suitable training environment, a Reinforcement Learning (RL) model is able to improve upon these policies, but training such a model, especially in a way that scales to thousands of buildings, presents many real world challenges. We propose a novel simulation-based approach, where a customized simulator is used to train the agent for each building. Our open-source simulator (available online: https://github.com/google/sbsim) is lightweight and calibrated via telemetry from the building to reach a higher level of fidelity. On a two-story, 68,000 square foot building, with 127 devices, we were able to calibrate our simulator to have just over half a degree of drift from the real world over a six-hour interval. This approach is an important step toward having a real-world RL control system that can be scaled to many buildings, allowing for greater efficiency and resulting in reduced energy consumption and carbon emissions.

Incomplete knowledge of the environment leads an agent to make decisions under uncertainty. One of the major dilemmas in Reinforcement Learning (RL) where an autonomous agent has to balance two contrasting needs in making its decisions is: exploiting the current knowledge of the environment to maximize the cumulative reward as well as exploring actions that allow improving the knowledge of the environment, hopefully leading to higher reward values (exploration-exploitation trade-off). Concurrently, another relevant issue regards the full observability of the states, which may not be assumed in all applications. Such as when only 2D images are considered as input in a RL approach used for finding the optimal action within a 3D simulation environment. In this work, we address these issues by deploying and testing several techniques to balance exploration and exploitation trade-off on partially observable systems for predicting steering wheels in autonomous driving scenario. More precisely, the final aim is to investigate the effects of using both stochastic and deterministic multi-armed bandit strategies coupled with a Deep Recurrent Q-Network. Additionally, we adapted and evaluated the impact of an innovative method to improve the learning phase of the underlying Convolutional Recurrent Neural Network. We aim to show that adaptive stochastic methods for exploration better approximate the trade-off between exploration and exploitation as, in general, Softmax and Max-Boltzmann strategies are able to outperform epsilon-greedy techniques.

It is desirable for policies to optimistically explore new states and behaviors during online reinforcement learning (RL) or fine-tuning, especially when prior offline data does not provide enough state coverage. However, exploration bonuses can bias the learned policy, and our experiments find that naive, yet standard use of such bonuses can fail to recover a performant policy. Concurrently, pessimistic training in offline RL has enabled recovery of performant policies from static datasets. Can we leverage offline RL to recover better policies from online interaction? We make a simple observation that a policy can be trained from scratch on all interaction data with pessimistic objectives, thereby decoupling the policies used for data collection and for evaluation. Specifically, we propose offline retraining, a policy extraction step at the end of online fine-tuning in our Offline-to-Online-to-Offline (OOO) framework for reinforcement learning (RL). An optimistic (exploration) policy is used to interact with the environment, and a separate pessimistic (exploitation) policy is trained on all the observed data for evaluation. Such decoupling can reduce any bias from online interaction (intrinsic rewards, primacy bias) in the evaluation policy, and can allow more exploratory behaviors during online interaction which in turn can generate better data for exploitation. OOO is complementary to several offline-to-online RL and online RL methods, and improves their average performance by 14% to 26% in our fine-tuning experiments, achieves state-of-the-art performance on several environments in the D4RL benchmarks, and improves online RL performance by 165% on two OpenAI gym environments. Further, OOO can enable fine-tuning from incomplete offline datasets where prior methods can fail to recover a performant policy. Implementation: https://github.com/MaxSobolMark/OOO

Temporal difference learning (TD) is a foundational concept in reinforcement learning (RL), aimed at efficiently assessing a policy's value function. TD($\lambda$), a potent variant, incorporates a memory trace to distribute the prediction error into the historical context. However, this approach often neglects the significance of historical states and the relative importance of propagating the TD error, influenced by challenges such as visitation imbalance or outcome noise. To address this, we propose a novel TD algorithm named discerning TD learning (DTD), which allows flexible emphasis functions$-$predetermined or adapted during training$-$to allocate efforts effectively across states. We establish the convergence properties of our method within a specific class of emphasis functions and showcase its promising potential for adaptation to deep RL contexts. Empirical results underscore that employing a judicious emphasis function not only improves value estimation but also expedites learning across diverse scenarios.

In modern communication systems, efficient and reliable information dissemination is crucial for supporting critical operations across domains like disaster response, autonomous vehicles, and sensor networks. This paper introduces a Multi-Agent Reinforcement Learning (MARL) approach as a significant step forward in achieving more decentralized, efficient, and collaborative solutions. We propose a Partially Observable Stochastic Game (POSG) formulation for information dissemination empowering each agent to decide on message forwarding independently, based on their one-hop neighborhood. This constitutes a significant paradigm shift from traditional heuristics based on Multi-Point Relay (MPR) selection. Our approach harnesses Graph Convolutional Reinforcement Learning, employing Graph Attention Networks (GAT) with dynamic attention to capture essential network features. We propose two approaches, L-DGN and HL-DGN, which differ in the information that is exchanged among agents. We evaluate the performance of our decentralized approaches, by comparing them with a widely-used MPR heuristic, and we show that our trained policies are able to efficiently cover the network while bypassing the MPR set selection process. Our approach is a first step toward supporting the resilience of real-world broadcast communication infrastructures via learned, collaborative information dissemination.

Inverse reinforcement learning~(IRL) is a powerful framework to infer an agent's reward function by observing its behavior, but IRL algorithms that learn point estimates of the reward function can be misleading because there may be several functions that describe an agent's behavior equally well. A Bayesian approach to IRL models a distribution over candidate reward functions, alleviating the shortcomings of learning a point estimate. However, several Bayesian IRL algorithms use a $Q$-value function in place of the likelihood function. The resulting posterior is computationally intensive to calculate, has few theoretical guarantees, and the $Q$-value function is often a poor approximation for the likelihood. We introduce kernel density Bayesian IRL (KD-BIRL), which uses conditional kernel density estimation to directly approximate the likelihood, providing an efficient framework that, with a modified reward function parameterization, is applicable to environments with complex and infinite state spaces. We demonstrate KD-BIRL's benefits through a series of experiments in Gridworld environments and a simulated sepsis treatment task.

Human hands possess remarkable dexterity and have long served as a source of inspiration for robotic manipulation. In this work, we propose a human $\textbf{H}$and$\textbf{-In}$formed visual representation learning framework to solve difficult $\textbf{Dex}$terous manipulation tasks ($\textbf{H-InDex}$) with reinforcement learning. Our framework consists of three stages: (i) pre-training representations with 3D human hand pose estimation, (ii) offline adapting representations with self-supervised keypoint detection, and (iii) reinforcement learning with exponential moving average BatchNorm. The last two stages only modify $0.36\%$ parameters of the pre-trained representation in total, ensuring the knowledge from pre-training is maintained to the full extent. We empirically study 12 challenging dexterous manipulation tasks and find that H-InDex largely surpasses strong baseline methods and the recent visual foundation models for motor control. Code is available at https://yanjieze.com/H-InDex .

Inverse reinforcement learning (IRL) aims to explicitly infer an underlying reward function based on collected expert demonstrations. Considering that obtaining expert demonstrations can be costly, the focus of current IRL techniques is on learning a better-than-demonstrator policy using a reward function derived from sub-optimal demonstrations. However, existing IRL algorithms primarily tackle the challenge of trajectory ranking ambiguity when learning the reward function. They overlook the crucial role of considering the degree of difference between trajectories in terms of their returns, which is essential for further removing reward ambiguity. Additionally, it is important to note that the reward of a single transition is heavily influenced by the context information within the trajectory. To address these issues, we introduce the Distance-rank Aware Sequential Reward Learning (DRASRL) framework. Unlike existing approaches, DRASRL takes into account both the ranking of trajectories and the degrees of dissimilarity between them to collaboratively eliminate reward ambiguity when learning a sequence of contextually informed reward signals. Specifically, we leverage the distance between policies, from which the trajectories are generated, as a measure to quantify the degree of differences between traces. This distance-aware information is then used to infer embeddings in the representation space for reward learning, employing the contrastive learning technique. Meanwhile, we integrate the pairwise ranking loss function to incorporate ranking information into the latent features. Moreover, we resort to the Transformer architecture to capture the contextual dependencies within the trajectories in the latent space, leading to more accurate reward estimation. Through extensive experimentation, our DRASRL framework demonstrates significant performance improvements over previous SOTA methods.