Learning a near optimal policy in a partially observable system remains an elusive challenge in contemporary reinforcement learning. In this work, we consider episodic reinforcement learning in a reward-mixing Markov decision process (MDP). There, a reward function is drawn from one of multiple possible reward models at the beginning of every episode, but the identity of the chosen reward model is not revealed to the agent. Hence, the latent state space, for which the dynamics are Markovian, is not given to the agent. We study the problem of learning a near optimal policy for two reward-mixing MDPs. Unlike existing approaches that rely on strong assumptions on the dynamics, we make no assumptions and study the problem in full generality. Indeed, with no further assumptions, even for two switching reward-models, the problem requires several new ideas beyond existing algorithmic and analysis techniques for efficient exploration. We provide the first polynomial-time algorithm that finds an $\epsilon$-optimal policy after exploring $\tilde{O}(poly(H,\epsilon^{-1}) \cdot S^2 A^2)$ episodes, where $H$ is time-horizon and $S, A$ are the number of states and actions respectively. This is the first efficient algorithm that does not require any assumptions in partially observed environments where the observation space is smaller than the latent state space.

Realistic manipulation tasks require a robot to interact with an environment with a prolonged sequence of motor actions. While deep reinforcement learning methods have recently emerged as a promising paradigm for automating manipulation behaviors, they usually fall short in long-horizon tasks due to the exploration burden. This work introduces MAnipulation Primitive-augmented reinforcement LEarning (MAPLE), a learning framework that augments standard reinforcement learning algorithms with a pre-defined library of behavior primitives. These behavior primitives are robust functional modules specialized in achieving manipulation goals, such as grasping and pushing. To use these heterogeneous primitives, we develop a hierarchical policy that involves the primitives and instantiates their executions with input parameters. We demonstrate that MAPLE outperforms baseline approaches by a significant margin on a suite of simulated manipulation tasks. We also quantify the compositional structure of the learned behaviors and highlight our method's ability to transfer policies to new task variants and to physical hardware. Videos and code are available at https://ut-austin-rpl.github.io/maple

This work addresses the problem of sensing the world: how to learn a multimodal representation of a reinforcement learning agent's environment that allows the execution of tasks under incomplete perceptual conditions. To address such problem, we argue for hierarchy in the design of representation models and contribute with a novel multimodal representation model, MUSE. The proposed model learns hierarchical representations: low-level modality-specific representations, encoded from raw observation data, and a high-level multimodal representation, encoding joint-modality information to allow robust state estimation. We employ MUSE as the sensory representation model of deep reinforcement learning agents provided with multimodal observations in Atari games. We perform a comparative study over different designs of reinforcement learning agents, showing that MUSE allows agents to perform tasks under incomplete perceptual experience with minimal performance loss. Finally, we evaluate the performance of MUSE in literature-standard multimodal scenarios with higher number and more complex modalities, showing that it outperforms state-of-the-art multimodal variational autoencoders in single and cross-modality generation.

Reinforcement learning is hard in general. Yet, in many specific environments, learning is easy. What makes learning easy in one environment, but difficult in another? We address this question by proposing a simple measure of reinforcement-learning hardness called the bad-policy density. This quantity measures the fraction of the deterministic stationary policy space that is below a desired threshold in value. We prove that this simple quantity has many properties one would expect of a measure of learning hardness. Further, we prove it is NP-hard to compute the measure in general, but there are paths to polynomial-time approximation. We conclude by summarizing potential directions and uses for this measure.

Popular off-policy deep reinforcement learning algorithms compensate for overestimation bias during temporal-difference learning by utilizing pessimistic estimates of the expected target returns. In this work, we propose a novel learnable penalty to enact such pessimism, based on a new way to quantify the critic's epistemic uncertainty. Furthermore, we propose to learn the penalty alongside the critic with dual TD-learning, a strategy to estimate and minimize the bias magnitude in the target returns. Our method enables us to accurately counteract overestimation bias throughout training without incurring the downsides of overly pessimistic targets. Empirically, by integrating our method and other orthogonal improvements with popular off-policy algorithms, we achieve state-of-the-art results in continuous control tasks from both proprioceptive and pixel observations.

Inferring the substitutable and complementary products for a given product is an essential and fundamental concern for the recommender system. To achieve this, existing approaches take advantage of the knowledge graphs to learn more evidences for inference, whereas they often suffer from invalid reasoning for lack of elegant decision making strategies. Therefore, we propose a novel Knowledge-Aware Path Reasoning (KAPR) model which leverages the dynamic policy network to make explicit reasoning over knowledge graphs, for inferring the substitutable and complementary relationships. Our contributions can be highlighted as three aspects. Firstly, we model this inference scenario as a Markov Decision Process in order to accomplish a knowledge-aware path reasoning over knowledge graphs. Secondly,we integrate both structured and unstructured knowledge to provide adequate evidences for making accurate decision-making. Thirdly, we evaluate our model on a series of real-world datasets, achieving competitive performance compared with state-of-the-art approaches. Our code is released on https://gitee.com/yangzijing flower/kapr/tree/master.

We teach goal-driven agents to interactively act and speak in situated environments by training on generated curriculums. Our agents operate in LIGHT (Urbanek et al., 2019)--a large-scale crowd-sourced fantasy text adventure game wherein an agent perceives and interacts with the world through textual natural language. Goals in this environment take the form of character-based quests, consisting of personas and motivations. We augment LIGHT by learning to procedurally generate additional novel textual worlds and quests to create a curriculum of steadily increasing difficulty for training agents to achieve such goals. In particular, we measure curriculum difficulty in terms of the rarity of the quest in the original training distribution--an easier environment is one that is more likely to have been found in the unaugmented dataset. An ablation study shows that this method of learning from the tail of a distribution results in significantly higher generalization abilities as measured by zeroshot performance on never-before-seen quests. Figure 1: The LIGHT questing environment presented as a 2 player game deployed in Messenger.

We develop a model selection approach to tackle reinforcement learning with adversarial corruption in both transition and reward. For finite-horizon tabular MDPs, without prior knowledge on the total amount of corruption, our algorithm achieves a regret bound of $\widetilde{\mathcal{O}}(\min\{\frac{1}{\Delta}, \sqrt{T}\}+C)$ where $T$ is the number of episodes, $C$ is the total amount of corruption, and $\Delta$ is the reward gap between the best and the second-best policy. This is the first worst-case optimal bound achieved without knowledge of $C$, improving previous results of Lykouris et al. (2021); Chen et al. (2021); Wu et al. (2021). For finite-horizon linear MDPs, we develop a computationally efficient algorithm with a regret bound of $\widetilde{\mathcal{O}}(\sqrt{(1+C)T})$, and another computationally inefficient one with $\widetilde{\mathcal{O}}(\sqrt{T}+C)$, improving the result of Lykouris et al. (2021) and answering an open question by Zhang et al. (2021b). Finally, our model selection framework can be easily applied to other settings including linear bandits, linear contextual bandits, and MDPs with general function approximation, leading to several improved or new results.

Prior AI successes in complex games have largely focused on settings with at most hundreds of actions at each decision point. In contrast, Diplomacy is a game with more than 10^20 possible actions per turn. Previous attempts to address games with large branching factors, such as Diplomacy, StarCraft, and Dota, used human data to bootstrap the policy or used handcrafted reward shaping. In this paper, we describe an algorithm for action exploration and equilibrium approximation in games with combinatorial action spaces. This algorithm simultaneously performs value iteration while learning a policy proposal network. A double oracle step is used to explore additional actions to add to the policy proposals. At each state, the target state value and policy for the model training are computed via an equilibrium search procedure. Using this algorithm, we train an agent, DORA, completely from scratch for a popular two-player variant of Diplomacy and show that it achieves superhuman performance. Additionally, we extend our methods to full-scale no-press Diplomacy and for the first time train an agent from scratch with no human data. We present evidence that this agent plays a strategy that is incompatible with human-data bootstrapped agents. This presents the first strong evidence of multiple equilibria in Diplomacy and suggests that self play alone may be insufficient for achieving superhuman performance in Diplomacy.

Off-policy reinforcement learning (RL) has proven to be a powerful framework for guiding agents' actions in environments with stochastic rewards and unknown or noisy state dynamics. In many real-world settings, these agents must operate in multiple environments, each with slightly different dynamics. For example, we may be interested in developing policies to guide medical treatment for patients with and without a given disease, or policies to navigate curriculum design for students with and without a learning disability. Here, we introduce nested policy fitted Q-iteration (NFQI), an RL framework that finds optimal policies in environments that exhibit such a structure. Our approach develops a nested $Q$-value function that takes advantage of the shared structure between two groups of observations from two separate environments while allowing their policies to be distinct from one another. We find that NFQI yields policies that rely on relevant features and perform at least as well as a policy that does not consider group structure. We demonstrate NFQI's performance using an OpenAI Gym environment and a clinical decision making RL task. Our results suggest that NFQI can develop policies that are better suited to many real-world clinical environments.