Pommerman is a hybrid cooperative/adversarial multi-agent environment, with challenging characteristics in terms of partial observability, limited or no communication, sparse and delayed rewards, and restrictive computational time limits. This makes it a challenging environment for reinforcement learning (RL) approaches. In this paper, we focus on developing a curriculum for learning a robust and promising policy in a constrained computational budget of 100,000 games, starting from a fixed base policy (which is itself trained to imitate a noisy expert policy). All RL algorithms starting from the base policy use vanilla proximal-policy optimization (PPO) with the same reward function, and the only difference between their training is the mix and sequence of opponent policies. One expects that beginning training with simpler opponents and then gradually increasing the opponent difficulty will facilitate faster learning, leading to more robust policies compared against a baseline where all available opponent policies are introduced from the start. We test this hypothesis and show that within constrained computational budgets, it is in fact better to "learn in the school of hard knocks", i.e., against all available opponent policies nearly from the start. We also include ablation studies where we study the effect of modifying the base environment properties of ammo and bomb blast strength on the agent performance.

For their study, Misra and his team chose a type of machine-learning algorithm based on the concept of reinforcement learning. Simply put, the software …

Thermal power generation plays a dominant role in the world's electricity supply. It consumes large amounts of coal worldwide, and causes serious air pollution. Optimizing the combustion efficiency of a thermal power generating unit (TPGU) is a highly challenging and critical task in the energy industry. We develop a new data-driven AI system, namely DeepThermal, to optimize the combustion control strategy for TPGUs. At its core, is a new model-based offline reinforcement learning (RL) framework, called MORE, which leverages logged historical operational data of a TPGU to solve a highly complex constrained Markov decision process problem via purely offline training. MORE aims at simultaneously improving the long-term reward (increase combustion efficiency and reduce pollutant emission) and controlling operational risks (safety constraints satisfaction). In DeepThermal, we first learn a data-driven combustion process simulator from the offline dataset. The RL agent of MORE is then trained by combining real historical data as well as carefully filtered and processed simulation data through a novel restrictive exploration scheme. DeepThermal has been successfully deployed in four large coal-fired thermal power plants in China. Real-world experiments show that DeepThermal effectively improves the combustion efficiency of a TPGU. We also report and demonstrate the superior performance of MORE by comparing with the state-of-the-art algorithms on the standard offline RL benchmarks. To the best knowledge of the authors, DeepThermal is the first AI application that has been used to solve real-world complex mission-critical control tasks using the offline RL approach.

This paper investigates the motion planning of autonomous dynamical systems modeled by Markov decision processes (MDP) with unknown transition probabilities over continuous state and action spaces. Linear temporal logic (LTL) is used to specify high-level tasks over infinite horizon, which can be converted into a limit deterministic generalized B\"uchi automaton (LDGBA) with several accepting sets. The novelty is to design an embedded product MDP (EP-MDP) between the LDGBA and the MDP by incorporating a synchronous tracking-frontier function to record unvisited accepting sets of the automaton, and to facilitate the satisfaction of the accepting conditions. The proposed LDGBA-based reward shaping and discounting schemes for the model-free reinforcement learning (RL) only depend on the EP-MDP states and can overcome the issues of sparse rewards. Rigorous analysis shows that any RL method that optimizes the expected discounted return is guaranteed to find an optimal policy whose traces maximize the satisfaction probability. A modular deep deterministic policy gradient (DDPG) is then developed to generate such policies over continuous state and action spaces. The performance of our framework is evaluated via an array of OpenAI gym environments.

Reinforcement learning in large-scale environments is challenging due to the many possible actions that can be taken in specific situations. We have previously developed a means of constraining, and hence speeding up, the search process through the use of motion primitives; motion primitives are sequences of pre-specified actions taken across a state series. As a byproduct of this work, we have found that if the motion primitives' motions and actions are labeled, then the search can be sped up further. Since motion primitives may initially lack such details, we propose a theoretically viewpoint-insensitive and speed-insensitive means of automatically annotating the underlying motions and actions. We do this through a differential-geometric, spatio-temporal kinematics descriptor, which analyzes how the poses of entities in two motion sequences change over time. We use this descriptor in conjunction with a weighted-nearest-neighbor classifier to label the primitives using a limited set of training examples. In our experiments, we achieve high motion and action annotation rates for human-action-derived primitives with as few as one training sample. We also demonstrate that reinforcement learning using accurately labeled trajectories leads to high-performing policies more quickly than standard reinforcement learning techniques. This is partly because motion primitives encode prior domain knowledge and preempt the need to re-discover that knowledge during training. It is also because agents can leverage the labels to systematically ignore action classes that do not facilitate task objectives, thereby reducing the action space.

Real world applications such as economics and policy making often involve solving multi-agent games with two unique features: (1) The agents are inherently asymmetric and partitioned into leaders and followers; (2) The agents have different reward functions, thus the game is general-sum. The majority of existing results in this field focuses on either symmetric solution concepts (e.g. Nash equilibrium) or zero-sum games. It remains vastly open how to learn the Stackelberg equilibrium -- an asymmetric analog of the Nash equilibrium -- in general-sum games efficiently from samples. This paper initiates the theoretical study of sample-efficient learning of the Stackelberg equilibrium in two-player turn-based general-sum games. We identify a fundamental gap between the exact value of the Stackelberg equilibrium and its estimated version using finite samples, which can not be closed information-theoretically regardless of the algorithm. We then establish a positive result on sample-efficient learning of Stackelberg equilibrium with value optimal up to the gap identified above. We show that our sample complexity is tight with matching upper and lower bounds. Finally, we extend our learning results to the setting where the follower plays in a Markov Decision Process (MDP), and the setting where the leader and the follower act simultaneously.

Reinforcement learning (RL) is one of the most active fields of AI research. Despite the interest demonstrated by the research community in reinforcement learning, the development methodology still lags behind, with a severe lack of standard APIs to foster the development of RL applications. OpenAI Gym is probably the most used environment to develop RL applications and simulations, but most of the abstractions proposed in such a framework are still assuming a semi-structured methodology. This is particularly relevant for agent-based models whose purpose is to analyse adaptive behaviour displayed by self-learning agents in the simulation. In order to bridge this gap, we present a workflow and tools for the decoupled development and maintenance of multi-purpose agent-based models and derived single-purpose reinforcement learning environments, enabling the researcher to swap out environments with ones representing different perspectives or different reward models, all while keeping the underlying domain model intact and separate. The Sim-Env Python library generates OpenAI-Gym-compatible reinforcement learning environments that use existing or purposely created domain models as their simulation back-ends. Its design emphasizes ease-of-use, modularity and code separation.

Offline reinforcement learning approaches can generally be divided to proximal and uncertainty-aware methods. In this work, we demonstrate the benefit of combining the two in a latent variational model. We impose a latent representation of states and actions and leverage its intrinsic Riemannian geometry to measure distance of latent samples to the data. Our proposed metrics measure both the quality of out of distribution samples as well as the discrepancy of examples in the data. We integrate our metrics in a model-based offline optimization framework, in which proximity and uncertainty can be carefully controlled. We illustrate the geodesics on a simple grid-like environment, depicting its natural inherent topology. Finally, we analyze our approach and improve upon contemporary offline RL benchmarks.

The softmax policy gradient (PG) method, which performs gradient ascent under softmax policy parameterization, is arguably one of the de facto implementations of policy optimization in modern reinforcement learning. For $\gamma$-discounted infinite-horizon tabular Markov decision processes (MDPs), remarkable progress has recently been achieved towards establishing global convergence of softmax PG methods in finding a near-optimal policy. However, prior results fall short of delineating clear dependencies of convergence rates on salient parameters such as the cardinality of the state space $\mathcal{S}$ and the effective horizon $\frac{1}{1-\gamma}$, both of which could be excessively large. In this paper, we deliver a pessimistic message regarding the iteration complexity of softmax PG methods, despite assuming access to exact gradient computation. Specifically, we demonstrate that softmax PG methods can take exponential time -- in terms of $|\mathcal{S}|$ and $\frac{1}{1-\gamma}$ -- to converge, even in the presence of a benign policy initialization and an initial state distribution amenable to exploration. This is accomplished by characterizing the algorithmic dynamics over a carefully-constructed MDP containing only three actions. Our exponential lower bound hints at the necessity of carefully adjusting update rules or enforcing proper regularization in accelerating PG methods.

In many contemporary applications such as healthcare, finance, robotics, and recommendation systems, continuous deployment of new policies for data collection and online learning is either cost ineffective or impractical. We consider a setting that lies between pure offline reinforcement learning (RL) and pure online RL called deployment constrained RL in which the number of policy deployments for data sampling is limited. To solve this challenging task, we propose a new algorithmic learning framework called Model-based Uncertainty regularized and Sample Efficient Batch Optimization (MUSBO). Our framework discovers novel and high quality samples for each deployment to enable efficient data collection. During each offline training session, we bootstrap the policy update by quantifying the amount of uncertainty within our collected data. In the high support region (low uncertainty), we encourage our policy by taking an aggressive update. In the low support region (high uncertainty) when the policy bootstraps into the out-of-distribution region, we downweight it by our estimated uncertainty quantification. Experimental results show that MUSBO achieves state-of-the-art performance in the deployment constrained RL setting.