Game-playing artificial intelligence (AI) systems have advanced to a new frontier. Stratego, the classic board game that's more complex than chess and Go, and craftier than poker, has now been mastered. Published in Science, we present DeepNash, an AI agent that learned the game from scratch to a human expert level by playing against itself. DeepNash uses a novel approach, based on game theory and model-free deep reinforcement learning. Its play style converges to a Nash equilibrium, which means its play is very hard for an opponent to exploit.

Reinforcement learning (RL) is an area of machine learning that focuses on training intelligent agents using related experiences so they can learn to solve decision making tasks, such as playing video games, flying stratospheric balloons, and designing hardware chips. Due to the generality of RL, the prevalent trend in RL research is to develop agents that can efficiently learn tabula rasa, that is, from scratch without using previously learned knowledge about the problem. However, in practice, tabula rasa RL systems are typically the exception rather than the norm for solving large-scale RL problems. Large-scale RL systems, such as OpenAI Five, which achieves human-level performance on Dota 2, undergo multiple design changes (e.g., algorithmic or architectural changes) during their developmental cycle. This modification process can last months and necessitates incorporating such changes without re-training from scratch, which would be prohibitively expensive.

Sampling-based Model Predictive Control (MPC) is a flexible control framework that can reason about non-smooth dynamics and cost functions. Recently, significant work has focused on the use of machine learning to improve the performance of MPC, often through learning or fine-tuning the dynamics or cost function. In contrast, we focus on learning to optimize more effectively. In other words, to improve the update rule within MPC. We show that this can be particularly useful in sampling-based MPC, where we often wish to minimize the number of samples for computational reasons. Unfortunately, the cost of computational efficiency is a reduction in performance; fewer samples results in noisier updates. We show that we can contend with this noise by learning how to update the control distribution more effectively and make better use of the few samples that we have. Our learned controllers are trained via imitation learning to mimic an expert which has access to substantially more samples. We test the efficacy of our approach on multiple simulated robotics tasks in sample-constrained regimes and demonstrate that our approach can outperform a MPC controller with the same number of samples.

Combinatorial optimisation problems framed as mixed integer linear programmes (MILPs) are ubiquitous across a range of real-world applications. The canonical branch-and-bound algorithm seeks to exactly solve MILPs by constructing a search tree of increasingly constrained sub-problems. In practice, its solving time performance is dependent on heuristics, such as the choice of the next variable to constrain ('branching'). Recently, machine learning (ML) has emerged as a promising paradigm for branching. However, prior works have struggled to apply reinforcement learning (RL), citing sparse rewards, difficult exploration, and partial observability as significant challenges. Instead, leading ML methodologies resort to approximating high quality handcrafted heuristics with imitation learning (IL), which precludes the discovery of novel policies and requires expensive data labelling. In this work, we propose retro branching; a simple yet effective approach to RL for branching. By retrospectively deconstructing the search tree into multiple paths each contained within a sub-tree, we enable the agent to learn from shorter trajectories with more predictable next states. In experiments on four combinatorial tasks, our approach enables learning-to-branch without any expert guidance or pre-training. We outperform the current state-of-the-art RL branching algorithm by 3-5x and come within 20% of the best IL method's performance on MILPs with 500 constraints and 1000 variables, with ablations verifying that our retrospectively constructed trajectories are essential to achieving these results.

This article proposes a model-based deep reinforcement learning (DRL) method to design emergency control strategies for short-term voltage stability problems in power systems. Recent advances show promising results in model-free DRL-based methods for power systems, but model-free methods suffer from poor sample efficiency and training time, both critical for making state-of-the-art DRL algorithms practically applicable. DRL-agent learns an optimal policy via a trial-and-error method while interacting with the real-world environment. And it is desirable to minimize the direct interaction of the DRL agent with the real-world power grid due to its safety-critical nature. Additionally, state-of-the-art DRL-based policies are mostly trained using a physics-based grid simulator where dynamic simulation is computationally intensive, lowering the training efficiency. We propose a novel model-based-DRL framework where a deep neural network (DNN)-based dynamic surrogate model, instead of a real-world power-grid or physics-based simulation, is utilized with the policy learning framework, making the process faster and sample efficient. However, stabilizing model-based DRL is challenging because of the complex system dynamics of large-scale power systems. We solved these issues by incorporating imitation learning to have a warm start in policy learning, reward-shaping, and multi-step surrogate loss. Finally, we achieved 97.5% sample efficiency and 87.7% training efficiency for an application to the IEEE 300-bus test system.

A critical challenge in multi-agent reinforcement learning (MARL) is for multiple agents to efficiently accomplish complex, long-horizon tasks. The agents often have difficulties in cooperating on common goals, dividing complex tasks, and planning through several stages to make progress. We propose to address these challenges by guiding agents with programs designed for parallelization, since programs as a representation contain rich structural and semantic information, and are widely used as abstractions for long-horizon tasks. Specifically, we introduce Efficient Multi-Agent Reinforcement Learning with Parallel Program Guidance (E-MAPP), a novel framework that leverages parallel programs to guide multiple agents to efficiently accomplish goals that require planning over 10+ stages. E-MAPP integrates the structural information from a parallel program, promotes the cooperative behaviors grounded in program semantics, and improves the time efficiency via a task allocator. We conduct extensive experiments on a series of challenging, long-horizon cooperative tasks in the Overcooked environment. Results show that E-MAPP outperforms strong baselines in terms of the completion rate, time efficiency, and zero-shot generalization ability by a large margin.

We consider task allocation for multi-object transport using a multi-robot system, in which each robot selects one object among multiple objects with different and unknown weights. The existing centralized methods assume the number of robots and tasks to be fixed, which is inapplicable to scenarios that differ from the learning environment. Meanwhile, the existing distributed methods limit the minimum number of robots and tasks to a constant value, making them applicable to various numbers of robots and tasks. However, they cannot transport an object whose weight exceeds the load capacity of robots observing the object. To make it applicable to various numbers of robots and objects with different and unknown weights, we propose a framework using multi-agent reinforcement learning for task allocation. First, we introduce a structured policy model consisting of 1) predesigned dynamic task priorities with global communication and 2) a neural network-based distributed policy model that determines the timing for coordination. The distributed policy builds consensus on the high-priority object under local observations and selects cooperative or independent actions. Then, the policy is optimized by multi-agent reinforcement learning through trial and error. This structured policy of local learning and global communication makes our framework applicable to various numbers of robots and objects with different and unknown weights, as demonstrated by numerical simulations.

Using massive datasets to train large-scale models has emerged as a dominant approach for broad generalization in natural language and vision applications. In reinforcement learning, however, a key challenge is that available data of sequential decision making is often not annotated with actions - for example, videos of game-play are much more available than sequences of frames paired with their logged game controls. We propose to circumvent this challenge by combining large but sparsely-annotated datasets from a \emph{target} environment of interest with fully-annotated datasets from various other \emph{source} environments. Our method, Action Limited PreTraining (ALPT), leverages the generalization capabilities of inverse dynamics modelling (IDM) to label missing action data in the target environment. We show that utilizing even one additional environment dataset of labelled data during IDM pretraining gives rise to substantial improvements in generating action labels for unannotated sequences. We evaluate our method on benchmark game-playing environments and show that we can significantly improve game performance and generalization capability compared to other approaches, using annotated datasets equivalent to only $12$ minutes of gameplay. Highlighting the power of IDM, we show that these benefits remain even when target and source environments share no common actions.

Unmanned combat air vehicle (UCAV) combat is a challenging scenario with continuous action space. In this paper, we propose a general hierarchical framework to resolve the within-vision-range (WVR) air-to-air combat problem under 6 dimensions of degree (6-DOF) dynamics. The core idea is to divide the whole decision process into two loops and use reinforcement learning (RL) to solve them separately. The outer loop takes into account the current combat situation and decides the expected macro behavior of the aircraft according to a combat strategy. Then the inner loop tracks the macro behavior with a flight controller by calculating the actual input signals for the aircraft. We design the Markov decision process for both the outer loop strategy and inner loop controller, and train them by proximal policy optimization (PPO) algorithm. For the inner loop controller, we design an effective reward function to accurately track various macro behavior. For the outer loop strategy, we further adopt a fictitious self-play mechanism to improve the combat performance by constantly combating against the historical strategies. Experiment results show that the inner loop controller can achieve better tracking performance than fine-tuned PID controller, and the outer loop strategy can perform complex maneuvers to get higher and higher winning rate, with the generation evolves.

Amazon and other e-commerce sites must employ mechanisms to protect their millions of customers from fraud, such as unauthorized use of credit cards. One such mechanism is order fraud evaluation, where systems evaluate orders for fraud risk, and either "pass" the order, or take an action to mitigate high risk. Order fraud evaluation systems typically use binary classification models that distinguish fraudulent and legitimate orders, to assess risk and take action. We seek to devise a system that considers both financial losses of fraud and long-term customer satisfaction, which may be impaired when incorrect actions are applied to legitimate customers. We propose that taking actions to optimize long-term impact can be formulated as a Reinforcement Learning (RL) problem. Standard RL methods require online interaction with an environment to learn, but this is not desirable in high-stakes applications like order fraud evaluation. Offline RL algorithms learn from logged data collected from the environment, without the need for online interaction, making them suitable for our use case. We show that offline RL methods outperform traditional binary classification solutions in SimStore, a simplified e-commerce simulation that incorporates order fraud risk. We also propose a novel approach to training offline RL policies that adds a new loss term during training, to better align policy exploration with taking correct actions.