We propose a novel framework for verifiable and compositional reinforcement learning (RL) in which a collection of RL sub-systems, each of which learns to accomplish a separate sub-task, are composed to achieve an overall task. The framework consists of a high-level model, represented as a parametric Markov decision process (pMDP) which is used to plan and to analyze compositions of sub-systems, and of the collection of low-level sub-systems themselves. By defining interfaces between the sub-systems, the framework enables automatic decompositons of task specifications, e.g., reach a target set of states with a probability of at least 0.95, into individual sub-task specifications, i.e. achieve the sub-system's exit conditions with at least some minimum probability, given that its entry conditions are met. This in turn allows for the independent training and testing of the sub-systems; if they each learn a policy satisfying the appropriate sub-task specification, then their composition is guaranteed to satisfy the overall task specification. Conversely, if the sub-task specifications cannot all be satisfied by the learned policies, we present a method, formulated as the problem of finding an optimal set of parameters in the pMDP, to automatically update the sub-task specifications to account for the observed shortcomings. The result is an iterative procedure for defining sub-task specifications, and for training the sub-systems to meet them. As an additional benefit, this procedure allows for particularly challenging or important components of an overall task to be determined automatically, and focused on, during training. Experimental results demonstrate the presented framework's novel capabilities.

Modern reinforcement learning (RL) commonly engages practical problems with large state spaces, where function approximation must be deployed to approximate either the value function or the policy. While recent progresses in RL theory address a rich set of RL problems with general function approximation, such successes are mostly restricted to the single-agent setting. It remains elusive how to extend these results to multi-agent RL, especially due to the new challenges arising from its game-theoretical nature. This paper considers two-player zero-sum Markov Games (MGs). We propose a new algorithm that can provably find the Nash equilibrium policy using a polynomial number of samples, for any MG with low multi-agent Bellman-Eluder dimension -- a new complexity measure adapted from its single-agent version (Jin et al., 2021). A key component of our new algorithm is the exploiter, which facilitates the learning of the main player by deliberately exploiting her weakness. Our theoretical framework is generic, which applies to a wide range of models including but not limited to tabular MGs, MGs with linear or kernel function approximation, and MGs with rich observations.

We propose ScheduleNet, a RL-based real-time scheduler, that can solve various types of multi-agent scheduling problems. We formulate these problems as a semi-MDP with episodic reward (makespan) and learn ScheduleNet, a decentralized decision-making policy that can effectively coordinate multiple agents to complete tasks. The decision making procedure of ScheduleNet includes: (1) representing the state of a scheduling problem with the agent-task graph, (2) extracting node embeddings for agent and tasks nodes, the important relational information among agents and tasks, by employing the type-aware graph attention (TGA), and (3) computing the assignment probability with the computed node embeddings.

An effective understanding of the contextual environment and accurate motion forecasting of surrounding agents is crucial for the development of autonomous vehicles and social mobile robots. This task is challenging since the behavior of an autonomous agent is not only affected by its own intention, but also by the static environment and surrounding dynamically interacting agents. Previous works focused on utilizing the spatial and temporal information in time domain while not sufficiently taking advantage of the cues in frequency domain. To this end, we propose a Spectral Temporal Graph Neural Network (SpecTGNN), which can capture inter-agent correlations and temporal dependency simultaneously in frequency domain in addition to time domain. SpecTGNN operates on both an agent graph with dynamic state information and an environment graph with the features extracted from context images in two streams. The model integrates graph Fourier transform, spectral graph convolution and temporal gated convolution to encode history information and forecast future trajectories. Moreover, we incorporate a multi-head spatio-temporal attention mechanism to mitigate the effect of error propagation in a long time horizon. We demonstrate the performance of SpecTGNN on two public trajectory prediction benchmark datasets, which achieves state-of-the-art performance in terms of prediction accuracy.

Action selection from many options with few constraints is crucial for improvisation and co-creativity. Our previous work proposed creative arc negotiation to solve this problem, i.e., selecting actions to follow an author-defined `creative arc' or trajectory over estimates of novelty, unexpectedness, and quality for potential actions. The CARNIVAL agent architecture demonstrated this approach for playing the Props game from improv theatre in the Robot Improv Circus installation. This article evaluates the creative arc negotiation experience with CARNIVAL through two crowdsourced observer studies and one improviser laboratory study. The studies focus on subjects' ability to identify creative arcs in performance and their preference for creative arc negotiation compared to a random selection baseline. Our results show empirically that observers successfully identified creative arcs in performances. Both groups also preferred creative arc negotiation in agent creativity and logical coherence, while observers enjoyed it more too.

We present a general optimization framework for emergent belief-state representation without any supervision. We employed the common configuration of multiagent reinforcement learning and communication to improve exploration coverage over an environment by leveraging the knowledge of each agent. In this paper, we obtained that recurrent neural nets (RNNs) with shared weights are highly biased in partially observable environments because of their noncooperativity. To address this, we designated an unbiased version of self-play via mechanism design, also known as reverse game theory, to clarify unbiased knowledge at the Bayesian Nash equilibrium. The key idea is to add imaginary rewards using the peer prediction mechanism (Miller et al., 2005), i.e., a mechanism for mutually criticizing information in a decentralized environment. Numerical analyses, including StarCraft exploration tasks with up to 20 agents and off-the-shelf RNNs, demonstrate the state-of-the-art performance.

We investigate the use of multi-agent allocation techniques on problems related to Earth observation scenarios with multiple users and satellites. We focus on the problem of coordinating users having reserved exclusive orbit portions and one central planner having several requests that may use some intervals of these exclusives. We define this problem as Earth Observation Satellite Constellation Scheduling Problem (EOSCSP) and map it to a Mixed Integer Linear Program. As to solve EOSCSP, we propose market-based techniques and a distributed problem solving technique based on Distributed Constraint Optimization (DCOP), where agents cooperate to allocate requests without sharing their own schedules. These contributions are experimentally evaluated on randomly generated EOSCSP instances based on real large-scale or highly conflicting observation order books.

When solving two-player zero-sum games, multi-agent reinforcement learning (MARL) algorithms often create populations of agents where, at each iteration, a new agent is discovered as the best response to a mixture over the opponent population. Within such a process, the update rules of "who to compete with" (i.e., the opponent mixture) and "how to beat them" (i.e., finding best responses) are underpinned by manually developed game theoretical principles such as fictitious play and Double Oracle. In this paper we introduce a framework, LMAC, based on meta-gradient descent that automates the discovery of the update rule without explicit human design. Specifically, we parameterise the opponent selection module by neural networks and the best-response module by optimisation subroutines, and update their parameters solely via interaction with the game engine, where both players aim to minimise their exploitability. Surprisingly, even without human design, the discovered MARL algorithms achieve competitive or even better performance with the state-of-the-art population-based game solvers (e.g., PSRO) on Games of Skill, differentiable Lotto, non-transitive Mixture Games, Iterated Matching Pennies, and Kuhn Poker. Additionally, we show that LMAC is able to generalise from small games to large games, for example training on Kuhn Poker and outperforming PSRO on Leduc Poker. Our work inspires a promising future direction to discover general MARL algorithms solely from data.

To build agents that can collaborate effectively with others, recent research has trained artificial agents to communicate with each other in Lewis-style referential games. However, this often leads to successful but uninterpretable communication. We argue that this is due to the game objective: communicating about a single object in a shared visual context is prone to overfitting and does not encourage language useful beyond concrete reference. In contrast, human language conveys a rich variety of abstract ideas. To promote such skills, we propose games that require communicating generalizations over sets of objects representing abstract visual concepts, optionally with separate contexts for each agent. We find that these games greatly improve systematicity and interpretability of the learned languages, according to several metrics in the literature. Finally, we propose a method for identifying logical operations embedded in the emergent languages by learning an approximate compositional reconstruction of the language.

Population-based multi-agent reinforcement learning (PB-MARL) refers to the series of methods nested with reinforcement learning (RL) algorithms, which produces a self-generated sequence of tasks arising from the coupled population dynamics. By leveraging auto-curricula to induce a population of distinct emergent strategies, PB-MARL has achieved impressive success in tackling multi-agent tasks. Despite remarkable prior arts of distributed RL frameworks, PB-MARL poses new challenges for parallelizing the training frameworks due to the additional complexity of multiple nested workloads between sampling, training and evaluation involved with heterogeneous policy interactions. To solve these problems, we present MALib, a scalable and efficient computing framework for PB-MARL. Our framework is comprised of three key components: (1) a centralized task dispatching model, which supports the self-generated tasks and scalable training with heterogeneous policy combinations; (2) a programming architecture named Actor-Evaluator-Learner, which achieves high parallelism for both training and sampling, and meets the evaluation requirement of auto-curriculum learning; (3) a higher-level abstraction of MARL training paradigms, which enables efficient code reuse and flexible deployments on different distributed computing paradigms. Experiments on a series of complex tasks such as multi-agent Atari Games show that MALib achieves throughput higher than 40K FPS on a single machine with $32$ CPU cores; 5x speedup than RLlib and at least 3x speedup than OpenSpiel in multi-agent training tasks. MALib is publicly available at https://github.com/sjtu-marl/malib.