New Zealand


Many real-world scenarios involve teams of agents who have to coordinate their policies to achieve a shared goal. Most previous studies mainly focus on learning decentralized control policies to maximize a common reward and rarely consider the coordinated control policies, which is critical in dynamic, complicated environments. Out of the fully decentralized rules, we propose a graph-based coordination strategy (GCS) which consists of a graph generator and a factored coordination policy. We adopt an encoder-decoder model as the generator to construct a directed acyclic graph (DAG), dynamically representing the underlying decision structure. We also apply the DAGness and depth constrained optimization in the graph generator to balance efficiency and performance. To exploit the underlying decision structure, we investigate the factored coordination policy to obtain the optimally coordinated behaviors among agents. We train the graph generator and the factored coordination policy simultaneously to maximize the discounted return. Empirical evaluations on Collaborative Gaussian Squeeze, Cooperative Navigation, and Google Research Football demonstrate the superiority of the proposed method.

AAMAS 2022

GCS: Graph-Based Coordination Strategy for Multi-Agent Reinforcement Learning

action coordination graph

multi-agent systems

reinforcement learning

AAMAS (International Conference on Autonomous Agents and Multiagent Systems) is the largest and most influential conference in the area of agents and multiagent systems. The aim of the conference is to bring together researchers and practitioners in all areas of agent technology and to provide a single, high-profile, internationally renowned forum for research in the theory and practice of autonomous agents and multiagent systems. AAMAS is the flagship conference of the non-profit International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS).

The AAMAS conference series was initiated in 2002 in Bologna, Italy as a joint event comprising the 6th International Conference on Autonomous Agents (AA), the 5th International Conference on Multiagent Systems (ICMAS), and the 9th International Workshop on Agent Theories, Architectures, and Languages (ATAL).

Subsequent AAMAS conferences have been held in Melbourne, Australia (July 2003), New York City, NY, USA (July 2004), Utrecht, The Netherlands (July 2005), Hakodate, Japan (May 2006), Honolulu, Hawaii, USA (May 2007), Estoril, Portugal (May 2008), Budapest, Hungary (May 2009), Toronto, Canada (May 2010), Taipei, Taiwan (May 2011), Valencia, Spain (June 2012), Minnesota, USA (May 2013), Paris, France (May 2014), Istanbul, Turkey (May 2015), Singapore (May 2016), São Paulo (2017) and Stockholm, Sweden (2018), Montreal (May 2019), Auckland (May 2020, Virtual), London (May 2021, Virtual).
<br>
<br>



A ticket is required to attend this event, please register using the link below:

https://aamas2022-conference.auckland.ac.nz/attending/registration/

Registration Is Required

AAMAS (International Conference on Autonomous Agents and Multiagent Systems) is the largest and most influential conference in the area of agents and multiagent systems. The aim of the conference is to bring together researchers and practitioners in all areas of agent technology and to provide a single, high-profile, internationally renowned forum for research in the theory and practice of autonomous agents and multiagent systems. AAMAS is the flagship conference of the non-profit International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS).

technical paper


We study a cooperative game setting where the grand coalition may change since the initial players can invite more players. We focus on the monotone games, i.e., adding more players to the grand coalition is not harmful. We model the invitation relationship as a directed acyclic graph. Our goal is to design a reward distribution mechanism for this new cooperative game setting such that players are incentivized to invite new players. In this paper, we propose the weighted permission Shapley value (inspired by permission structure and the weighted Shapley value) to achieve the goal. Our solution offers the very first attempt to incentivize players to invite more players to form a larger coalition via their private invitations in cooperative settings.

Incentives to Invite Others to Form Larger Coalitions


We initiate the study of the social welfare loss caused by corrupt auctioneers, both in single-item and multi-unit auctions. In our model, the auctioneer may collude with the winning bidders by letting them lower their bids in exchange for a (possibly bidder-dependent) fraction \gamma of the surplus. We consider different corruption schemes. In the most basic one, all winning bidders lower their bid to the highest losing bid. We show that this setting is equivalent to a \gamma-hybrid auction in which the payments are a convex combination of first-price and the second-price payments. More generally, we consider corruption schemes that can be related to \gamma-approximate first-price auctions (\gamma-FPA), where the payments recover at least a \gamma-fraction of the first-price payments. Our goal is to obtain a precise understanding of the robust price of anarchy (POA) of such auctions. If no restrictions are imposed on the bids, we prove a bound on the robust POA of \gamma-FPA which is tight (over the entire range of \gamma) for the single-item and the multi-unit auction setting. On the other hand, if the bids satisfy the no-overbidding assumption a more fine-grained landscape of the price of anarchy emerges, depending on the auction setting and the equilibrium notion. Albeit being more challenging, we derive (almost) tight bounds for both auction settings and several equilibrium notions, basically leaving open some (small) gaps for the coarse-correlated price of anarchy only.

Corruption in Auctions: Social Welfare Loss in Hybrid Multi-Unit Auctions


Bargaining can be used to resolve mixed-motive games in multi-agent systems. Although there is an abundance of negotiation strategies implemented in automated negotiating agents, most agents are based on single fixed strategies, while it is widely acknowledged that there is no single best-performing strategy for all negotiation settings.

In this paper, we focus on bargaining settings where opponents are repeatedly encountered, but the bargaining problems change. We introduce a novel method that automatically creates and deploys a portfolio of complementary negotiation strategies using a training set and optimise pay-off in never-before-seen bargaining settings through per-setting strategy selection. Our method relies on the following contributions. We introduce a feature representation that captures characteristics for both the opponent and the bargaining problem. We model the behaviour of an opponent during a negotiation based on its actions, which is indicative of its negotiation strategy, in order to be more effective in future encounters.

Our combination of feature-based methods generalises to new negotiation settings, as in practice, over time, it selects effective counter strategies in future encounters. Our approach is tested in an \gls{anac}-like tournament, and we show that we are capable of winning such a tournament with a \(5.6\% \) increase in pay-off compared to the runner-up agent.

Automated Configuration and Usage of Strategy Portfolios for Bargaining



We study strategic network formation games in which agents at- tempt to form (costly) links in order to maximize their network centrality. Our model derives from Jackson and Wolinsky’s symmetric connection model, but allows for heterogeneity in agent utilities by replacing decay centrality (implicit in the J-W model) by a va- riety of classical centrality measures, as well as game-theoretic measures of centrality. We are primarily interested in characteriz- ing the asymptotically pairwise stable networks, i.e. those networks that are pairwise stable for all sufficiently small, positive edge costs. We uncover a rich typology of stability:

- we give an axiomatic approach to network centrality that allows us to predict the stable network for a rich set of com- bination of centrality utility functions, yielding stable networks with features reminiscent of structural properties such as "core periphery" and "rich club" networks.
- We show that a simple variation on the model renders it universal, i.e. every network may be a stable network.
- We also show that often we can infer a significant amount about agent utilities from the structure of stable networks.

Being Central on the Cheap: Stability in Heterogeneous Multiagent Centrality Games


The potential of reinforcement learning (RL) to deliver aligned and performant agents is partially bottlenecked by the reward engineering problem. One alternative to heuristic trial-and-error is preference-based RL (PbRL), where a reward function is inferred from sparse human feedback. However, prior PbRL methods lack interpretability of the learned reward structure, which hampers the ability to assess robustness and alignment. We propose an online, active preference learning algorithm that constructs reward functions with the intrinsically interpretable, compositional structure of a tree. Using both synthetic and human-provided feedback, we demonstrate sample-efficient learning of tree-structured reward functions in several environments, then harness the enhanced interpretability to explore and debug for alignment.

Interpretable Preference-based Reinforcement Learning with Tree-Structured Reward Functions


As artificial agents become increasingly prevalent in our daily lives, it becomes imperative to equip them with an awareness of societal norms; specifically, the ability to account for and be considerate towards others they may cohabit with. In this work, we explore the ability for an agent trained through reinforcement learning to exhibit sympathetic behaviours towards another (independent) agent in the environment. We propose to achieve such behaviors by first inferring the reward function of the independent agent, through inverse reinforcement learning, and subsequently learning a policy based on a sympathetic reward function - a convex combination of the inferred rewards together with the agent's own rewards. The corresponding weighting function, which we refer to as the sympathy function, is computed online as a function of the estimated return of the agent's taken action relative to that of other possible actions it could have taken. We evaluate our approach on adversarial as well as assistive environment settings, and demonstrate the ability of our sympathetic agent to perform well at its own goal, while simultaneously giving due consideration to another agent in its environment. We also perform systematic ablation studies to identify the influence of the various components of our proposed framework.

Sympathy based Reinforcement Learning agents


As reinforcement learning (RL) continues to improve and be applied in situations alongside humans, the need to explain the learned behaviors of RL agents to end-users becomes more important. Strategies for explaining the reasoning behind an agent's policy, called policy-level explanations, can lead to important insights about both the task and the agent's behaviors. Following this line of research, in this work, we propose a novel approach that summarizes an agent's policy in the form of a directed graph with natural language descriptions. A decision tree based clustering method is utilized to abstract the state space of the task into fewer, condensed states which makes the policy graphs more digestible to end-users. This abstraction allows the users to control the size of the policy graph to achieve their desired balance between comprehensibility and accuracy. In addition, we develop a heuristic optimization method to find the most explainable graph policy and present it to the users. Finally, we use the user-defined predicates to enrich the abstract states with semantic meaning. We test our approach on five RL tasks, using both deterministic and stochastic policies, and show that our method is: (1) agnostic to the algorithms used to train the policies, and (2) comparable in accuracy and superior in explanation capabilities to existing baselines. Especially, when provided with our explanation graph, end-users are able to accurately interpret policies of trained RL agents 80% of the time, compared to 10% when provided with the next best baseline. We uploaded our code and datasets as supplementary material for ensuring the reproducibility of our research findings.

CAPS: Comprehensible Abstract Policy Summaries for Explaining Reinforcement Learning Agents


We address the problem of routing a team of drones and trucks over large-scale urban road networks. To conserve their limited flight energy, drones can use trucks as temporary modes of transit en route to their own destinations. Such coordination can yield significant savings in total vehicle distance traveled, i.e., truck travel distance and drone flight distance, compared to operating drones and trucks independently. But it comes at the potentially prohibitive computational cost of deciding which trucks and drones should coordinate and when and where it is most beneficial to do so. We tackle this fundamental trade-off by decoupling our overall intractable problem into tractable sub-problems that we solve stage-wise. The first stage solves only for trucks, by computing paths that make them more likely to be useful transit options for drones. The second stage solves only for drones, by routing them over a composite of the road network and the transit network defined by truck paths from the first stage. We design a comprehensive algorithmic framework that frames each stage as a multi-agent path finding problem and implement two distinct methods for solving them. We evaluate our approach on extensive simulations with up to 100 agents on the real-world Manhattan road network containing nearly 4500 vertices and 10000 edges. Our framework saves on more than 50% of vehicle distance traveled compared to independently solving for trucks and drones, and computes solutions for all settings within 5 minutes on commodity hardware.

Coordinated Multi-Agent Path Finding for Drones and Trucks over Road Networks


High-performing teams learn intelligent and efficient communication and coordination strategies to maximize their joint utility. These teams implicitly understand the different roles of heterogeneous team members and adapt their communication protocols accordingly. Multi-Agent Reinforcement Learning (MARL) seeks to develop computational methods for synthesizing such coordination strategies, but formulating models for heterogeneous teams with different state, action, and observation spaces has remained an open problem. Without properly modeling agent heterogeneity, as in prior MARL work that leverages homogeneous graph networks, communication becomes less helpful and can even deteriorate the cooperativity and team performance. We propose Heterogeneous Policy Networks (HetNet) to learn efficient and diverse communication models for coordinating cooperative heterogeneous teams. Building on heterogeneous graph-attention networks, we show that HetNet not only facilitates learning heterogeneous collaborative policies per existing agent-class but also enables end-to-end training for learning highly efficient binarized messaging. Our empirical evaluation shows that HetNet sets a new state of the art in learning coordination and communication strategies for heterogeneous multi-agent teams by achieving an 8.1% to 98.7% performance improvement over the next-best baseline across multiple domains while simultaneously achieving a 200x reduction in the required communication bandwidth.

Learning Efficient Diverse Communication for Cooperative Heterogeneous Teaming


In online advertising, search engines sell ad placements for keywords continuously through auctions. This problem can be seen as an infinitely repeated game since the auction is executed whenever a user performs a query with the keyword. As advertisers may frequently change their bids, the game will have a large set of equilibria with potentially complex strategies. In this paper, we propose the use of natural strategies for reasoning in such setting as they are processable by artificial agents with limited memory and/or computational power as well as understandable by humans users. To reach this goal, we introduce a quantitative version of Strategy Logic with natural strategies in the setting of imperfect information. In a first step, we show how to model strategies for repeated keyword auctions and take advantage of the model for proving properties evaluating this game. In a second step, we study the logic in relation to the distinguishable power, expressivity, and model-checking complexity for strategies with and without recall.

GCS: Graph-Based Coordination Strategy for Multi-Agent Reinforcement Learning

Downloads

Next from AAMAS 2022

Incentives to Invite Others to Form Larger Coalitions

Similar lecture

Multi-Agent Curricula and Emergent Implicit Signaling

GCS: Graph-Based Coordination Strategy for Multi-Agent Reinforcement Learning

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from AAMAS 2022

Incentives to Invite Others to Form Larger Coalitions

Similar lecture

Multi-Agent Curricula and Emergent Implicit Signaling

Downloads