Sequential Optimality and Coordination in Multiagent Systems

By Craig Boutilier. 1999

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Multi-Agent Common Knowledge Reinforcement Learning (2018, background): “Boutilier (1999) augments the state space with a coordination mechanism, to ensure coordination be-tween agents is possible in a fully observable multi-agent setting.”
Cooperative Inverse Reinforcement Learning (2016, background): “This is not the same as ‘solving’ a CIRL game, as a real world implementation of a CIRL agent must account for coordination problems and strategic uncertainty (Boutilier, 1999).”
Learning to Cooperate via Policy Search (2000, background): “Craig Boutilier and Michael Schwarz gave relevant ideas and comments.”
Evolution of the GPGP/TÆMS Domain-Independent Coordination Framework (2002, methodology): “This incremental and evolving approach to coordination is in contrast to a distributed MDP optimal policy where all contingencies based on the execution behavior of activities in agents are pre-analyzed and factored into the design of an optimal coordination policy [5, 80, 81].”
A Framework for Sequential Planning in Multi-Agent Settings (2005, background): “Some recent work in AI follows that tradition [5, 14, 17, 18].”
Reinforcement learning of coordination in cooperative multi-agent systems (2002, background): “Learning to coordinate in cooperative multi-agent systems is a central and widely studied problem, see, for example (Lauer & Riedmiller 2000), (Boutilier 1999), (Claus & Boutilier 1998), (Sen & Sekaran 1998), (Sen, Sekaran, & Hale 1994), (Weiss 1993).”
Multiagent Learning of Coordination in Loosely Coupled Multiagent Systems (2015, background): “Fundamental concepts are clarified and notations are established for further description.”
Coordination in multiagent reinforcement learning: a Bayesian approach (2003, background): “Boutilier [1] argues that the solution of games like this, assuming some (generally history dependent) mechanism for resolving these stage game coordination problems, requires explicit reasoning about the odds and benefits of coordination and its expected cost, and comparing these to the alternative”
Theoretical considerations of potential-based reward shaping for multi-agent systems (2011, unknown): “To demonstrate the theorised effects of potential-based reward shaping, an empirical study of a game based on Boutilier’s coordination game [4] will be presented here.”
An Algorithm for Distributed Reinforcement Learning in Cooperative Multi-Agent Systems (2000, background): “Some suggestions for solving the problem ofharmonization are given in Boutilier (1999).”
Hierarchical multi-agent reinforcement learning (2001, background): “of MDP (the MMDP model [5]), the multiagent extension of SMDP is not straight forward.”
A Cordial Sync: Going Beyond Marginal Policies for Multi-Agent Embodied Tasks (2020, background): “, a single agent with access to all observations from all agents [9,71,93].”
Lenient Learning in Independent-Learner Stochastic Cooperative Games (2016, background): “%) end Climb State 1 11 -30 0 -30 7 6 0 0 5 End < any, any > Climb Partially Stochastic State 1 11 -30 0 -30 14 / 0 6 0 0 5 End < any, any > Climb Fully Stochastic State 1 10 / 12 5 / -65 8 / -8 5 / -65 14 / 0 12 / 0 5 / -5 5 / -5 10 / 0 End < any, any > Penalty From (Claus and Boutilier, 1998).”
A Survey of Inverse Reinforcement Learning: Challenges, Methods and Progress (2018, background): “These classes include models for fully-observable state spaces (Markov games [82], multi-agent Markov decision processes [83], interaction-driven Markov games [84]) and partially-observable states (partially observable identical-payoff stochastic games [85], multi-agent team decision problems [86], ”
Accelerating Reinforcement Learning through Implicit Imitation (2003, background): “For example, see the fully cooperative multiagent MDP model proposed by Boutilier (1999). the set of joint actions; then T : S × A → ∆(S), with T (si, a)(sj) = Pr(si, a, sj) denoting the probability of ending up in state sj ∈ S when joint action a is performed at state si.”
Dynamic Programming for Partially Observable Stochastic Games (2004, background): “The theory of stochastic games provides a foundation for much recent work on multi-agent planning and learning [17, 3, 13, 4, 11].”
Multiagent Reinforcement Learning for Multi-Robot Systems: A Survey (2004, methodology): “In tackling the problem of coordination in multiagent systems, Boutilier [55] proposed a method for solving sequential multiagent decision problems by allowing agents to reason explicitly about specific coordination mechanisms.”