Improved Memory-Bounded Dynamic Programming for Decentralized POMDPsProceedings of the Twenty-Third Conference on Uncertainty in Artificial Intelligence (2012)
AbstractMemory-Bounded Dynamic Programming (MBDP) has proved extremely effective in solving decentralized POMDPs with large horizons. We generalize the algorithm and improve its scalability by reducing the complexity with respect to the number of observations from exponential to polynomial. We derive error bounds on solution quality with respect to this new approximation and analyze the convergence behavior. To evaluate the effectiveness of the improvements, we introduce a new, larger benchmark problem. Experimental results show that despite the high complexity of decentralized POMDPs, scalable solution techniques such as MBDP perform surprisingly well.
Publication DateJune 20, 2012
Citation InformationSven Seuken and Shlomo Zilberstein. "Improved Memory-Bounded Dynamic Programming for Decentralized POMDPs" Proceedings of the Twenty-Third Conference on Uncertainty in Artificial Intelligence (2012)
Available at: http://works.bepress.com/shlomo_zilberstein/5/