Monte Carlo Localization with Field Lines Observations for Simulated Humanoid Robotic Soccer

Cooperative Localization for Multiple Soccer Agents Using Factor Graphs and Sequential Monte Carlo

“…Content may change prior to final publication. On the other hand, previous results from our robotics team reported in [27] show that a non-cooperative Monte Carlo localization implementation running with N p = 200 particles in a similar scenario would have a CPU time of 0.025 ms if run on an Intel® Core™ i7-4720HQ CPU @ 2.60 GHz, a typical setup for the robot soccer simulation. The third column in Table 2 shows the projected CPU time for the proposed algorithms using that value as a time estimate for an alternative, efficient implementation.…”

“…On the other hand, since each landmark, DOA or ally position measurement can be independently assimilated, we can use different assimilation schemes within the observation step leading to different versions of the proposed filter. Specifically, in order to compare to which degree the cooperation improves the state estimation, we propose four variants of the cooperative filter: 1) P L. Non-cooperative filter in which the observation step consists solely of landmark assimilation, similar to [27]. Steps 1 and 2 of Algorithm 1 are performed.…”

“…) Robot dynamicsWe use the dynamic model given in[27],[35],[36]. The system input is the vector u r,n given by u r,n = ∆x r,n ∆y r,n ∆ψ r,n T ,(2)VOLUME 4, 2016…”

Influence of Location and Number of Landmarks on the Monte Carlo Localization Problem

“…Desde que foi proposta, MCL vêm sendo aplicada e generalizada para diversos contextos onde surgem problemas de localização [Torma et al 2010, Payá et al 2010, Elinas and Little 2005, Milstein 2008, Muzio et al 2016, Metropolis et al 1953, Howard 2006]. Entretanto, nenhum destes trabalhos abordou a questão da quantidade e posicionamento dos pontos de referência no desempenho do MCL, queé o foco deste trabalho.…”

“…Uma técnica para resolver este problema de localização globalé a Monte Carlo Localization (MCL), que apesar de proposta em 1999 [Fox et al 1999, Thruna et al 2001, vem sendo aplicada em cada vez mais cenários [Muzio et al 2016]. De forma sucinta, o MCL utiliza inferência bayesiana recursiva, gerando amostras aleatórias de possíveis locais para o robô (chamadas de partículas), dando mais peso para locais que condizem com suas observações (pontos de referência) e ações do robô.…”