Planning under uncertainty in the continuous domain: A generalized belief space approach

摘要: This work investigates the problem of planning under uncertainty, with application to mobile robotics. We propose a probabilistic framework in which robot bases its decisions on generalized belief, is description own state and external variables interest. The approach naturally leads dual-layer architecture: an inner estimation layer, performs inference predict outcome possible decisions, outer decisional layer charge deciding best action undertake. does not discretize or control space, allows continuous domain. Moreover, it relax assumption maximum likelihood observations: predicted measurements are treated as random considered given. Experimental results show that our produces smooth trajectories while maintaining uncertainty within reasonable bounds.