Safe path planning in an uncertain-configuration space

Abstract

The objective of this paper is to bring an effective response to the safe path planning problem which should be solved in an uncertain-configuration space. Firstly, a path planning method dealing with localization uncertainties is proposed, where the uncertainties in both position and orientation of a non-holonomic mobile robot are considered. The safety of this method is due to the mixing of the planning phase and the naviyation phase using the same process of localization (the Kalman filter). Secondly, while previous works planned safe paths in the configuration space, we show that it is nece s s a y to plan safe paths in an uncertain-configuration space. Then, we introduce the novel concept of "towers of uncertainties" and show the effectiveness of this concept with some examples.0-7803-7736-2/03/$17.00 02003 IEEE

A Bayesian framework for optimal motion planning with uncertainty

“…In some cases, the plan is a function from the compressed belief space to the actions [10]. In some other cases ( [1]), the plan is a reference trajectory, along which the motion is guaranteed to be safe, in spite of uncertainty -this is also our case. Finally, the output of some methods ( [13], [4]) is a 'sensorbased motion': a sequence of action/termination condition, where the condition depends on the sensory input ("walk until you see the wall").…”

“…The straightforward possibility is to propagate the mean/covariance of the distribution (e.g., [1], [9]). Some methods operate in a reduced information space ( [10]), in which the belief is compressed by using a small set of parameters.…”

“…The methods reviewed here have solved the problem in a variety of ways, covering most, if not all, known planning methods: back-projection [13], value/policy iteration [10], gradient descent [5], analytic closed-form solution [6], multi-sine optimization [14], classical graph search [1], dynamic programming [15], RRT [2], POMDPs [16].…”

“…This paper tries to solve the problem by working in the extended space of poses×covariances, as has already been done in [1], [2]. The following are the main contributions of this paper with respect to previous work.…”

“…• We extend the forward search algorithm first described in [1]. The robot is given the option to stay still for better localizing itself, and we use a more powerful dominance heuristics.…”