Software Code for POMDP File Creator Tool (PCMRPP)

This is the code for the PCMRPP software package which implements an algorithm for automatically creating POMDP file for mobile robot path planning in a two-robot scenario. The Partially Observable Markov Decision Process (POMDP), a mathematical framework for decision-making in uncertain environments suffers from the curse of dimensionality. The POMDP model is made of three matrices - State Transition Matrix (T), Observation Matrix (O) and Reward Matrix (R). For a simplest case of two mobile robots moving in a bounded field, the number of states can easily cross 1000 leading to huge sizes of these matrices. The sizes of these matrices are of the order of square of no. of states. Even though the matrices are sparse, manually creating the high-dimension matrices is a cumbersome and sometimes even impossible task. For accurately defining the states and observations in discrete space, the size of POMDP model becomes very big. Creating such a file manually involves meticulously formulating the probabilities of huge number of state Transitions and Observations. The PCMRPP (POMDP file Creator for Mobile Robot Path Planning) is a novel algorithm to programmatically generate the POMDP model. This is implemented in C on Linux. This version is limited to only two mobile robots moving in an open 2D space with possibility of collision between them. The software creates a pomdp file which can be passed through a pomdp solver (e.g. APPL Toolkit - https://bigbird.comp.nus.edu.sg/pmwiki/farm/appl/). The resultant policy file either in the form of a graph or a set of alpha vectors is used by an executioner program running on-board the robot to generate the optimum robot path. The PCMRPP allows control of the sizes and the sparseness of the matrices to achieve trade-off between time complexity and the level of robustness of the POMDP solution. This is done by controlling the spread of the probability distribution of the state transition and the granularity of discretization of the components of the state. Refer to the figures included in the uploaded data set files to get more clarity on this.