A linear programming approach for probabilistic robot path planning with missing information of outcomes

In practical robot motion planning, robots usually do not have full models of their surrounding, and hence no complete and correct plan exists for the robots to be executed fully. In most real-world problems a robot operates in just a partially-known environment, meaning that most of the environment is known to the robot at the time of planning, but there exists incomplete information about some `hidden´ variables which represent potential blockages (e.g. open/closed doors, or corridors congested with other robots or obstacles). For these hidden variables, the robot has a probability distribution estimation and a prioritized preference over their possible values. In this paper, to deal with the problem of choosing an optimal policy for planning in offline mode, a stochastic dynamic programming model is developed, which is converted to and solved by linear programming. Next, a heuristic method is proposed for conditional planning in the presence of numerous hidden variables which produces near-optimal plans.