Path planning and learning strategies for mobile robots in dynamic partially unknown environments
Kristo Heero · DSpace repository (University of Tartu) · 2006
This thesis investigates path planning strategies for mobile robots in large partially unknown dynamic environments.The aim of this work is to reduce collision risk and time of path following in cases when robot repeatedly traverses between predefined target points (e.g. transportation or surveillance tasks).A novel path selection strategy is examined.The method creates innovative paths between pre-defined target points and learns to use paths that are more reliable.This approach is implemented on the research robot Khepera and verified against the shortest path following by a wave transform algorithm.Experimental data show that new approach is able to reduce collision risk, travel time and distance.The robot is also able to learn and adapt quickly in a changing environment.Test results show that trajectory planned by a wave transform algorithm is very difficult to predict and control, because even little unmodelled obstacles can cause a large deviation from the pre-planned path.The approach used in this thesis makes robot motion more predictable.This thesis also suggests that the behaviour of the robot depends strongly on the knowledge about it's surrounding but not on the path planning strategy used.It is concluded that in order to optimise travel time, distance and deviation one has to minimize the occurrence of unknown obstacles since the last one influences the former parameters.Finally, this thesis addresses the problem of the utility of exploration on time-critical mobile robot missions.It is argued that in large environments mission-oriented mobile robot applications can become more efficient if the exploration strategy considers knowledge already gained and its applicability during the rest of the mission. Path Selection for Mobile Robots in Dynamic Environments