Abstract:
Real-time navigation and mapping of an autonomous robot is one of the major challenges in autonomous robot systems. In this dissertation, a novel sensor-based biologically inspired neural network algorithm to real-time collision-free navigation and mapping of an autonomous mobile robot in a completely unknown environment is proposed. The topologically organized neural network with nonlinear analog neurons is efficient for trajectory planning with obstacle avoidance, in which each neuron in the neural network is described by a Glasius and Komoda equation. The proposed biologically-inspired neural network model for motion planning with safety consideration is studied in known and unknown map of environments. A virtual obstacle methodology is developed to enlarge the obstacle to have safer distance from the mobile robot to the obstacles.
A two-level LIDAR-driven hybrid real-time mapping and navigation system is proposed by using the biologically inspired neural network algorithm of an autonomous robot. A local map composed of cells is built up through the proposed neural dynamics for robot navigation with restricted incoming sensory information. According to the measured sensory information, an accurate map with grid representation of the robot with local environment is dynamically built for the robot navigation. The proposed model for autonomous robot navigation and mapping is capable of planning a real-time reasonable and safe trajectory of an autonomous robot.
A distance matrix scheme associated with the biologically inspired neural network model is developed to achieve safe navigation in an indoor environment. A safe navigation scheme is developed to reduce collision risk considering distance from the robot to obstacles.
Simulation and comparison studies are carried out to demonstrate the effectiveness and efficiency of the proposed methodologies that concurrently performs collision-free navigation and mapping of an autonomous robot.
