A Distributed Mobile Robot Navigation by Snake Coordinated Vision Sensors

To date research in intelligent mobile agent have mainly focused on the development of a large and smart “brain” to enable robot autonomy (Arkin 2000; Murphy 2000). They are, however, facing a bottleneck of complexity due to the dynamics of the unstructured environments. Steering away from this smart brain approach, this chapter investigates the use of low level intelligence, such as insect eyes, and combines them to produce highly intelligent functions for autonomous robotic control by exploiting the creativity and diversity of insect eyes with small nervous systems (Land & Nilsson 2002) that mimics a mosaic eye. A mosaic eye transmits information through the retina to the insect's brain where they are integrated to form a usable picture of the insect's environment in order to co-ordinate their activities in response to any changes in the environment. Applying this concept to robotic control, the mosaic eye is used to assist a robot to find the shortest and safest path to reach its final destination. This is achieved through path planning in a dynamic environment and trajectory generation and control under robot non-holonomic constraints with control input saturations, subject to pre-defined criteria or constraints such as time and energy. By utilising pervasive intelligence (Snoonian 2003) distributed in the environment, robots can still maintain a high degree of mobility while utilising little computational functions and power. However, navigation techniques assisted by an environment with distributed information intelligence are different from conventional ones that rely on centralised intelligence implemented in the robot itself. These distributed navigation techniques need to be reconsidered and developed. The contour snake model (Kass et al. 1988) is broadly used and plays an important role in computer vision for image segmentation and contour tracking. Similar concepts have been applied to path planning with centralised robot navigation control using onboard sensors, such as elastic bands and bubbles (Quinlan & Khatib 1993; Quinlan 1994), connected splines (Mclean 1996) and redundant manipulators (Mclean & Cameron 1993; Cameron 1998). They all require a global planner to gather and process information. On the other hand, most of the existing work for robot navigation in an environment with small scale sensor networks considers only the high level path or discrete event planning(Sinopoli et al. 2003; Li & Rus 2005), ignoring issues related to low level trajectory planning and motion control due to sensor communication delay, timing skew, and discrete decision making. Low level