Robotics Research

Applications: Navigation

In order for a mobile robot to move to a desired location in an efficient manner without collision its surroundings, some form of navigation is required. This apparently simple task has been the subject of extensive research since the first mobile robots in the early ’70s, and is still an open problem.

In the past, the general approach was to use advanced perception abilities to gather as much information as possible about the environment, and use this information to build a detailed internal representation of this environment which the robot would use to plan a safe path to the destination. This approach was popular because the intuitive sense-model-plan-act technique seemed to approximate the human solution to the problem. However, while this approach did have a measure of success in certain kinds of highly structured environments, overall it performed poorly because of the unreliable nature of the perceived data, and the high computational cost of modelling complex environments.  Also, any change in the robot’s surroundings required a reiteration of the algorithm, which led to poor performance in dynamic environments.

These perceived limitations in the traditional approach led to a new approach to mobile robot navigation, whereby a number of locally simple behaviours compete for control of the robot under a fixed arbitration scheme. Assuming appropriately designed behaviours, this approach led to complex emergent behaviour which performed very well in complex, dynamic environments. Rather than storing a detailed model of the world this technique effectively use the world as its own model. However, in order to achieve effective navigation in environments where not all information is available to the robot's sensors, it often necessary to use a high-level world model and planning architecture, while leaving low level navigation tasks in the hands of behaviours. This hybrid model is often used today. Also common today are a number of biologically inspired methods which often allow novel solutions to problems which are difficult to solve using conventional systems.

Navigation Research in Australia

• David Austin at the ANU has developed an operating system DROS. This employs a SLAM alogrithm which is used in navigation.
  
• Nick Barnes at ANU has developed a biologically inspired system to enable a mobile robot to circumnavigate an arbitrary object using vision.
  
• Gu Fang from the University of Western Sydney (UWS) is working on optimal trajectory planning in robotics.
  
• The IRRC at Monash University is also working on intelligent, collision free path planning.
  
• The Mechatronic Engineering Department at UNSW has developed an infra-red system able to passively detect remote beacons or use active reflectance to navigate.
  
• Phillip McKerrow at the University of Wollongong Intelligent Robotics Lab has developed a landmark-based navigation system which has been applied to a Titan outdoor robot.
• Shane Degen from Queensland University of Technology - ARCAA, is looking at UAV Guidance and Control in the application to the Uncooperative Collision Avoidance problem.

• 

  The Titan outdoor robot at the University of Wollongong uses a landmark navigation system to navigate in an outdoor environment.

• The ACFR is working on a number of navigation technologies including terrain-aided navigation and a hybrid model-based (quantitative) and behaviour-based (qualitative) system.
• Claude Sammut at UNSW is supervising a project to develop a mobile indoor guide robot. The robot can navigate around obstacles and map (model) the environment, and navigation capabilities allowing the robot to move to desired interesting locations are currently being developed.