Robotics Research

Applications: Systems Infrastructure

The overall architecture and design of robotic systems has recently been the subject of an increasing amount of research. The research is aimed at finding new solutions to difficult problems in robotics by the use of novel system structures. Historically, most robotic systems were fairly simple in structure – information perceived by sensors was routed to some form of central processor, which used this sensor data to generate actuator commands. This type of system was well suited to the traditional sense–plan–act approach used in navigation and similar tasks, and is still the most commonly found system design. It is an intuitive architecture, and with appropriate software in the central processor it can be used successfully in most applications.

However, there are some drawbacks to such a system. The centralised nature of the structure can lead to bottlenecks if the central processor is unable to cope with the large amount of processing required by a complex system. In order to ensure that these bottlenecks do not occur a fast, expensive processor is often required, which leads to increased system cost. The centralisation also results in a fairly brittle system, where damage to the processing unit will cause the entire system to fail. This kind of system is also criticised for a lack of modularity – once the system is in place it is difficult to add further functionality without significant and often costly changes.  

These concerns have led to extensive research into alternative system structures able to offer increased modularity, robustness and stability. Many of these systems are also simpler to implement when designed for appropriate applications. Most of these systems are based around some form of distribution, whether over multiple robots or multiple elements within the same robot.

System Infrastructure and Design Research in Australia

• David Austin from ANU has been researching multi-robot mapping (modelling) using multiple mobile robots to explore an area quicker and with less error. He has also developed an operating system DROS which consists mainly of support functions for modular programming and modules for mobile robots.
• Roy Featherstone, also from ANU, has developed a method for calculating the dynamics of articulated systems very quickly using a parallel computing architecture.
• Researchers at UNSW Mechatronic Engineering department have conducted experiments in multi-robot systems, particularly in the use reinforcement, fuzzy, neural and genetic techniques to produce coordinated movement and sharing of experience.
• The ACFR is involved in a number of systems related projects. They are using local data fusion techniques to produce a decentralised sensing system which is highly modular and requires no global communication channel or central processor. They also have a group of fixed wing aircraft cooperating to perform autonomous sensing and navigation.
• The RSL at ANU is developing a school of small-scale submersibles with the aim of creating a scalable, fault-tolerant network of autonomous yet cooperating vehicles.

• 

  The submersibles under development in the Serafina project at the Australian National University. The goal of the project is to create a group of autonomous yet cooperating vehicles.

• Research at the University of Queensland has demonstrated strategies for team coordination in distributed mobile robot systems using data fusion of the disparate world models.