The challenge
Deploying robots into Australia's difficult environments
Australia presents a myriad of difficult environments which challenge the automation of any tasks within them. Whether deep inside a mine, operating non-stop to maintain an array of solar panels, or preserving biosecurity in a rainforest, successful deployment of robots into such environments would provide significant societal and economic impacts, but has to date eluded researchers.
Our response
Evolving robots piece by piece
Although robots struggle in these harsh environments, natural evolution consistently provides specialised life-forms that excel in them. Evolutionary Computing, the artificial analogue of natural evolution is, in essence, a stochastic population-based search algorithm. Using a technique inspired by Darwinian evolution, a population of candidate solutions are iteratively improved for a specific task in a given environment - in evolutionary field robotics, these solutions are either robots or their controllers.
The key benefits of the Evolutionary Field Robotics project include automatically generating control systems using advanced machine learning, and the ability to evolve parts of our robots to specific tasks and environmental conditions.
We focus on automatically generating high-performance controllers for real robots, and developing 'test beds' that allow us to perform such optimisation, for either legged or flying robots.
Bioinspired neural controllers allow our multirotors to adapt to environmental conditions faster than traditional controllers, and our advanced modelling algorithms automatically generate human-readable mathematical models of the robot.
Our robots need to work in the real world, so we also develop methods to make the transfer between simulation and reality more seamless.
As well as controller generation, we also study morphological evolution, for example evolving a hexapods leg for high performance in specific environments. Our science continues to generate many publications that are accepted for top-ranked venues.
The results
Creating more adaptive robots
The future will see our algorithms deployed across an increasing number of CSIRO's robotics platforms, and the development of increasingly complex and capable robot morphologies.