Published in the journal Nature Machine Intelligence, the CSIRO-led research shows robots could soon be taking their engineering cues from evolution, using a new system of multi-level robotic design.
Acting as a universal designer, the system – known as Multi-Level Evolution (MLE) – would use algorithms based on natural evolution to automatically design robots for specific missions and adapt to different environments, by combining a variety of materials, components, sensors and behaviours.
La Trobe’s Professor Winkler said advanced, computer-based modelling could then rapidly test these robots in simulated, ‘real world’ scenarios to decide which prototype is best suited to each mission.
“The end result would be simple,” Professor Winkler said.
“Small, highly integrated, highly specialised, and highly cost-effective robots would be precision engineered for their task, environment, and terrain.
“MLE allows them to adapt on their own and automatically improve their performance.”
The research shows that within 20 years, using this cutting-edge technology, robots could be designed from the molecular level up to perform their mission in extremely challenging environments.
Lead author from the CSIRO, Dr David Howard, said current robots struggle in unstructured, complex environments because they aren't specialised enough, and should emulate the incredibly diverse adaptation animals have undergone to survive in their environment.
"Evolution doesn't care what something looks like. It searches a much wider design space and comes up with effective solutions that wouldn't be immediately obvious to a human designer," Dr Howard said.
"An animal like a manta ray or a kangaroo may look unusual to human eyes, but is perfectly calibrated for its environment."
Professor Winkler said MLE offers a more scalable process than conventional engineering approaches that require teams of engineers to design just one robot.
“The end result is the creation of a variety of robots that are designed to be specialised to specifics tasks or environments, as opposed to the versatile generalist robots we currently produce,” Professor Winkler said.
“One example would be a robot designed for basic environmental monitoring in extreme environments. It would need to traverse difficult terrain, gather data, and eventually fully degrade so as to not pollute the environment.
“Another example would be a robot designed for work in the Sahara Desert, using materials that can survive punishing heat, sand and dust. It could be solar powered, slide across sand dunes, and use the harsh UV light as a trigger to eventually degrade.”
The international research team comprises the CSIRO, La Trobe University, Vrije University in the Netherlands, the University of Lorraine in France, and Monash University.
Media contact: Dragana Mrkaja, d.mrkaja@latrobe.edu.au, 0447 508 171
