Test Bed Vision:
This research test bed focuses on automatically designing robots, or robot components, with certain desired physical properties that are derived from the environments in which they function. It is poised to solve the problem of “Task-based design” – the notion that bespoke, specialized parts are often higher-performing than common-off-the-shelf options when opeating under certain conditions. The main technologies are both software (autonomous science algorithms, autonomous product verification) and hardware (physical robots or components).
From the design of space-bound aerials for NASA satellites, to the generation of parts for modern automobiles, and the intelligence behind our stock trading systems, evolutionary computing has been successfully deployed again and again to meet real-world requirements. Taking their inspiration from genetics and Darwinian “survival of the fittest”, these population-based optimisation algorithms generate unconventional solutions to problems that human optimisers find difficult – or impossible – to solve.
Robotics is a key area for the future economic health of the nation, but the uptake of robotics is being hindered by a lack of capability: robots are complex to design and produce, tricky to program, and have limited off-the-shelf configurations, with no option to physically adapt to their environments and specific tasks. The need for adaptation is especially prevalent in cases with unforgiving environmental pressures, including disaster response, and long-term remote sensing. Autonomous design is poised to deliver solutions through the use of Evolutionary Robotics, to generate robots and components that are formed though, and tailored to, their environment. Through a number of generations, we iteratively breed promising traits in the robots that we create, and use a combination of simulation and real-world testing to evaluate their fitness for purpose. We are harnessing machine learning to discover unconventional materials, and combining those materials with advanced processing methods, easing the usually laborious construction process and allowing us to create entire robotic lifeforms without human intervention.
Due to their specialisation, autonomously designed robots are likely to perform much better in challenging conditions than generic off-the-shelf solutions. Autonomous design aims to ignite a “Cambrian explosion” of different robotic species, each tailored to their own task and environmental requirements, and designed and constructed autonomously.