Antoine Cully, Jeff Clune, Danesh Tarapore, Jean-Baptiste Mouret, Robots that may adapt like animals, Nature, 2015.
Summary: Robots have remodeled many industries, most notably manufacturing, and have the ability to ship large advantages to society, reminiscent of in search and rescue, catastrophe response, well being care and transportation. They’re additionally invaluable instruments for scientific exploration in environments inaccessible to people, from distant planets to deep oceans. A significant impediment to their widespread adoption in additional complicated environments outdoors factories is their fragility. Whereas animals can rapidly adapt to accidents, present robots can not ‘assume outdoors the field’ to discover a compensatory behaviour when they’re broken: they’re restricted to their pre-specified self-sensing talents, can diagnose solely anticipated failure modes and require a pre-programmed contingency plan for each kind of potential harm, an impracticality for complicated robots. A promising method to lowering robotic fragility includes having robots study acceptable behaviours in response to break, 11, however present strategies are sluggish even with small, constrained search areas. Right here we introduce an clever trial-and-error algorithm that enables robots to adapt to break in lower than two minutes in massive search areas with out requiring self-diagnosis or pre-specified contingency plans. Earlier than the robotic is deployed, it makes use of a novel approach to create an in depth map of the area of high-performing behaviours. This map represents the robotic’s prior information about what behaviours it could possibly carry out and their worth. When the robotic is broken, it makes use of this prior information to information a trial-and-error studying algorithm that conducts clever experiments to quickly uncover a behaviour that compensates for the harm. Experiments reveal profitable diversifications for a legged robotic injured in 5 other ways, together with broken, damaged, and lacking legs, and for a robotic arm with joints damaged in 14 other ways. This new algorithm will allow extra strong, efficient, autonomous robots, and will make clear the ideas that animals use to adapt to harm.
Article reference:
发布者：Dr.Durant，转转请注明出处：https://robotalks.cn/robots-that-can-adapt-like-animals/
