Instructing robots to navigate new environments is hard. You possibly can prepare them on bodily, real-world knowledge taken from recordings made by people, however that’s scarce, and costly to gather. Digital simulations are a fast, scalable technique to train them to do new issues, however the robots typically fail once they’re pulled out of digital worlds and requested to do the identical duties in the actual one.
Now, there’s probably a greater possibility: a brand new system that makes use of generative AI fashions together with a physics simulator to develop digital coaching grounds that extra precisely mirror the bodily world. Robots skilled utilizing this methodology labored with a better success price than these skilled utilizing extra conventional methods throughout real-world assessments.
Researchers used the system, referred to as LucidSim, to coach a robotic canine in parkour, getting it to scramble over a field and climb stairs, regardless of by no means seeing any actual world knowledge. The method demonstrates how useful generative AI might be in relation to instructing robots to do difficult duties. It additionally raises the chance that we may in the end prepare them in fully digital worlds. Learn the total story.
—Rhiannon Williams
Africa’s AI researchers are prepared for takeoff
Once we speak in regards to the world race for AI dominance, the dialog typically focuses on tensions between the US and China, and European efforts at regulating the know-how. Nevertheless it’s excessive time we discuss one other participant: Africa.
African AI researchers are forging their very own path, creating instruments that reply the wants of Africans, in their very own languages. Their story shouldn’t be solely certainly one of persistence and innovation, however of preserving cultures and combating to form how AI applied sciences are used on their very own continent. Nevertheless, they face many limitations. Learn the total story.
—Melissa Heikkilä