🔁Mario Nawfal:
🚨🇺🇸 AI ROBOTS CAN JUMP 41% HIGHER - AND STICK THE LANDING LIKE A GYMNAST
MIT’s CSAIL researchers just proved AI isn’t only good for writing essays or making cat memes - it’s also designing robots that can jump higher than human engineers ever managed.
Their latest creation leaps about 2 feet into the air, which is 41% higher than the version humans built on their own - all thanks to a generative AI model that thought outside the box (and apparently outside the laws of normal geometry).
Instead of plain old straight parts, A.I. redesigned the robot’s linkages into curved, drumstick-looking blobs that store more energy for the jump without snapping like twigs, solving the classic “jump higher without breaking” problem humans kept failing to crack.
MIT’s team didn’t just settle for high jumps, though: they used A.I. to optimize the robot’s feet too, slashing crash-landings by 84%.
To get there, they ran 5 rounds of simulations, sampling 500 design ideas and steadily tweaking an “embedding vector” (basically a digital vibe-check) until the robot was a high-jumping champ.
Researchers say this A.I. creativity could lead to robots that do way more complex stuff - like picking up trash or handling power tools - and hint that future designs might bounce even higher if they swap plastic for lighter materials.
So yeah: robots are learning to jump higher and fall down less often than most humans… and AI might be coming for the Olympic gold next.
Source: MIT News
https://x.com/MarioNawfal/... https://x.com/MarioNawfal/...
🚨🇺🇸 AI ROBOTS CAN JUMP 41% HIGHER - AND STICK THE LANDING LIKE A GYMNAST
MIT’s CSAIL researchers just proved AI isn’t only good for writing essays or making cat memes - it’s also designing robots that can jump higher than human engineers ever managed.
Their latest creation leaps about 2 feet into the air, which is 41% higher than the version humans built on their own - all thanks to a generative AI model that thought outside the box (and apparently outside the laws of normal geometry).
Instead of plain old straight parts, A.I. redesigned the robot’s linkages into curved, drumstick-looking blobs that store more energy for the jump without snapping like twigs, solving the classic “jump higher without breaking” problem humans kept failing to crack.
MIT’s team didn’t just settle for high jumps, though: they used A.I. to optimize the robot’s feet too, slashing crash-landings by 84%.
To get there, they ran 5 rounds of simulations, sampling 500 design ideas and steadily tweaking an “embedding vector” (basically a digital vibe-check) until the robot was a high-jumping champ.
Researchers say this A.I. creativity could lead to robots that do way more complex stuff - like picking up trash or handling power tools - and hint that future designs might bounce even higher if they swap plastic for lighter materials.
So yeah: robots are learning to jump higher and fall down less often than most humans… and AI might be coming for the Olympic gold next.
Source: MIT News
https://x.com/MarioNawfal/... https://x.com/MarioNawfal/...
3 days ago
