"Robots excel at precise, repetitive, high-speed movements, but they struggle to pick up simple items and manipulate them. Even doing our laundry is extremely difficult."
"Dex-Net didn't have any preexisting data representing robot actions to train on, so the team created a video game containing a vast database of three-dimensional shapes for their virtual robot to pick up."
"The more items it picked up, the higher the robot's score. Across millions of iterations, Dex-Net learned to pick up objects it had never seen before by recognizing features like handles on mugs."
Robots excel in precise, repetitive tasks but struggle with simple manipulations like picking up items. Ambi Robotics aims to bridge this dexterity gap. Founded by UC Berkeley graduates, the company originated from research in AI and robotics. Their project, Dex-Net, uses a video game to train robots on object manipulation, allowing them to learn from millions of iterations. This approach helps robots recognize and pick up unfamiliar objects, addressing a significant challenge in robotics.
Read at www.berkeleyside.org
Unable to calculate read time
Collection
[
|
...
]