OpenAI Creates Robot With Perfect Mimicking Capabilities, Imitates After One Demo

OpenAI is the non-profit organization that is chaired by Tesla boss Elon Musk and created by multiple giants in the tech industry. Recently, the group revealed an artificial intelligence that could imitate actions after being shown how it’s done just once. This is a huge jump in the AI technology sector, which brings humanity closer to creating robots that can do anything that humans can.

Details regarding the new breakthrough are on the published paper titled “One-Shot Imitation Learning,” which basically goes through the significance of this development and what led to it. What’s really great about this innovation, however, is the fact that it’s all open to developers who are developing their own AI technologies.

“Imitation learning has been commonly applied to solve different tasks in isolation,” the paper reads. “This usually requires either careful feature engineering, or a significant number of samples. This is far from what we desire: ideally, robots should be able to learn from very few demonstrations of any given task, and instantly generalize to new situations of the same task, without requiring task-specific engineering. In this paper, we propose a meta-learning framework for achieving such capability, which we call one-shot imitation learning.”

As to how it works, it’s best to look at it as a system with two main parts. The first is the vision system, which is what analyzes a ton of images and practices in order to learn object placements in the real world. The second is the imitation system, which takes what it learns from virtual training and applies it to reality by perfectly mimicking human actions, Futurism reports.

For the early tests, OpenAI used block-stacking exercises to see if the AI can mimic not only the actions of stacking blocks but also the order with which those blocks are stacked. Even with some variations, the AI still responded perfectly.