Optimus Takes out the Trash

Big week this week for us Robot lovers. We have exciting news on Tesla’s Optimus, and Nvidia unleashes some truly groundbreaking robotics work.

Optimus Takes out the Trash

The robot news is bright this week! Tesla’s released a video this week of their humanoid robot, Optimus, doing chores. I’m extremely excited about this video because the robots are not doing backflips, parkour or dancing. While backflips and dances make for very entertaining YouTube content, it doesn’t actually help us in our daily lives.

The video is a montage of Optimus doing different tasks around the home… well, not really a home, it’s a simulated home, but close enough. The robots specifically tackle two items from our chore list (take out the trash and vacuum the floor).

Tesla’s goal with the video was showing that they are training the robot on videos. This is a big deal because we have (or can get) videos of people doing chores easily so that should make training the robots faster. The post on X is below and it’s definitely worth a watch.

These robot demo videos are always awesome, but there’s often reality lurking in the corners. Here’s what we see:

1) The robots are all tethered. It’s unclear if those tethers are just for power or if they require that connection to communicate with an off board computer. This IS a lab setting, but if it’s working in my kitchen, I don’t want the robot to be tethered. I need to walk around my kitchen, dodge kids, clean up mess etc… we’ll keep an eye on the tether in further videos.

2) There’s LOTS of people around the robots watching monitors. I’m assuming I won’t need a legion of machine learning guys to make my robot work at my house, but it appears Optimus still requires a babysitter.

Elon called this the biggest product ever, and we agree. The applications for a humanoid robot that can do general tasks is limitless. While this is very exciting for us, but we still cannot buy one!

Optimus looks incredible. As soon as there’s a sales page, we’ll be first in line to preorder, but until then, we’re waiting.

Nvidia Trains Robots with Dreams

You may be tempted to think I’m taking some creative liberty with that title, but you’d be wrong. Nvidia CEO Jensen Huang made a few big announcements last week at COMPUTEX Taipei. In contrast to Boston Dynamics, who thinks human generalist robots are 10 years away, Nvidia said “The Age of Generalist Robotics Has Arrived”.

Here’s a summary of the Nvidia announcement in plain english.

• New - Isaac GR00T N1.5: This is an updated AI model that helps robots understand tasks and environments more effectively. This model enables robots to perform actions like sorting cereal boxes or adapting to new kitchen layouts with improved efficiency. Basically, it’s a better AI model designed specifically for robotics. Very cool, but this is just a better mousetrap. The next part is really awesome.

• GR00T-Dreams: A tool that creates synthetic motion data, allowing robots to learn new behaviors without extensive real-world training. For example, if we have an plain, flat image of a robot holding an apple. GR00T-Dreams can take that plain image, and generate a video of a robot actually picking up the apple. The video is called a ‘dream’ because it didn’t really happen, but then another robot can use this ‘dream’ to learn how to actually pick up an apple. Pretty trippy, right? This is called ‘synthetic data’ and it’s important because if we have to train the robots to pick up apples, we humans have to sit there and actually pick up apples while the robot learns from our actions. If we use synthetic data (fake videos generated from real world pictures), the training can happen exponentially faster. This is a really big deal, and we’ll probably do another post only on synthetic data. Stay tuned.

• GR00T-Mimic: While dreams uses images to generate synthetic training data, Mimic allows humans to teach the robot how to do something a handful of times. Then Mimic uses the human examples to generate thousands of training models from the few minutes of training from the human.

Huang said humanoid robots will likely be the next multi trillion-dollar industry, which give us a tremendous sense of optimism that the robots will be here soon to help us with the laundry. Until the next time.

The future is bright,
Ferol

Did you enjoy this week’s newsletter? Shoot me a reply and let me know what you think. If you like it, I’d appreciate you sending to any friends you have who might like the newsletter.