Tue. May 21st, 2024



Quick Links

While everyone is fixated on generative AI and its implications for knowledge and creative work, robotics have been quietly approaching a similar apparently sudden point of advancement. The year 2024 may well go down in history as the start of the true AI automation revolution for physical labor, and even this early in the year there have been some jaw-dropping demonstrations.



Figure 01

Figure is a nascent robotics startup that has attracted a massive amount of investment from major technology players. While the company has been fairly reserved with demos showing its humanoid robot project, directly designed to supplement human labor shortfalls in the years ahead, each one has been jaw-dropping. The latest demo as of this writing feels like a historical turning point.


In it, we see the Figure 01 robot at a table with objects on it. A human then converses with the robot in natural language, and it responds in a perfectly human voice, relating what it sees and performing tasks per request. This is thanks to a combination of Figure’s technology and software along with a partnership with OpenAI that has improved the robot’s reasoning skills and machine vision abilities. Most impressive is that it can figure out how to perform tasks on the fly.

According to Figure, this is not an edited or cut video, which feels likely to be true thanks to the long processing pauses as the robot “thinks” about what to do or say next, but as always, we only have their word for this until these systems make it into third party hands. Figure is coming in hot with these robots though, with trials already happening in partnership with BMW.


Agility Robotics & Amazon’s Digit

Agility Robotics in partnership with Amazon seems to have the same general idea as Figure when it comes to their humanoid robot. The idea is to create a robot that can slot into the same spaces that human workers currently occupy. Digit is already undergoing testing in real Amazon facilities, as we’ve been shown in this video.

However, perhaps even more impressive was this earlier demo of Digit showing its reasoning ability. What really makes this an amazing demo is that we can see the bot’s internal monologue and thought processes as it dynamically reasons its way through the task it’s been given verbally.


Boston Dynamics Spot V.4 Update

You can’t have a list of robot demo videos without Boston Dynamics on the list, except this time it’s sadly not an adorable dancing robot video, but perhaps something far more impressive. Boston Dynamics shows off how they’ve used virtual simulations and a form of machine learning known as reinforcement to give their Spot robot a serious upgrade when it comes to mobility. Before, Spot relied entirely on a predictive model fine-tuned by Boston Dynamics to predict how it should move when traversing terrain, but now with the latest update it uses a hybrid of that model and a model trained using reinforcement.


The end result is where the old Spot software version would trip and fall on loose or slippery surfaces, the latest version can keep it together even when grip is nonexistent or the terrain is unpredictable. It might not be the flashiest thing anyone is showing off, but it’s a big deal when you need robots to go into places that don’t offer a perfectly predictable environment, such as disaster sites or inclement weather.

Tesla’s Optimus Gen 2

In 2021, I wrote that Tesla’s proposed robot made total sense, and since then we’ve slowly seen more and more of how far this project has progressed. What started out as a human dancer in a suit is now a fully-realized humanoid robot platform. In the Gen 2 demo, we see “Optimus” showing smooth human-like motion, walking at a reasonable (if sedate) pace, and doing the classic egg-manipulation trick with the help of fingertip sensors. A common trope in videos of robot arms for decades now.


The video is quite impressive, but it’s not clear what’s autonomous, what’s pre-programmed, and what’s direct tele-operation. Still, to go from having nothing to show to a working hardware platform in such a short time is worth applauding. In another clip posted by Elon Musk on X (formerly Twitter) we can see Optimus folding clothes, however Musk then clarified that this isn’t being done autonomously yet. Getting the hopes of lazy people everywhere, just to knock them down again,

Stanford’s Mobile ALOHA

The robots we’ve seen so far look like they’re right from a 2000s sci-fi movie, but they also come with high price tags running into hundreds of thousands of dollars.


Mobile ALOHA, which is a project collaborated on by Google and Stanford, provides a versatile robotic platform for under $20,000. That sounds like a lot, but if you watch the demo of what ALOHA can do, it seems like a bargain. A tele-operator can easily do complex dextrous tasks with this bare-bones robot, and it can learn to do tasks autonomously by watching demonstrations by humans. Best of all, the hardware and software for ALOHA are open source, so I expect plenty of advancement ahead for it as numerous contributors from around the world work on making it better.


The year has hardly got a good start, and already it looks like we’re in for even more mind-blowing automation demos before the next holiday season rolls around. Suddenly, that robot vacuum doesn’t feel so futuristic, does it?





Source link

By John P.

Leave a Reply

Your email address will not be published. Required fields are marked *