Brand new videos on the Tesla and Unitree humanoids perfectly illustrate the different approaches behind these two cutting-edge robots. Optimus is clearly focused on useful work, ASAP – while the Unitree G1 is becoming surprisingly agile and athletic.
It's an interesting distinction, given how Elon Musk has typically pushed Tesla to be more "fun" than other automakers, with video games on the dash console, fart noises for indicators, Spaceballs-inspired names like Ludicrous and Plaid modes for extreme acceleration, and any number of other touches you certainly wouldn't expect from a Mercedes-Benz or BMW.
But despite the endless opportunities a humanoid might offer for such silly fun, it's clear Tesla's Optimus is more business focused at this point – and I say this despite the recent Cybercab/Robovan launch event, in which a squad of Optimus robots did some podium dancing, served drinks, and interacted with the crowd (albeit apparently in a manner driven by humans in telepresence suits).
Take a look at this Optimus autonomy update video, posted just 18 hours ago – and filmed without human teleoperation:
While we were busy making its walk more robust for 10/10, we’ve also been working on additional pieces of autonomy for Optimus!
— Milan Kovac (@_milankovac_) October 17, 2024
The absence of (useful) GPS in most indoor environments makes visual navigation central for humanoids. Using its 2D cameras, Optimus can now navigate… https://t.co/R448099uEQ
The key Optimus updates here:
- Smoother walking – although it still looks like it needs a diaper change
- Autonomous exploration of new areas, using only visual inputs to map the space and avoid obstacles
- Swarm-wide map updates
- It autonomously goes and plugs itself into a charging dock when required
- It can pick up and transport heavier items
- It can now go up and down stairs by itself
- It can now react to humans through voice or gesture communications
All of which are important steps toward getting Optimus out into the workforce at mass scale, doing useful work. But they're also ... well, sort of not that exciting or fun. Especially when Chinese company Unitree is putting out videos like this one, just three hours later:
The G1 humanoid is shorter, smaller and lighter than Optimus. It's got three fingers instead of five, and while Musk says he hopes to get Optimus down to a consumer price around US$20-30,000 per unit once Tesla begins manufacturing at scale, Unitree is selling the G1 humanoid for US$16,000, right now – albeit for "purposes of education and research or industrial applications."
And its physical capabilities as an athlete make Optimus look pretty stodgy at this point. The speed and fluidity with which the G1 walks, jogs, negotiates stairs, rips out dance moves, picks itself up from the floor and shows off karate warmup stretches that'd put me in hospital ... It's super impressive, and a lot of fun to watch.
It looks quick and nimble even in real-time 1X video playback, where most of the Optimus video is sped up to between 2X and 10X so nobody falls asleep.
You can see more of the G1's dance moves, as well as how well it handles being pushed and tripped, in the video below, from a month ago.
It's weird to see what looks like a direct competitor having more fun than Tesla – but then, perhaps it's inaccurate to see these bots as direct competitors. While the G1 is the "mass production version" for Unitree, it's a production version that's still focused on customers that want to do R&D. Tesla, on the other hand, wants Optimus working in its own auto production plants yesterday.
So it could well be a matter of priorities; G1 the research bot is looking fun and athletic, not unlike the extraordinary Atlas robot from Boston Dynamics, because that's where the focus has been.
The Optimus team, on the other hand, is postponing these sorts of shenanigans 'til later, and focusing much more tightly on the skills this bot will need to start paying its way. The simplest of autonomous factory floor tasks: sorting things into boxes, picking them up from here and putting them down in the right spot over there. Operating safely in a dynamic environment full of humans. Boring but important.
Either way, both teams are doing remarkable work, and it's fascinating to watch this space developing in real time. Progress is slower here than it has been with language model AIs, but it might not be long before general-purpose robots start proving their worth and taking over the kinds of boring, repetitive, dangerous labor jobs that put a lot of stress on the human body.
The disruptive potential in this space is nearly unlimited, so while there's a long way yet to go, we've got our eyes glued to the feeds.
