full-stack ceo @innate_bot prev @stanford @theoceancleanup
Palo Alto, CA, USA
- Tweets 971
- Following 878
- Followers 4,604
- Likes 5,096
i have long been convinced that robotics adoption will be faster once everyone can get one at home and teach it intuitively!
today we are releasing the @innate_bot OS in open-source so that you can do just that, in simulator, on our robots, or on yours
Today, we are open-sourcing the Innate OS, the intuitive agentic OS for general-purpose robots
🎮 Try it today in our simulator and make the smartest robot
⬇️ Here is a demo with two robots collaborating
We are entering the physical agents era, but building for mobile robots is still reserved for PhDs. We designed the Innate OS for intuitiveness, so that everyone can start automating their lives.
Innate OS runs on our <$1k open-source robot or on your computer. It takes just a minute to build your first agent in our simulator.
10
11
70
8,076
Axel retweeted
👀
Innate is a team of 7 building capable robots for everyday life.
"The seven of us here do the work of 50," says Vignesh Anand, co-founder of @innate_bot.
We spent time with the team to capture what building means to them.
1
2
8
912
the miracle of silicon valley is that small teams on a mission can make magic, and ultimately win
loved to collab with our friends at @cursor_ai and show how it's like to work at innate
and this does now even show the biggest part of our work this year!
Innate is a team of 7 building capable robots for everyday life.
"The seven of us here do the work of 50," says Vignesh Anand, co-founder of @innate_bot.
We spent time with the team to capture what building means to them.
8
1
69
8,960
You can fine-tune Flux 3 Action from @bfl_ai to do mobile manipulation
Pretty insane speed to a first result in simulation by @DDobas
First results: @bfl_ai Flux 3 Action doing mobile manipulation on @innate_bot MARS 👇
We finetuned the SO101 checkpoint and replaced the action head with our 8DoF (5 arm, 1 gripper, 2 base)
Generated synthetic episodes in innate-os sim, with randomised positions and height of the box
2
6
1,215
Many @innate_bot robots in the wild!
Squeeee. Got Gemma 4 working on my innate Mars with meshtastic. Can trigger the wave function or status updates with Gemma responses from miles away with no internet. Going to do some cleanup and write up how this worked later.
3
589
Axel retweeted
Squeeee. Got Gemma 4 working on my innate Mars with meshtastic. Can trigger the wave function or status updates with Gemma responses from miles away with no internet. Going to do some cleanup and write up how this worked later.
2
8
45
3,937
Axel retweeted
Okay builders, we have some new hardware to play with.
Meet MARS from @innate_bot I have a lot of experiments in mind.
What are we building first?
2
6
32
1,455
Axel retweeted
Okay this is cracked. Put the models in robots, give them something physical to compete over, and suddenly you’re testing a lot more than who gives the nicest text response.
We made GPT-6 Astra and Gemini 3.8 Flash play capture the flag on real robots while roasting each other and...
Gemini won 70% of the time!
Here is a flaming Gemini win, more in the thread
This is surprising given the recent astra hype, but not entirely. Findings below ⬇️
5
33
4,773
At @innate_bot, until GPT-6 came out, we used Gemini the most, since it seemed to perform better than GPT and Claude at spatial understanding, and was cheaper and faster
The Gemini team has included robotics data in their models for a while, with specialized models like the ER series; seems to have paid off!
1
2
288
If you want to run complete mobile manipulation benchmarks like this, this is what our open-source robot MARS was made for!
Grab it today for $995 at innate.bot or play with it straight in simulator at github.com/innate-inc/innate…
5
236
We use 2 low-cost @innate_bot MARS robots for these experiments. They run innate OS, which allows to run physical agents very quickly.
Across 29 runs, robots completed the task 10 times.
7 of these were won by @GeminiApp (including attempt above)
1
8
591
Flash runs faster than astra; this partly explains why it catches the sock faster then runs with it
But this is not the only reason
On several occasions, Gemini seemed to understand better how far the gripper is from the sock. It was also better at quickly retrying and adjusting after a failure
Astra would also fail at picking it up, but continue to the drop point as if it had it in its gripper
1
2
354
We made GPT-6 Astra and Gemini 3.8 Flash play capture the flag on real robots while roasting each other and...
Gemini won 70% of the time!
Here is a flaming Gemini win, more in the thread
This is surprising given the recent astra hype, but not entirely. Findings below ⬇️
5
9
99
14,904
Axel retweeted
Got a robot running?
@innate_bot founder @ax_pey shows his.
I said last week that it is the Apple II of robots. Says it is more like Apple I with their Apple II in labs.
17
11
107
20,649
figure is awesome but this video makes this look bigger than it is
note the language
"we saw success" -> it works sometimes and not frequently enough to say anything else
The holy grail for robotics is being able to generalize: doing work in unseen places
We rented 30 homes in the Bay Area and are doing tasks without any new training
7
1
90
16,062
Axel retweeted
when you make an agentic robot, collaboration just works
one agentic robot leads another to clean up a room
they communicate by voice and the lead organizes
watch until the end :)
funny to me how the lead repeats the follower's name, and how they self-congratulate
it's one of my favorite things with agentic robots, they feel so alive
1
6
640