Building Synthetic Neurons and Artificial Brains for Robots
69637.4K19
Tags
<500KB2D (developers)GlobalFeature launchPre-launchDemoVertical AI
TLVC Rating
2.80
Hook
Editing / Creativity
Copy
Sentiment of launch
Distribution strategy
Community Rating
No ratings yet
to rate this launch.
Artificial Brains AI is developing compact, biology-inspired neural controllers for robots, targeting robotics teams that want alternatives to large end-to-end learned policies. The company positions its work around human-like brains for robots and real-time learning on the edge , meaning controllers small enough to run on the robot itself rather than through a heavy cloud stack.
This launch highlights a concrete portability result. A roughly 120-neuron primitive, first validated in simulation, is now driving a different physical robot arm using the same underlying architecture, with an opposing extend and flex pair per joint that mirrors how antagonist muscle groups operate in biology. For readers tracking robotics foundation models, the interesting signal is that the same low-neuron-count module transferred to a new body without being rebuilt from scratch, which cuts against the assumption that motor skills must live inside a large monolithic policy trained per embodiment.
The update was shared by Alexander A. Wolf, who is publicly building the project. If the same primitive continues to hold across more arms and morphologies, it becomes a small but meaningful data point for anyone weighing neuromorphic and structured-network approaches against scaled behavior-cloning pipelines in general-purpose robotics.
Comments (14)
Sign in to join the discussion.
N
Nkechi Okonkwo18d ago
120 neurons doing the job of a whole ROS stack is either genius or terrifying. leaning genius but keeping one hand on the e-stop.
M
Mira P.18d ago
ok wait, so you basically wrote motor primitives once and now they hot-swap between bodies? that's the part of the tweet that deserved the hook, not the buried lede.
T
Tomas Lindqvist18d ago
cool result but the arm in the clip looks suspiciously well-lit and the trajectory suspiciously smooth. how many takes?
R
Rahul Iyer18d ago
is there an sdk or do i have to email you and hope? asking because 'reusable neural functions' sounds like a package i want to npm install.
P
Priya Ramanathan18d ago
webhooks when a neuron misfires would go hard, don't @ me.
K
Kenji Watabe18d ago
opposing extend/flex per joint is such a clean abstraction. feels like you rediscovered muscles and i mean that as a compliment.
S
Sofía Carrasco18d ago
the video needed 3 more seconds at the start showing the sim-to-real jump side by side. you buried the whole thesis in b-roll.
A
Aditya Ghosh18d ago
hot take: 120 neurons is not a flex, it's a constraint you'll regret the moment someone asks for bimanual manipulation.
F
Faisal Qureshi18d ago
120 neurons and no mention of latency or jitter. staff eng brain says the demo is doing a lot of quiet averaging under the hood.
L
Lian Zhao18d ago
any paper or preprint on the primitive architecture? curious how you're handling the coupling between antagonistic pairs across joints.
B
Benoît Marchal18d ago
the thumbnail is a robot arm on a white table. every robotics launch this year has been a robot arm on a white table. give us a barn or something.
M
Marcus Adeyemi18d ago
the interesting bit isn't that it transferred, it's that you didn't need to retrain. that's the sentence that should've been the tweet.
Y
Yuki Tanaka18d ago
one more thing: publish a leaderboard of which bodies the same primitive controls. founders love a leaderboard, journalists love a leaderboard, you love a leaderboard.
I
Ivana Reva18d ago
skeptical until i see it pick up a wet noodle. arms on rigid blocks is the mnist of manipulation.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Artificial Brains AI
Building Synthetic Neurons and Artificial Brains for Robots
Building Synthetic Neurons and Artificial Brains for Robots
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Artificial Brains AI is developing compact, biology-inspired neural controllers for robots, targeting robotics teams that want alternatives to large end-to-end learned policies. The company positions its work around human-like brains for robots and real-time learning on the edge , meaning controllers small enough to run on the robot itself rather than through a heavy cloud stack. This launch highlights a concrete portability result. A roughly 120-neuron primitive, first validated in simulation, is now driving a different physical robot arm using the same underlying architecture, with an opposing extend and flex pair per joint that mirrors how antagonist muscle groups operate in biology. For readers tracking robotics foundation models, the interesting signal is that the same low-neuron-count module transferred to a new body without being rebuilt from scratch, which cuts against the assumption that motor skills must live inside a large monolithic policy trained per embodiment. The update was shared by Alexander A. Wolf, who is publicly building the project. If the same primitive continues to hold across more arms and morphologies, it becomes a small but meaningful data point for anyone weighing neuromorphic and structured-network approaches against scaled behavior-cloning pipelines in general-purpose robotics.
Comments (14)
120 neurons doing the job of a whole ROS stack is either genius or terrifying. leaning genius but keeping one hand on the e-stop.
ok wait, so you basically wrote motor primitives once and now they hot-swap between bodies? that's the part of the tweet that deserved the hook, not the buried lede.
cool result but the arm in the clip looks suspiciously well-lit and the trajectory suspiciously smooth. how many takes?
is there an sdk or do i have to email you and hope? asking because 'reusable neural functions' sounds like a package i want to npm install.
webhooks when a neuron misfires would go hard, don't @ me.
opposing extend/flex per joint is such a clean abstraction. feels like you rediscovered muscles and i mean that as a compliment.
the video needed 3 more seconds at the start showing the sim-to-real jump side by side. you buried the whole thesis in b-roll.
hot take: 120 neurons is not a flex, it's a constraint you'll regret the moment someone asks for bimanual manipulation.
120 neurons and no mention of latency or jitter. staff eng brain says the demo is doing a lot of quiet averaging under the hood.
any paper or preprint on the primitive architecture? curious how you're handling the coupling between antagonistic pairs across joints.
the thumbnail is a robot arm on a white table. every robotics launch this year has been a robot arm on a white table. give us a barn or something.
the interesting bit isn't that it transferred, it's that you didn't need to retrain. that's the sentence that should've been the tweet.
one more thing: publish a leaderboard of which bodies the same primitive controls. founders love a leaderboard, journalists love a leaderboard, you love a leaderboard.
skeptical until i see it pick up a wet noodle. arms on rigid blocks is the mnist of manipulation.