What is the way to go in hobby robotics today? I'm more interested in high level, and want the lower level to 'just work' with minimum efforts from my side. Having mechanical part and vision what would be the right choice for low-middle software to control robotic arm and car, may be attached one to another. ROS2?
HuggingFace LeRobot. You can build the reference arm easily and cheaply and the software is designed to train AI. There's a lot to explore and extend there and the community is growing rapidly. It's based on the Stanford Aloha project.
https://huggingface.co/lerobot
Probably not much.
Also the LeRobot reference arm (SO-ARM100) is 6DoF, but it's very hackable and there are already project with different grippers, etc.
Thanks, but no. It's going to be robotic arm with gripper and camera. The rest can be either Raspberry Zero (for cam, control, with net or blutooth, something big for high level). Another option, not exclusive, NVidia Jetson Nano instead of Zero. It could be Raspberry Pi 3, but I don't want to do video processing on it. All this I have, just need to put them together. Adding AGX Orin will be a big thing. That's actually the goal. With video processing and LLM all in one mobile robot. As it's hobby R/D it will be configured and reconfigured many times. That's why I don't want to do low level by hand every time.
Another interesting option is Raspberry Pico * N + Tiny PC. For control and thinking. They can be connected via wifi or blutooth.
3D reconstruction, from stereo or mono camera, have both. Object detection, text reading. Ideally it should recognize the speaker, simple gestures. Take audio, feed speech to LLM, get the output. Being able to detect and move out of the way of walking humans. Most of it has been done, like 3d structure and localization from motion. There are reference implementations. I've done it before too.
It's sort of open ended project. Having LLM with vision on mobile robot with arm.. has a lot of applications. AGX Orin 64GB is capable of running serious models.
I have long assumed that we won’t be getting robot butlers partly because it’s really really hard, but also because most of not all things we want robots for it’s easier to reconfigure the environment than make a flexible humaniod
So factories are obvious but the real mass uptake is the home - and honestly I think something that cleans and tidies an hour a day might actually be achievable
Wait till they get access to the "MCP" for making tiktok content - "I'm a tradbot, and I like it!" / "You won't believe what this roomba did to my cleaning routine!" ...
The "Berkeley Humanoid" is a distinct robot (they have the "Berkeley Humanoid Lite" named "ours" and colored in orange as the rightmost point on their graph).
the cost-effectiveness/performance factor benchmark is interesting, but it feels slightly misleading - I just don't see how "average peak torque of all actuated DoFs, normalized by the robot's size" is related to measuring "accessibility and customizability" of the robot.
What is interesting is that on their own metric, the Berkley Humanoid is only twice as expensive as the Berkley Humanoid Lite but has more than twice the "performance factor" (0.36 vs 0.14).
It shows they threw away too much while creating the lite version.
Depends on the relative market size for performance factor though. If 90 percent of the market is captured by a 0.14 performance factor then that extra in price could be put towards solving another problem.
Rather, I think we can say based on those datapoints that for their design, performance scales superlinearly with cost. Not surprising given fixed costs!
As much as I like the concept, 3D printing everything is not the way to lower cost.
Mass-produced (stamped / extruded / whatever) mechanical parts + hackable 'brains' is.
Robots do lend themselves well w/ respect to that last part. Worst case is rip out its control electronics wholesale & replace with your own motor drivers etc.
Hm, perhaps not - but maybe give the users an option to print such parts, and warn that they may affect longevity of said parts if they do decide to go full manufacturing route.
My potential concern is the "Apple" gatekeeping of parts.
It depends what you're doing. High volume parts, absolutely. It's one of the things that bugs me about the "3D printers printing printers" type projects. 3D printing is terrible for mass producing parts. If you're making 1000+ of something, injection mold it.
Low volume, probably customized parts like R&D robotics tends to need? 3D printing is great, especially if the design files are available so you can modify the parts as required before printing. And then if you break something you can print another one off overnight instead of stalling your project for weeks waiting for new parts to arrive.
As weird as it sounds, but carbon fiber is the most accessible material for making a DIY robot. Anything that uses metal requires expensive machinery. Carbon fiber is labor intensive (i.e. bad for mass manufacturing), but doesn't need much equipment beyond a curing oven for the epoxy (around $2000) for state of the art results.
The comment criticizes a chart or metric comparing "performance factor" to torque and degrees of freedom (DOFs) in robotics, calling it "the most silly thing" the commenter, a licensed mechanical engineer, has seen. By referencing "Kony 2012"—a widely mocked internet campaign—they emphasize their point about the chart's perceived absurdity. ([The performance factor vs. torque vs. DOFs is the most silly thing as ...](https://news.ycombinator.com/item?id=43801052&utm_source=cha...))
The critique likely stems from the idea that combining performance factor, torque, and DOFs into a single comparison oversimplifies complex engineering concepts. Torque and DOFs are distinct mechanical properties, and "performance factor" is a vague term without a clear definition. Such a chart might misleadingly suggest direct correlations where none exist, leading to confusion or misinterpretation.
In essence, the commenter is expressing frustration over what they see as a technically flawed and potentially misleading representation of robotic performance metrics.
What is the way to go in hobby robotics today? I'm more interested in high level, and want the lower level to 'just work' with minimum efforts from my side. Having mechanical part and vision what would be the right choice for low-middle software to control robotic arm and car, may be attached one to another. ROS2?
HuggingFace LeRobot. You can build the reference arm easily and cheaply and the software is designed to train AI. There's a lot to explore and extend there and the community is growing rapidly. It's based on the Stanford Aloha project. https://huggingface.co/lerobot
LeRobot is 6DoF.
How much does this matter in practice vs 7DoF arm?
Probably not much. Also the LeRobot reference arm (SO-ARM100) is 6DoF, but it's very hackable and there are already project with different grippers, etc.
I’d say start with a sumo robot, line follower, or maze solver. That’ll keep you very entertained for a good while.
Thanks, but no. It's going to be robotic arm with gripper and camera. The rest can be either Raspberry Zero (for cam, control, with net or blutooth, something big for high level). Another option, not exclusive, NVidia Jetson Nano instead of Zero. It could be Raspberry Pi 3, but I don't want to do video processing on it. All this I have, just need to put them together. Adding AGX Orin will be a big thing. That's actually the goal. With video processing and LLM all in one mobile robot. As it's hobby R/D it will be configured and reconfigured many times. That's why I don't want to do low level by hand every time.
Another interesting option is Raspberry Pico * N + Tiny PC. For control and thinking. They can be connected via wifi or blutooth.
What kind of vision peocessing are you envisioning?
3D reconstruction, from stereo or mono camera, have both. Object detection, text reading. Ideally it should recognize the speaker, simple gestures. Take audio, feed speech to LLM, get the output. Being able to detect and move out of the way of walking humans. Most of it has been done, like 3d structure and localization from motion. There are reference implementations. I've done it before too.
It's sort of open ended project. Having LLM with vision on mobile robot with arm.. has a lot of applications. AGX Orin 64GB is capable of running serious models.
Micro mouse is a good way to get into robotic.
I think this is a great idea. It seems like we are entering the phase where the core hardware problems are solved and we now need to:
A) bring down cost and expand the design space for the hardware and
B) minimise the barriers to working on the "software" problems where there still seem to be huge areas of mostly unaddressed challenges.
An open source platform seems like a good thing for both.
I have long assumed that we won’t be getting robot butlers partly because it’s really really hard, but also because most of not all things we want robots for it’s easier to reconfigure the environment than make a flexible humaniod
So factories are obvious but the real mass uptake is the home - and honestly I think something that cleans and tidies an hour a day might actually be achievable
How does that work for things like taking out the trash, doing dishes, and folding laundry?
Robots that take over the old fashioned traditional housewife job will cause an outrage.
Wait till they get access to the "MCP" for making tiktok content - "I'm a tradbot, and I like it!" / "You won't believe what this roomba did to my cleaning routine!" ...
https://lite.berkeley-humanoid.org/static/comparision.png
why does it say the Berkeley Humanoid is closed source here? Is it a typo, was this paper peer-reviewed?
The "Berkeley Humanoid" is a distinct robot (they have the "Berkeley Humanoid Lite" named "ours" and colored in orange as the rightmost point on their graph).
!!! hunter eye. thanks for clarification!
the cost-effectiveness/performance factor benchmark is interesting, but it feels slightly misleading - I just don't see how "average peak torque of all actuated DoFs, normalized by the robot's size" is related to measuring "accessibility and customizability" of the robot.
What is interesting is that on their own metric, the Berkley Humanoid is only twice as expensive as the Berkley Humanoid Lite but has more than twice the "performance factor" (0.36 vs 0.14).
It shows they threw away too much while creating the lite version.
Depends on the relative market size for performance factor though. If 90 percent of the market is captured by a 0.14 performance factor then that extra in price could be put towards solving another problem.
Rather, I think we can say based on those datapoints that for their design, performance scales superlinearly with cost. Not surprising given fixed costs!
As much as I like the concept, 3D printing everything is not the way to lower cost.
Mass-produced (stamped / extruded / whatever) mechanical parts + hackable 'brains' is.
Robots do lend themselves well w/ respect to that last part. Worst case is rip out its control electronics wholesale & replace with your own motor drivers etc.
Hm, perhaps not - but maybe give the users an option to print such parts, and warn that they may affect longevity of said parts if they do decide to go full manufacturing route.
My potential concern is the "Apple" gatekeeping of parts.
It depends what you're doing. High volume parts, absolutely. It's one of the things that bugs me about the "3D printers printing printers" type projects. 3D printing is terrible for mass producing parts. If you're making 1000+ of something, injection mold it.
Low volume, probably customized parts like R&D robotics tends to need? 3D printing is great, especially if the design files are available so you can modify the parts as required before printing. And then if you break something you can print another one off overnight instead of stalling your project for weeks waiting for new parts to arrive.
A 3D printed robot that costs $5000 exerts pressure on the price of mass-produced competitors.
As weird as it sounds, but carbon fiber is the most accessible material for making a DIY robot. Anything that uses metal requires expensive machinery. Carbon fiber is labor intensive (i.e. bad for mass manufacturing), but doesn't need much equipment beyond a curing oven for the epoxy (around $2000) for state of the art results.
Robot gets Piercing in Berkeley:
https://youtu.be/0Gkl1H2eKsM?t=99
Servitude: Robot Waiter:
https://www.youtube.com/watch?v=NXsUetUzXlg
Empathy: Broken Robot:
https://www.youtube.com/watch?v=KXrbqXPnHvE
I was hoping "Lite" would be a smaller humanoid that I could build for <$5K, but this looks expensive.
The performance factor vs. torque vs. DOFs is the most silly thing as a licensed mechanical engineer I have ever seen. And I was around for Kony 2012.
Can you explain why to the layman?
https://chatgpt.com/share/680cb5ae-10d8-8007-a580-b7c3266138...
The comment criticizes a chart or metric comparing "performance factor" to torque and degrees of freedom (DOFs) in robotics, calling it "the most silly thing" the commenter, a licensed mechanical engineer, has seen. By referencing "Kony 2012"—a widely mocked internet campaign—they emphasize their point about the chart's perceived absurdity. ([The performance factor vs. torque vs. DOFs is the most silly thing as ...](https://news.ycombinator.com/item?id=43801052&utm_source=cha...))
The critique likely stems from the idea that combining performance factor, torque, and DOFs into a single comparison oversimplifies complex engineering concepts. Torque and DOFs are distinct mechanical properties, and "performance factor" is a vague term without a clear definition. Such a chart might misleadingly suggest direct correlations where none exist, leading to confusion or misinterpretation.
In essence, the commenter is expressing frustration over what they see as a technically flawed and potentially misleading representation of robotic performance metrics.
Why do you think this excessively verbose bit of LLM vomit contributes to the conversation?
Very cool! Open source robotics is something I always imagined to be a part of the future. Hope the idea catches on.
A left handed robot!
been cool watching robots go open source like this, always gets me thinking how much i could hack together something dumb just to see if it works
[flagged]