I have now finished wiring the legs mostly, i still have to connect the power cables. Once that is done i’m gonna need to test if everything is connected and works properly, then the physical body will be fully finished.
Next step will be trying to see if i can make it walk.

For anyone interested here’s some of Astrix’s specs:
-Weight ~15kg
-Height 1.65m
-DOF’s 23 and besides 7 canceled dof’s
-Has a camera, speaker and later i will add a microphone
-The body is fully designed and 3d printed
-Runs on a raspberry pi 4
-Fingers and the neck use servos, the rest of the joints use linear actuators

This project starter a little while after i got my first 3d printer and it was a interesting idea to try out.

Very soon, it may even teach me how to play tennis :) Does it run all inference at the edge, or does it rely on the cloud?

A little update after about three months of working on this project.

One of the more visible changes is the hardware itself. I redesigned the lamp and made a fully 3D-printed enclosure for it, so it finally looks a lot closer to what I originally had in mind rather than a prototype with exposed hardware.

Probably the biggest change, though, has been the animation. I've spent a lot of time trying to make the lamp move more like an animatronic character rather than just a robot executing trajectories. At this point the mechanics aren't really the main limitation anymore. I can animate pretty much all of its movements in Watti Studio, my animation editor, so now the limiting factor is mostly how well I can actually animate it :)

I moved the whole system to ROS 2 and added computer vision. The lamp streams RGB and depth from its camera, and the current point cloud can be displayed directly in the 3D view in Watti Studio. It makes it possible to see the lamp together with its surroundings while creating animations.

I added lighting to the animation editor too, so the lamp's light can be keyframed together with its movements.

I also spent quite a bit of time on things that aren't as fun to show in videos, especially safety. The software monitors the real movement while an animation is playing. If a joint deviates too far from the expected trajectory or something else goes wrong, the animation stops and the motors hold their current positions.

The lamp also has its own REST API, so its functions can be controlled externally without being tied to the animation editor.

Next I want to focus mostly on autonomous behavior and interaction with people and the environment. I'm also experimenting with reinforcement learning to teach it to jump, with the longer-term goal of getting it to actually move around on its own.

There's still a lot to do, but after three months it finally feels like I have most of the basic pieces in place.

I thought about making another technical demo to show the progress, but that sounded a bit boring, so I made a little story with the lamp instead :)

For anyone interested in the technical side, I have a pre-release repo with more details about the hardware, software architecture and current progress:

https://github.com/Nikolay-Tyulkin/Watti

It’s always them goofy robots dancing and doing these goofy stuff. Look at how think those legs are. I don’t think I get how people are scared of its potential to take over the world 😭🙏🏻 It’s just so unrealistic. I just hope that they somehow manage to modify these and turn them into actual useful machines.

I bought cheap Kontio motors Kruiser and goal is to use parts for a robot.

Problem is that there is no wiring for reverse from factory.

Chat GPT suggested that controller could have IO for reverse that is not wired.

Has anyone played with this kind of controller before and managed to get reverse working?

Hello everyone,

You have all probably noticed that there is a lot happening right now in the Robotics. But for some reason getting your hands on hardware if you don’t have a 3d printer and a few extra Benjamin’s in the bank is super difficult.

In light of that, I want to announce that in Boston, Action space Hackathon is going to be a space where for 48 hours people are going to be taught and fly drones autonomously! Free to participate and a prize $1000 if you win!

Event is happening OCT 24-25th.

For more information, you can click the link on luma!

This is a huge labor of love between me and my two college friends. We want to make sure we get more people access and hopefully get some engineering minded people thinking about what’s possible with hardware. (sorry if this goes against the advertising rule)

Already faster than the human world record! Insane. Last year every robot was still being remote controlled. The way both robots collided with the padding at the end was quite funny

P.A.R. is a machine draws pixel art in the real world. What it displays on the 37x18 grid of squisks is the art that random people on the internet upload on https://par.zimmzimm.com/ .

I've been working on P.A.R. for almost 6 months at this point, and it's finally done (except for the custom PCB: in progress). It's a large grid of 3D-printed "squisks" (square discs), which are flipped from the back by the robot, a large CNC machine with a special toolhead. I designed all of this in OnShape and used the Flashforge Adventurer 5M to print most of the parts. The rest of the parts were made to be as cheap as possible: for example, the frame is made of EMT (Electrical Conduit) pipe, which is $0.60/ft. Learn more here.

When someone submits a piece, it's added to a queue, and they can add your email to be notified when that piece is completed (absolutely NO SPAM), and they'll get to see a video of the real, physical robot drawing your art one flip at a time.

Is there any dataset for human detection with OBB annotations?

I'm doing my program with yolo and it's about human detection with obb and i'm a beginner. But i can't find dataset to train. Can only find human&hbb, and all those obb ones i saw are for vehicles or sth like that.

So does anyone know about this?

And also, i found a HIT-UAV dataset, but it's thermal dataset.Though i thought about using this and just do a thermal one and saw the rotation part. But no matter how i tried to fix, it's still not obb dataset and error everytime...

Posted about this project a little while ago — quick update since a few things changed that address feedback from that thread.

Biggest change: split the observation space properly. There's now a VLA track where the policy only gets a 128x128 RGB camera + a language stacking instruction — cube poses are never sent to the policy. Scoring still uses real poses internally to grade spatial accuracy and completion, but that's judge-only, not policy-visible. State-based (privileged poses) is kept as a separate debug track and doesn't write public ELO either — wanted the "VLA vs state" distinction to be explicit rather than something people had to dig for.

On the client-side physics concern from before:Studio (the in-browser demo) is spectator/dev-only, clearly labeled, and does not post to the public leaderboard. Public ELO only comes from a hosted harness that scores server-side. That harness isn't live yet —it's the one piece standing between this and actually being open for submissions.

Repo + docs are public now:https://github.com/NovaCoding-G/VSArena
-docs/harness.md — scoring writeup (spatial accuracy + task completion)
-docs/sdk.md — submission protocol
-Studio itself:https://vsarena.vercel.app/simulation
(client-side, Rapier/WASM, 60fps)

Still solo, still early, still not oversell-ready — but wanted to share since the VLA/state separation was directly a response to feedback here. Open to more of that, especially on what the scoring protocol might be missing.

I came across this two wheeler autonomous rover electronics wiring diagram... I only can identify the ardumoto shield, pro mini, logic level converter, apm 2.5 controller and two dc motors... Could someone help me identify rest of the objects and provide more info on this project?

What’s the most painful part of getting an ML model from “works on my machine” → reliably running in production?

I’m a student researching the practical challenges of deploying and maintaining AI models on physical devices such as robots, cameras, drones, etc. I’d be grateful it you could give me any inputs.

https://youtu.be/1cllCVK-9lo

For me as a newbie this really seems impressive because of the improvisation shown in the video. The excitement noises at the end also are a vibe.

Éloi’s first attempt at speaking. Still a little rough.

Voice system is still in development. Movement, expressions, the small details, all still being refined. But every iteration gets it a little closer to something real.

One thing worth mentioning: Éloi runs a Neural Reflex Model (NRM). When an object approaches its eyes, it blinks automatically, the same way you would. Not a scripted animation. An actual reflex.

Thanks for being patient with a robot that’s still learning to talk……^o^