knox byte

building ML frameworks for Unitree Robotics

HkFLX8ASQwBEM4d764nN1QDeZtb7HGjp4oj7mKV4pump

[knox byte @knoxsbyte]

Can everyone see the stream output correctly?

I want to make sure everything's in place

[knox byte @knoxsbyte]

I am live

https://t.co/A9lLj5e5qk

[knox byte @knoxsbyte]

I am live

https://t.co/A9lLj5e5qk

[knox byte @knoxsbyte]

While I'm setting up the stream I wanted to recap what ARGOS for anyone new here and create a pinned tweet for the community to see.

ARGOS is a multi-robot swarm intelligence platform that coordinates fleets of Unitree humanoid and quadruped robots to autonomously clean, organize, and maintain physical spaces.

What makes it so exciting:

- Heterogeneous Swarm Intelligence: G1 humanoids (29 DOF, bimanual arms) and Go2 quadrupeds (12 DOF, 3.5 m/s trot) work side-by-side in the same coordinated fleet, each assigned tasks that match their physical capabilities

- LLM-Powered Task Planning: Claude decomposes natural language goals like "clean the kitchen" into optimized task graphs with dependencies, timing estimates, and robot assignments

- Auction-Based Task Allocation: robots bid on tasks using real-time cost functions that factor in distance, battery level, and current workload.

- Vision-Language Policy Execution: state-of-the-art ML policies (VLA, Diffusion Policy, ACT) translate camera frames directly into 29 or 12-DOF joint commands at up to 30 Hz

Full Training Pipeline: ingest demonstration videos, label actions with pose estimation, build LeRobot-format datasets, LoRA fine-tune on custom hardware, and deploy to robots in minutes

Live Fleet CLI & TUI: a Rich-based terminal interface gives operators real-time battery telemetry, zone coverage maps, task queues, and swarm logs in a beautiful hacker-aesthetic dashboard

Fault-Tolerant Rebalancing: if a robot fails mid-task, the allocator instantly reassigns its workload to remaining robots without any operator input

The impact: ARGOS is being integrated with Unitree Robotics framework suite to be used mainstream and can (potentially) be shipped out worldwide to customers for use within the next several months after conducting thorough testing.

The video below is the first demo of ARGOS in action at Unitree's Studio and with their G1 humanoid robot.

[knox byte @knoxsbyte]

Sorry for the delay on the livestream everyone.

It seems like I'm not able to stream directly through my PumpFun wallet, so I'm getting a stream key from the creator of the coin.

Setting this up shortly

[knox byte @knoxsbyte]

👀

[knox byte @knoxsbyte]

I've endorsed the token on my profile and I've heard some suggestions about getting even a temporary livestream going so I will set that up.

[knox byte @knoxsbyte]

As promised I've rolled out my biggest change yet.

1000+ lines of code 😄

Good start to the day, I will attach details below

https://t.co/PlX8KHfsSS

[knox byte @knoxsbyte]

I have a big update rolling out shortly

Very big update for everyone considering it will provide support for lower budget humanoid robots.

I am excited to hopefully open the doors to new researchers and engineers!

[knox byte @knoxsbyte]

My manager also gave me the go ahead on streaming, but needs to clear a few things with legal and HR.

I'm also required to meet certain video quality standards, so I will be using my fees to buy streaming gear that's been chosen by them.

[knox byte @knoxsbyte]

Hello everyone, just wanted to update everyone with my agenda for today:

- Adding support for Unitree Go2 robots

- Rolling out some reworks for the training pipeline to increase efficiency and cleaning accuracy

- Redesigning the CLI

- Few bug fixes

[knox byte @knoxsbyte]

Hello everyone, just wanted to update everyone with my agenda for today:

- Adding support for Unitree Go2 robots

- Rolling out some reworks for the training pipeline to increase efficiency and cleaning accuracy

- Redesigning the CLI

- Few bug fixes

[knox byte @knoxsbyte]

Recording and releasing my first demo of ARGOS was already a long and pain staking process with Unitree's legal team so please bare with me.

I'm trying my hardest here and I will deliver for you.

[knox byte @knoxsbyte]

@notHidro I will set everything up to make that possible tomorrow

I'm working with Unitree's studio and G1 robots so it's not as simple as live streaming all of it. Lot of legality issues to consider and deal with first.

[knox byte @knoxsbyte]

Several new commands available and made as easy to use as possible!

[knox byte @knoxsbyte]

About 400 lines of new code written 😎

Respect and thank you to everyone's support today, I will continue working hard tomorrow.

https://t.co/qoE7fJznPn

[knox byte @knoxsbyte]

flipped, recoloured, sped up, with slight noise added to the robot's movement data. A dataset of 10 episodes becomes 30 or 40 with one command. (2/2)

[knox byte @knoxsbyte]

I've just pushed out my last update for tonight!

The training pipeline can now multiply your dataset without recording any new footage.

This means: you give it your existing demonstration videos, tell it how many copies you want, and it generates transformed variations (1/2)

[knox byte @knoxsbyte]

I've claimed the pump fun funds

The contract address for the token is:

HkFLX8ASQwBEM4d764nN1QDeZtb7HGjp4oj7mKV4pump

I will continue putting out updates and building ARGOS in public for all of you. Thank you for your support.

[knox byte @knoxsbyte]

app and I am in the process of getting things together to do so.

[knox byte @knoxsbyte]

I want to ensure that I'm transparent about the whole building and testing process, similar to how I would have to share testing results with researchers, I would be doing the same with all of you.

I have spoken with @Kimbazxz about setting up a live stream in the PumpFun (1/2)

[knox byte @knoxsbyte]

- GPU/Computation costs for training ~$400/mo
- Cloud Hosting ~ $50/mo
- Unitree Go2 $1600 x 3 ~ $4800 (one time)

I've already acquired several small research stipends to support half of these costs for testing but this token could potentially help me cover the remaining amount

[knox byte @knoxsbyte]

ARGOS started as a side project but has grown into something much greater than that over the past 1.5 months.

Some people don't know this already, but I also have a full time position as a Deep Learning Researcher at Unitree Robotics - What this means: (1/4)

[knox byte @knoxsbyte]

about releasing this under Unitree's open-source framework suite. This opens up a lot of doors to make this dramatically more mainstream, but it will require more validation testing and that means more funding to do so.

I require funding for the following things: (3/4)