Comment by amacneil
5 days ago
Robotics is definitely getting more accessible! But keep in mind it’s a whole different world from traditional web or desktop software development, so don’t be surprised by the relatively steep learning curve. Keep at it though!
- Buy a small robot kit from Amazon or a local reseller. Yahboom make some good robot toy car kits. Hugging Face have the open source SO-ARM101 that plenty of companies manufacture and sell now. Expect to spend about $250 USD including a Jetson Nano for a good kit, up to $1000 USD if you want some more sensors
- If you can’t afford a real robot, play around with simulators like Isaac Sim and Mujoco
- Check out LeRobot, excellent framework for ML robotics from Hugging Face
- Learn the basics of ROS (pubsub), even if you don’t end up using it, a lot of the industry jargon and design patterns come from ROS so it helps to understand it. Think of ROS like Ruby on Rails, it’s a heavyweight batteries-included framework with lots of opinions.
- ROS does have some nice libraries for manipulation (MoveIt) and navigation (Nav2) using more classical (non-ML) methods
- Leverage AI tools such as ChatGPT and Cursor when you get stuck, it’s a lot faster than Googling when you’re just getting started and don’t even know the right term to search for.
- (Shameless plug) Check out two tools I’m working on: mcap.dev for logging and foxglove.dev for visualization
I highly recommend starting with the SO-ARM101 and the LeRobot tutorial. They're super cheap, its insanely quick to get started, and you can even buy pre-made kits like at https://partabot.com . It's the "Hello World" of robotics now, imo.
Don't bother with a Jetson Nano, you don't need that to get started, and by the time you need that you'll know a lot already. You can just drive the robot from your laptop!
Getting to training your own VLA fine-tuned model is a super quick and easy process. You can see examples of other people completing the tutorial and uploading their training/evaluation datasets here (shameless plug for my thing): https://app.destroyrobots.com
I wouldn't bother much with ROS at first tbh. It'll bog you down, and startups are moving toward using other approaches that are more developer friendly, like Rust-based embedded.
You can go far with a robot connected to USB though!
Maybe ROS is the way to get started in robotics and for AGVs it's probably the most resourced thing one could learn.
But that is really only for AGVs. Once you get into the air, it's all Mavlink (always has been), ASVs and AUVs still run MOOS, and robotic arms are far more likely to use ABB studio than Moveit and the research is going towards LeRobot with end-to-end control, not classical methods. IoT systems are mostly Home Assistant, and well, some just use MQTT directly instead.
ROS is probably not as ubiquitous as us ROS devs would like to believe, what one should learn really depends on the field one wants to be in.
Definitely! I agree with everything you said.
ROS is still very heavily used in industry and is growing in absolute terms, but most of the newer frameworks and research are learning towards e2e methods.
But even if you don't use ROS, it really helps to have a working knowledge of how it approaches things like pubsub, logging + replay, transforms, etc.