You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Would you be interested in some python bindings if I wrote them?
Things I've tried so far and why I'm tempted to bind this library
My current use-case is for robot arm teleop for recording episodes for training models, so nothing ground-breaking (I did some playing around on a branch over at huggingface/lerobot@main...alsuren:lerobot:teleop a while back, but mostly exploratory and not super coherent).
The current options for ik in python seem to be:
moveit/moveit2 - quite big, and tied into the ROS/ROS2 ecosystem. Quite painful to get working on macos.
https://github.com/Phylliade/ikpy - allowed me to make a good start, but the terminology mismatch with URDF is a bit jarring, and the types that mypy/pylance infers are a bit confusing (from memory).
https://github.com/UM-ARM-Lab/pytorch_kinematics - allows you to set initial position and will give you answers that don't quite converge if you need (useful because the SO-100 arm is not able to reach all positions/orientations). Feels like it's really designed for putting in the training loop in simulations or something, because everything is vectorised and the "i only have one robot and I want to control it in real-time" API is slightly less obvious as a result.
I stumbled across this library when searching crates.io. It seems to handle URDF well, and I expect it to have reasonable types that will result in an easy-to-follow python api.
My vague plan is to set things up with maturin and publish some wheels to pypi using some bogus crate name (alsuren-openrr-k-py or something) or just use it as a path/git dependency. If this works well then I can send you a PR and you can pick a better pypi project name to publish under.
Does this sound like a valuable thing?
The text was updated successfully, but these errors were encountered:
Would you be interested in some python bindings if I wrote them?
Things I've tried so far and why I'm tempted to bind this library
My current use-case is for robot arm teleop for recording episodes for training models, so nothing ground-breaking (I did some playing around on a branch over at huggingface/lerobot@main...alsuren:lerobot:teleop a while back, but mostly exploratory and not super coherent).
The current options for ik in python seem to be:
moveit/moveit2 - quite big, and tied into the ROS/ROS2 ecosystem. Quite painful to get working on macos.
https://github.com/Phylliade/ikpy - allowed me to make a good start, but the terminology mismatch with URDF is a bit jarring, and the types that mypy/pylance infers are a bit confusing (from memory).
https://github.com/UM-ARM-Lab/pytorch_kinematics - allows you to set initial position and will give you answers that don't quite converge if you need (useful because the SO-100 arm is not able to reach all positions/orientations). Feels like it's really designed for putting in the training loop in simulations or something, because everything is vectorised and the "i only have one robot and I want to control it in real-time" API is slightly less obvious as a result.
I stumbled across this library when searching crates.io. It seems to handle URDF well, and I expect it to have reasonable types that will result in an easy-to-follow python api.
My vague plan is to set things up with maturin and publish some wheels to pypi using some bogus crate name (alsuren-openrr-k-py or something) or just use it as a path/git dependency. If this works well then I can send you a PR and you can pick a better pypi project name to publish under.
Does this sound like a valuable thing?
The text was updated successfully, but these errors were encountered: