traversaro.bsky.social
Robotics Applied Scientist
@ami_iit @iCub @IITalk .
Robotics Software @condaforge @RoboStack .
15 posts
341 followers
826 following
Regular Contributor
Conversation Starter
comment in response to
post
I love having cappuccino in the afternoon! The only annoying thing is to have to deal with reactions of other Italians.
comment in response to
post
This is interesting, which bioconda packages use defaults dependencies? As far as I know bioconda stopped using defaults a long time ago, see github.com/bioconda/bio... .
comment in response to
post
Just migrate 3.13 . Adding free-threading support is a different migration that is currently paused: conda-forge.org/status/migra... . As you can read in github.com/conda-forge/... there are some problems in free-threading in some core python libraries that do not make sense to start that migration.
comment in response to
post
It's basically like Mujoco but with more advanced materials/rendering/solvers, written all in Python thanks to being powered by Taichi, which makes it much more accessible.
I like it a lot. It's very accessible.
They went too far with marketing, but willing to ignore it for now.
comment in response to
post
unfortunately not yet. It is only fast for simple scenes but for more realistic scenarios (robust locomotion, manipulation) it seems to be slower (rendering or state only simulation) than NVIDIA isaac lab and my labs simulator maniskill. Sharing a report next week with accurate benchmark numbers
comment in response to
post
Unless I am missing something, as far as I know CUDA is not supported on macOS, so I am not sure if any package manager can help in that situation.
comment in response to
post
I completely agree on this being a social problem (and on social problems being harder to solve then tech ones). Why do you think there will be an increase of pressure on PyPI to solve those problems, if the demand is already met by conda-forge ?
comment in response to
post
Interesting, do you have any reference for the work in progress to support dynamic linking in wheels?
comment in response to
post
For example, Julia package manager started effectively to package non-Julia dependencies for its packages, but it adopted an extremely centralized model with github.com/JuliaPackagi... that seems quite different from the distributed model of how PyPI wheels are built as of late 2024.
comment in response to
post
While PyPI can host also non-Python dependencies (and it does, see for example the cmeel C++ distribution on PyPI, pypi.org/search/?q=cm...), I am not sure how it can scale to handle ABI migrations without a centralized tooling.
comment in response to
post
What I think is relevant in the comparison is not only the comparison between conda and pip as package managers, but also of PyPI and conda-forge as distributions/package indexes, where PyPI is extremely decentralized while in conda-forge the tooling for building binaries is centralized.
comment in response to
post
In Robot Utility Models, we collected a large number of demonstration videos — 1000 examples in 40 environments — using a tool called “the Stick." The lead author even demonstrated it live at CoRL
on completely new furniture. This one really does 100% actually work - I’ve used it in my house.
comment in response to
post
I guess you could also be interested in gitlab.kuleuven.be/meco-softwar..., that is developed by the same group developing CasADi itself.
comment in response to
post
Feel free to open GitHub issues if you have any curiosity! Some of the authors of the libraries are not on bsky, while for liecasadi and adam I just realized that the author (@giulero.bsky.social) is indeed here.
comment in response to
post
The nice aspect of using a Python-based implementation for the kinematics and dynamics algorithm is that is more readable for lab members. The semantics used for the algorithms class is inspired by iDynTree, a in-depth comparison w.r.t. to Pinocchio is available at github.com/ami-iit/idyn... .
comment in response to
post
We have some work in progress in that area (i.e. planning via direct transcription of optimal control problems with CasADi) at github.com/ami-iit/hipp... . We do not use Pinocchio but a Python-based implementation of the usual robotics dynamics algorithms available at github.com/ami-iit/adam .
comment in response to
post
It seems to me that the main focus now is to support really well multiple nodes that start all together as part of the same process, while they are still working on the "distributed" part (see github.com/dora-rs/dora...).
comment in response to
post
I have no direct experience, but I am lurking a bit on their GitHub/Discord. I think is interesting especially as it is quite different from ROS 2 (the pub/sub part). They seem to have a huge focus on reproducible/deterministic dataflow programming, as opposed to distributed nature of ROS.