Making the Most Out of the Shade Registry

Why We Built the Shade Registry

We love robotics. The fact that we’re able to take a piece of hardware, slap on a Jetson Nano, control some servos, and make it do amazing things is super exciting (okay that’s a little scrappy but you get the point). Yet the tools that are out there that enable us to build out these robots are far and few between. Yes we have the Robot Operating System and a few amazing packages like Nav2, MoveIt and more, but what about all the countless perception algorithms, classification, and SLAM algorithms that have been built out to be THE state of the art research?

We’ve built out the shade registry out of the struggles to try to even get our robots to run. After consistently failing at getting various open source SLAM algorithms to compile and run for a ROS 2 robot along with the constant rewriting of code and wrappers to support new algorithms like YOLOv5 and the Vision Transformers, we decided to open source them all and provide readily available docker containers so that anyone can pull them and get their robot up and running in seconds.

If we are to advance the robot industry, we need to provide these systems and tools for researchers and companies alike to quickly conceive and create these proof of concepts at light speed, not being bogged down by large amounts of time and effort just setting up the code and getting the robot to communicate with various output. With the Shade Registry, our goal is to support the state of the art research and algorithms that are being published by the top AI researchers and the top roboticists across universities and companies.

How do I get started?

Getting started is simple. All you have to do is head out to any of the algorithms that you’d like and run the docker pull command for the ROS distribution and the configuration that you need. For example, for Shifting Windows Transformer (SWIN) for image classification, all I have to do is pull the docker container for foxy like so:

docker pull swin:foxy-base-patch4-window12-384

Now to get this up and running locally on my computer I simply need to run:

docker run -v /dev/shm:/dev/shm -t swin:foxy-base-patch4-window12-384

Or for your LAN network:

docker run --net=host swin:foxy-base-patch4-window12-384

If you have any issues! Feel free to reach out to us at hello@shaderobotics.com or join our discord

So Where is the Shade Registry Going?

Right now, we’ve created a sample set of over 18 different algorithms ranging from image classification, object detection, SLAM, and semantic segmentation. Our goal is to continue to increase this set of algorithms to encompass motion and path planning, scene graph construction, depth estimation, NLP, and audio for a variety of uses cases. We also want to enable all of our docker containers to be finetuneable to a downstream task of your choice.

In addition to algorithms, our hope is to be the aggregator of all research and work in the robotics space. This includes providing easy access to critical datasets via a Shade CLI and directly from the registry. This allows users to easily test out their algorithms and robots on these datasets ranging from video, to LiDAR, to RADAR, and more.

Finally, we hope to aggregate URDF models that can easily be adopted for simulation and testing of new algorithm techniques. From our own experience of building out these models, these take the most amount of time when conducting research. Creating general URDF models enables better reproducibility of research and ultimately faster time to market.

For more examples and applications, make sure to keep up to date with our blog and our medium.

How you can help the Shade Registry

We’re just getting started and we’d love to hear your thoughts and support as we continue to do it. For starters, try out our docker containers, let us know where things are working, what isn’t, and we’ll be prompt to fix them all. We’ve thoroughly tested all of them across devices, and manage a CI/CD pipeline for you. Second, join our discord and ask for the next set of algorithms that you’d like to see supported. If you’re even more ambitious, consider building out your own supported ROS2 wrappers, and we’ll build and push them up to the registry for everyone to use!

If you have datasets that you want to upload and store for the rest of the world, we’d be happy to get in contact with you as well.

Finally, start using them in your projects, post an issue on any of the GitHub’s, we’re so excited to see what you build.

Who are we and what are building?

We’re a startup that wants to build out the tools and infrastructure needed for the next generation of roboticists and companies. As roboticists ourselves, we’ve struggled with the existing tools provided and want to build out the tools that enable anyone to do cutting-edge research and bring robots to the masses. Some of the projects that we’re working on now include a data platform for robotics that includes smaller storage requirements for ROSbags, data observability into use, automatic scene understanding for indexing, and integrations with third party labeling and cloud platforms. We’re also exploring robot release management and deployment including feature flagging and A/B testing.

If any of these problems is one that you’re facing now, we’d love to talk to you at hello@shaderobotics.com, or feel free to contact us here.

Best of luck to all and happy building!