Today’s issue is a nice one project-wise. I enjoyed learning about all the projects featured in this issue. And the two articles are the icing on the cake! As usual, the publication of the week section is manned by Rodrigo.
Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation
This system has been making rounds on social media in the last couple of days. ALOHA is a whole body manipulation system consisting of a mobile base and two manipulators. The total cost of the system is $32k. The system is used for imitation learning, where the user can demonstrate a task for the robot. When co-training, the researchers obtained over 80% success rate for the tasks with as little as 50 demonstrations per task. I recommend checking out the project page for some good project videos. For more information about this project, check out this paper.
Spanner Rescue Robot
What do you do if you drop a spanner and can’t reach it? Of course, you build the robot to fetch it. And then you iterate on it. At least, that’s what Keegan Neave did with this project. I love this kind of story.
Logging on Embedded Devices
In this post, Victor Lai does a deep dive into logging on embedded devices with some good practices and issues you will likely encounter, not only if you develop low-level systems.
Lessons Learned by a Software Guy Venturing into Hardware
Here is an interesting article from the maker of SidecarT, a Raspberry Pi-powered ROM cartridge emulator for Atari computers. I found the insights contained in the article quite helpful, but one thing I would appreciate in an article like this is learning how to get a CE marking as a self-made maker.
If you’re not developing with this, you’re wasting your time
I’ve had a significant FOMO on dev containers for ROS development for quite a while now, but with this video by Josh from Articulated Robotics, I don’t have any excuse left not to look into them. Thanks Josh!
Remote-Control Kinetic Sand Table Uses A Single Arduino
Amazing what you can do with just an Arduino Uno, a CNC shield, and two stepper motors!
Publication of the Week - ODIN: A Single Model for 2D and 3D Perception
We’ve seen many models that separately are capable of segmenting images or labeling them, such as the famous Segment Anything Model (SAM) from Meta. This paper presents a single model for 2D and 3D segmentation and labeling based on a transformer architecture. The results achieved state-of-the-art performance in some benchmarks. Check out their GitHub page to test the code and get more content.