Mat is on a well-deserved vacation, probably lying on a beach somewhere. We don’t want this to become Monthly Robotics, so I, Ignat Georgiev, am preparing this and the next issue. It will still be the good ol’ Weekly Robotics newsletter you love, but with my robot learning spin on it! Namely, “Publication of the Week” will temporarily be rebranded as “The Robot Learning Corner”, bringing you the latest research in making robots smarter! Oh, and I’m bringing back the AI Image of the Week!
Weekly Robotics is being developed thanks to the Patreon supporters and the following business sponsors:
How Robots-as-a-Service Companies Can Fundraise In A Challenging EconomyRaaS Fundraising Webinar
Looking to fund your robotics startup? Last week we hosted a roundtable on fundraising for robotics companies, featuring Tessa Lau (CEO of Dusty Robotics), Nag Murty (CEO of Electric Sheep Robotics), and Mike Oitzman (Editor at The Robot Report). Read the summary and watch the full webinar here.
Linux Embracing Rust Will Boost Robotics Community
In this article, Brandon Minor discusses how introducing Rust into the Linux Kernel can affect the robotics industry, which is increasingly relying on Linux. I am a huge fan of Rust, which will hopefully accelerate its popularity within the robotics community. Why bother waiting for C++ to modernize itself when Rust is already here, just as fast, easier to use and most importantly, memory safe!
Game Changing Development Program 2022 Highlights
NASA published a compilation of their Game Changing Program that, unsurprisingly, contains many robotic systems. A favourite of mine is their CADRE project where a “momma” lunar rover deploys a swarm of smaller robots that go around and explore extraterestrial planets.
ROS 2 — DDS Shaping The Future of Next Generation Robotics and Autonomous Vehicles
In my opinion, DDS is biggest advance that ROS 2 made back when it came out. It makes ROS as a whole more scalable, robust and efficient, bringing it much closer to production-level applications. In this artcile Rohan presents a quick into to DDS and how it integrates into ROS.
What Robotics Experts Think of Tesla’s Optimus Robot
A comprehensive set of opinions on the new Tesla Humanoid robot from academics and industry experts. I found the diverse viewpoints on Optimus refreshing and eye-opening. The general sentiment seems that people aren’t impressed at present but are excited to see where the momentum Tesla has built will take them. Personally, I am happy that Elon Musk did what he did best - bring bold moonshot ideas to the forefront and make headlines. Even if the project fails, it will inspire many future roboticists!
The Robot Learning Corner
I am thinking of creating a dedicated robot learning newsletter. If you are interested in that, you can show interest here!
Mastering Atari With Decrete World Models
Reinforcement Learning (RL) has made some splash in the past couple of years, but we rarely see it applied to robotics in some breakthrough way. I think the reason is that people have primarily focused on model-free RL, which requires heaps of data to become good at something. Sadly robots tend to break before they gather much data. That’s why I am a massive fan of model-based RL, which is more data-efficient and one of the most exciting directions for robot learning. This paper introduces the Dreamer algorithm, which pushed model-based RL to new heights. Simulations are boring, though, so here it is applied to actual robots.
LM-Nav: Robotic Navigation with Large Pre-Trained Models of Language, Vision, and Action
Imagine how you tell DALL-E to paint you whatever you want. Well, imagine DALL-E but applied to a wheeled robot that can drive a path that you ask it to drive, such as “After passing a white building, take a right next to a white truck”. It sounds straight out of a sci-fi movie, but this is what the authors did here by combining large pre-trained language, vision, and navigation networks. Interestingly enough, this work actually builds on the DALL-E work! Your painting AI is now also your driving AI! I’m excited to see more cross-disciplinary work like this!
Amazon ends testing of Scout delivery robots
Amazon appears to be shutting down yet another of its big robotics projects. I know some folks who have been working very passionately on that for years, and it is a shame to hear they have to abandon their work now. The last-mile robot delivery scene is quite hot. Maybe that will give some breathing room to some of their competitors like Starship, Kiwibot, or Theo?