When you hear Machine Learning(ML) what is the first thing that comes to your mind? Do you imagine algorithms running on hardware with lots of computational power? Or perhaps GPUs and CPUs performing complex tasks and continuously learning with AI. When there are no software and hardware restraints, designing Machine Learning systems and algorithms that solve real-world problems is possible; however, the real challenge is when there are limitations, such as the lower computational power of robots. Today we'll learn about how researchers are playing a pivotal role in identifying these problems and are coming up with real solutions to overcome them!
While ML concepts and practical solutions are seen deployed in many large industrial robots, the same cannot be said for smaller, affordable, and resource-constrained bots. Today, smaller robots, also known as tiny robots (resource-constrained, low-cost, low-weight autonomous robots), are possibly the most affordable, usable, and safe solutions for the general public.
To tackle these challenges and provide a bedrock of research for developers, manufacturers, and designers, researchers from Harvard, University of Virginia, CMU, Google Brain, and Delft University of Technology came up with an excellent paper titled "Tiny Robot Learning: Challenges and Directions for Machine Learning in Resource-Constrained Robots". The paper starts by outlining 4 fundamental issues that come in the way of ML with Tiny Robots. These issues are called SWAP (stands for Size, Weight, Area, and Power).
The article also provides a clear understanding of what Tiny Robots are & outlines the differences between a larger-size robot and Tiny Robots. As you read through the paper (link provided at the end of this article) and see in the image below, you'll see numerous references to Petoi Bittle programmable robot dog as one of the Tiny Robots of choice that is used as part of this study.
The paper shines as it gets into the crux of the matter, identifying problems and suggesting solutions as seen in the Challenges and Opportunities section. Broadly, the 4 major challenges that are highlighted are:
- How to work around the SWAP constraints
- Sensor and actuator limitations
- Implementing end-to-end design
- Designing robots for various use cases and scenarios
All in all, the research paper is a brilliant read for anyone interested in learning about robotics and Machine Learning. We are grateful to the team for featuring Petoi Bittle as well! The research paper can be found here: "Tiny Robot Learning: Challenges and Directions for Machine Learning in Resource-Constrained Robots"
Tiny robots are not only for home use, but they are often seen in crucial situations like search and rescue, industrial planning/structural health monitoring, and inspection.
Most importantly, with the advent of STEM robotics, they form the backbone of STEM/robotics education in classrooms around the world as students get their first taste of coding and STEM while interacting with these programmable robots. With so many applications, ease of access, and utility for all, integrating Machine Learning with tiny robotics is a no-brainer. Petoi Bittle robot dog is designed with free robotics curriculums for educators for engaging robotics and AI education.
Interested in learning more about how Bittle can help advance AI robotics learning? Click on the image below to learn how you can talk to your Bittle Robot Dog using ChatGPT!
