The Eye of AI #11

The Eye of AI #11

Computer vision is a truly transformational technology for the retail industry, but it has applications that touch nearly every aspect of our lives. This is our weekly summary of what happens when cameras, computers and great ideas get together.

Robot Dog Enhancements

MIT people pushing hard against the stereotypes with a super-smart robot dog. Photo:MIT

We've all seen videos of various Boston Dynamics robots doing all sorts of borderline-un-nerving things. Not many of us saw those things and thought "What this incredibly agile machine needs is the ability to look around and figure out what is important to it", apart from a team of engineers at MIT. They have developed "Clio," a new system that helps robots identify the objects that matter most to their tasks. Clio uses natural language processing and computer vision to zoom in on relevant objects while ignoring distractions—a big win for future rescue missions, domestic robots, or anyone who wishes they could play 'fetch' with their multi-million dollar robotic dog.

Real Dog Enhancements

Picking up the pieces. Photo: Caleb Olson

We've been domesticating and cohabiting with animals for thousands of years, but efficiently managing their waste has proved elusive (although jazz composing legend Charles Mingus wrote a pamphlet on toilet training cats). A system invented by Caleb Olson looks like every pet owner's dream, arming a drone with real-time computer vision and machine learning to detect, scoop, and dispose of dog waste, all hands-free. The "world’s first aerial dog poop removal system" features a bell-shaped container that picks up the mess with precision landing guided by clever off-board algorithms. Currently a prototype, Olson hopes it could revolutionize backyard cleanup or even inspire a commercial service.

For the Computer Vision Hobbyists

Raspberry Pi's AI Camera. Photo: Sony

Some things are just better if you make them yourself - like bread, pasta, and image recognition systems. If you're looking to identify wildlife in your back yard, work out how much time your pets spend on the furniture while you're at work, Raspberry Pi has a new solution for those who like to build their own things. The AI Camera, developed with Sony, packing 12.3 megapixels and Sony’s IMX500 image sensor. This compact camera can handle advanced computer vision tasks like facial recognition and motion tracking, all without the need for an external computer. Thanks to its highly-functional microcontroller, machine learning models run directly on the device. Priced at USD$70, it's ideal for AI projects in robotics, home automation, and more—making cutting-edge tech accessible to students and developers alike.

Cats Eyes

Cats are pretty good at manipulating people and seeing things. Photo: Universal

We have already covered how researchers have tried to see the world through various kinds of eyes. Well, inspired by the vision of cats, researchers have developed a cutting-edge artificial vision system for robots. Mimicking feline eyes, the system features a custom aperture and silicon photodiode array that enhance low-light vision and improve object tracking, even in challenging lighting conditions. This tech, which breaks camouflage and reduces computational load, could be a game-changer for applications like autonomous drones and surveillance robots. Although its field of view is limited, future advancements may overcome this hurdle.

More More Efficient Vision Transformers

Visual depictions of AI are hard to get right. Photo: RTInsights

It seems every week there is a new breakthrough in the efficiency of computer vision modelling. This week is no exception. Researchers from City University of Hong Kong and Tianjin University have introduced the Visual State Space Duality (VSSD) model, a breakthrough in vision transformers. This new model enhances the efficiency and performance of computer vision tasks by processing long sequences of image data more effectively than traditional transformers. VSSD's innovative approach, which integrates motion data and refines predictions using memory mechanisms, shows superior results in image classification and segmentation. Its potential applications range from autonomous driving to mobile robotics, offering faster and more accurate visual processing


That's everything for this week. Please keep an eye on the SAI Group blog for everything that we're thinking and talking about.

Got some cool tech to share? Whether its your own project, or just something you saw and thought "I want people to know about this!", let us know about it and we'll include it in upcoming editions.

要查看或添加评论,请登录

SAI的更多文章

社区洞察