MIT’s Revolutionary Home Robotic System PIGINet Simplifies Home Life

Mornin’ miners⛏️,

Happy Thursday!

Welcome to "Digger Insights" - your daily 5-minute enlightenment on recent tech updates! In our enjoyable and easily digestible dispatches, we break down the latest tech trends, giving you a quick and comprehensive view.

Join us, and in just a few minutes each day, you'll gain an advantageous perspective on the ever-evolving tech landscape. Ready to dive in?

Let’s get to it!

Today’s Highlights:

• 🏠MIT’s Revolutionary Home Robotic System PIGINet Simplifies Home Life🤖

• 💸Drone Startup BRINC Raises Over $80M💰

MIT's Home Robotic System PIGINet Makes Housework a Breeze

It is irrefutable that at least some of us are familiar with robots, mostly from fiction, be it protocol droids like C3PO from Star Wars or very human-like, crime-fighting cyborgs like Robocop.

Though mainly circulating fictional worlds, various different kinds of robots have been seen plenty in real life, too, whether they be delivery robots for pizza establishments like Domino’s, articulated robots used to assist in surgical procedures, or robot vacuum cleaners like Roomba.

Although the varieties of robots have started becoming endless, robots for home assistance are still relatively scarce and underdeveloped, and it is a wonder why. Having an autonomous helper right at home, simplifying your day-to-day housework, seems so ideal, but is it simple to achieve?

This is what researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) are trying to develop using a machine learning system they call PIGINet.

Plans, Images, Goal, and Initial Facts

Autonomous assistance systems have become quite common on warehouse and factory floors, assigned to clean, pack, stock, sort, and even perform inventory work. This is due to the structured nature of warehouses and factories, which helps robots easily navigate their way to work.

On the contrary, homes, especially those owned by big families with pets or little children, tend to be, at the very least, a little more chaotic. Homes are dynamic places that are often full of “unfriendly” obstacles for robots, with furniture getting moved around and things scattered on the floor. As a result, a more intricate system is required for robots to operate well. This is where MIT CSAIL’s PIGINet system plays its part.

PIGINet, short for Plans, Images, Goal, and Initial facts, is a system designed with a neural network, giving robots the ability to create plans of action in varying environments.

Typically, household robots are trained to do a task by practicing the moves they need to do over and over until they eventually find a feasible solution. This form of training entails a sort of predefined recipe robots have to follow, which tends to be time-consuming and inefficient, especially when the robots are faced with different obstacles or environments every time. With PIGINet, the iterative process of task planning can be greatly decreased.

PIGINet utilizes a transformer encoder designed to operate on data sequences. The system inputs sequences of information, which include the task plan, the images of the environment, the initial state of the environment, and the desired goal. The encoder combines said information and generates a prediction of whether or not the selected task plan would be feasible to perform.

Kitchen Tasks

The team tested the PIGINet system by conducting a series of kitchen-based activities, simulating home environments with various different elements like counters, cabinets, fridges, sinks, etc.

The key idea of this project is to train robots to become adaptable problem-solvers instead of simple recipe followers. To do so, the team required good training data, which proved to be scarce. This scarcity was one of the major challenges the CSAIL team faced and resulted in quite a slow starting process.

After incorporating pre-trained vision language models and data augmentation tricks, however, the team observed that the robots could grasp spatial arrangements and object configurations more easily.

According to a paper the MIT CSAIL team published regarding this experiment, the team managed to reduce the system’s planning time by 80% and about 20-50% for more complex situations and tasks involving not only seen objects but also previously unseen objects, using zero-shot generalization.

Future Endeavors

Although focusing on kitchen-related tasks to further develop PIGINet at present, the team believes that the practical applications of PIGINet will not be confined to households, as it has been proven that the system can indeed navigate through complex and dynamic environments.

The MIT CSAIL team aims to further refine PIGINet by suggesting alternate task plans when an infeasible action is identified. If the team manages to make this happen, the generation of practicable task plans can be done much more swiftly without big training datasets.

The work tackled by this ambitious team is certainly far from simple, and can surely pave the way for revolutionizing robot development and training.

Funding News

BRINC Raises Over $80 Million

Seattle-based drone and aerospace startup BRINC has raised over $80 million following its Series B funding, supported by major players in technology and venture capital such as OpenAI’s Sam Altman, Index Ventures, Tusk Venture Partners, and former LinkedIn CEO Jeff Weiner’s Next Play Ventures.

Big celebrations are in order for CEO and founder Blake Resnick, who, during a board meeting in late July, heard the news of the startup’s Lemur 2 drone’s first successful production test flight.

Already thoroughly ambitious and accomplished from a very young age, once interning at McLaren Automotive, Tesla, and DJI before dropping out of Northwestern University’s mechanical engineering program, the 23-year-old CEO founded Brinc in 2017 and is currently overseeing a team of nearly 100 people, serving over 400 customers.

With the funds raised, Resnick plans to develop and refine the public safety-focused Lemur 2 further before its release and first shipment later this year.

Meme & AI-Generated Picture

Job Posting

• Motorola Solutions - IT Systems Administrator - United States (Remote)

• Intercom - Coupa Administrator - United States (Remote)

• Merck - Scientific Computing Systems Administrator - New York City, NY (Hybrid)

• Vendavo - Technical Engagement Manager - United States (Remote)

Promote your product/service to Digger Insights’ Community

Advertise with Digger Insights. Digger Insights’ Miners are professionals and business owners with diverse Industry backgrounds who are looking for interesting and helpful tools, products, services, jobs, events, apps, and books. Email us [email protected]

Your feedback means a lot to us, please send it to [email protected]