Robotics and machine learning in the kitchen

All industries which care about consistently producing high-quality output have increasingly invested in automation.

Car manufacturers mounted welding machines on robotic arms to get consistent MIG welds. CNC machines have revolutionised manufacturing. Anyone who has even a little experience machining realises how difficult and time consuming it is to cut metal precisely and repeatedly within tolerances. The ice cream cone we all love is an output of a manufacturing line which churns thousands of ice cream cones in a day, keeping the taste, texture and packaging consistent and within acceptable variation. There are countless other examples.

Packaged food industry embraced automation a long time ago. I remember visiting a Parle-G biscuit factory as part of a school trip in 2002. A flood of biscuits on a conveyor flowed through the factory and the aroma of thousands of fresh, identically and perfectly baked biscuits in the air made us all crave for them. Thankfully they gave all of us large packets of biscuits at the end of the tour.

In contrast, restaurants and kitchens have remained relatively low-tech. Some of the most sophisticated electronic equipment found in kitchens are fridges, dishwashers and mixers. It seems like innovation in the kitchen has stagnated. The most we have come up with is "fusion dishes" and even that didn't really take off.

So why should there be innovation in the kitchen?

A process view of the kitchen:

Kitchens are supposed to take raw resources and process them into healthy and tasty food for the people consuming from that kitchen. There are thousands of kitchens in any city. Homes, hotels, restaurants, offices, cruise ships, offshore rigs, military bases and any place where humans spend a considerable amount of time have kitchens.

It'd be interesting to measure how good these "kitchen instances" are in meeting the needs of their customers. Let's look at the data extracted below from Zomato:

Note: Zomato normalises its ratings but if you look at the raw data curve and the normalised curve, the data doesn't change much in the 4+ ratings range presented above. Normalisation has impacted the 3.5 - 4 range and a lot of restaurants which were supposed to be in that range as per customer ratings have been pushed down.

Essentially, less than 10% of kitchens are rated "Good", if we agree on that definition. One can dig deeper to find that even within this set of 4+ rated restaurants, there is a good correlation between the cost for two and ratings.

Enter technology

We'd all love if our food was cheaper. We'd all love if we didn't have to worry about the hygiene of the place we were ordering food from. We'd all love if we could eat out more while also looking after our health and our pockets. But that would be a perfect world, right?

Well, at least we think it's achievable. Automation of cooking:

• Brings consistency and predictability in the product
• Lowers the cost of food preparation
• Makes it easy to deploy new kitchens
• Increases productivity
• Lowers dependency on skilled labour
• Allows personalisation of food

We look at food preparation from an engineer's lenses and design our complete tech in-house to make it happen. Our machine learning models (CNNs) try to mimic a human sight to learn the cooking process. We prototype new designs almost every week to rapidly improve the quality of the food being made by our robots. Over time we have built our own museum of prototypes and will continue to be so, but its great fun and challenge! We do a lot more that might excite you!

If you are an engineer (Mechanical design engineer, Electronics design engineer, Machine Learning Engineer, Mechatronics etc) and this interests you, we'd love to talk to you!