Chef B Makes It Look Easy; Yet Teaching a Robot Nuance is Work

This is part two of a two-part interview with Venki Ayalur, one of the founders at Blendid. You can read part one.

Imagine you’re cooking your morning scramble. You add the potatoes and the bell pepper and tear apart the fridge to see if you have any gouda left. After you add the eggs, you taste the mix to see if your seasoning is right. Nope, a bit more salt, needs pepper and maybe a little cayenne. It’s all nuance: a little of this, a little of that, bit by bit so you don’t over do the salt or peppers. Thank goodness for taste, it’s an essential part of cooking – except when it isn’t!

Creating the perfect blend when taste is critical for the final result.

“I had to figure out how to marry the software and the hardware to allow the robot to properly access eight solids, eight powders and four liquids. Each of them would be used in a different combination to create a different kind of blend.” Venki explains it’s much more detailed than it looks. “Consider the number of combinations: it’s close to infinite if you consider the quantity of each ingredient is also a variable. We had to include machine learning and artificial intelligence to help us navigate the variables.”

The way the machine learns is through a continuous feedback flow. Venki and team reviews that data as the robot begins to adjust. But it’s still not that easy.

There’s the vision and then there’s the reality of how it really works.

“Just to give you an idea of how meticulous the programming is, let’s start with the cup. What is the weight of the cup? Will the robot pull just one or will they stick together?” Venki kind of laughs when explains, “It’s a delicate dance figuring out how every step impacts the final product.”

With creating great blends as the vision, managing just the cups had to be right and right 100% of the time. Or at least that was the goal. Venki and team created models and then started testing and learning. “Oh my gosh we had some surprises! It grabs two cups by accident. We fix that. It grabs the cups in the wrong place, we fixed that. We were making micro adjustments all the time,” Venki said. “We are endlessly collecting data; piles of data, tons of data!”

Basically, it’s just a blender right? Wrong.

“I don’t think people fully appreciate what’s really happening here,” Venki explains. “It’s so much more. We manage all aspects of dispensing a completed blend. It’s more than cups and ingredients. There’s the location of the cup. If it’s out of place, we could end up with a blend on the floor. There are the ingredients. If things stick together, that affects the weight and a blend might be more than 16 ounces.”

“This becomes even more complicated as we allow our customers to personalize their blends,” Venki continues. “Let’s say you want a Green Warrior (a kale-based blend) but you want extra blueberries. Now Chef B has to understand how to adapt the ingredients to accommodate that change. There’s the nuance.”

Working on the Blendid technical team is a lot like a forensic scientist. You must pay attention to the data so you can learn and make changes. “That’s exactly what we’re doing. Every single step. We have data and we collect it, send it to the Cloud and then we analyze the data. From that we can create models and apply what we learn.”

Now when you watch Chef B make your blend, you’ll realize how hard it was to make something look so easy!