Expo

A worker using the Expo interface to control the Picnic Pizza Station

Introduction

When Picnic launched the Pizza Station robot, I led design for its interface.

Expo lets you control a half-ton robot that can produce over a hundred pizzas an hour, placing ingredients on dough like a 3D printer. It’s designed to deploy into overwhelmed commercial kitchens—workers use Expo even with little training.

I shipped the first version of Expo in 2020; with market feedback, I retooled key workflows and introduced a high-visibility design language to improve its fit in the kitchen.

I led this project under director Kyu Han in collaboration with designer Jack Lo and software engineer Derek Soike.

Worker using the Picnic Pizza Station.

Capabilities

Expo consolidates the robot’s workflows and mediates them with the labor cycles of real kitchens. As a worker, you need to be able to do three things with maximum efficiency:

enter orders
make pizzas
clean and disassemble the robot

Capabilities:

Queue

Expo's Queue is a to-do list of upcoming pizza production. Some activities need user input while others are autonomous. It breaks large tasks into steps you can follow without memorizing.

The Queue decomposes kitchen tasks to follow. The chevron points to the current task

Capabilities:

Utilities

Expo’s Utilities cover tasks beyond pizza-making, like internal cleaning. The utility page arranges these functions in line with the kitchen workflow: ingredient priming, cleaning, then disassembly, and remembers progress along each workflow.

Utilities cover the remainder of the robot's functionality

Development

When I joined Picnic in 2019, we faced a six-month deadline: build the alpha version of our robot, withstand quality inspection, and install it at T-Mobile Stadium, home of the Mariners.

Integrating the robot so quickly took rapid engineering compromises. To center the project, I generalized a worker’s labor activities as Start of Day, Mid Shift, and End of Day. Cross functional teams used this model to describe the multivariate impact of their work in common terms. The user became the focal point of development.

Picnic-branded control board: a hardware integrator

SOD, Mid Shift, EOD

Development:

Food Safety

I teamed with engineering to include self-cleaning subroutines for the robot to meet National Food Safety standards.

These snippets of machine code run cleaning fluid through the robot's inaccessible mechanisms, but you need to be there to load and drain for each one.
‍
For the worker, I framed these as interactive walkthroughs. Given our complex hardware system, we also built a test environment to try out different control logic and operating procedures.

Without NFS certification, the robot is too dangerous to sell.

Utility screen featuring ingredient prep and cleaning walkthroughs

The utility screen condenses operating instructions

Our machine code test environment. The bucket was part of initial operating procedures we managed to prune

Over time, I was able to streamline the cleaning process to the point that you can now access all cleaning walkthroughs on one page, letting you complete and check their work using a single end of day screen.

Development:

Order Entry

Data streaming to a restaurant’s point of sale is cost-prohibitive. Before we shipped, we needed a speedy way to enter pizza orders directly on the robot.

Our customers’ top-selling pizzas are unmodified, but any pizza can be customized. I hid these options directly on their parent element until active; adding a pizza reveals ingredients, and adding an ingredient lets you change its portion.

Adding base pizzas takes one tap, and the screen is uncluttered while customization requires no additional navigation. In our trials, this method was faster to use than our generic POS.

Between visual clarity and feature-richness, we chose visual clarity. This would slightly favor the experience of the untrained user, but fit our demographics: the average seniority in a pizza restaurant is eight weeks.

Progressive disclosure of pizza options

Deployment Lessons

With remarkable effort, we installed our first robot in the Mariners’ stadium kitchen...and made a lot pizzas with sparse, unsalable ingredient coverage. I joined the onsite team to find out why.

While engineering tried to diagnose a problem with our dispersal hardware, I looked at the robot’s labor interdependencies in context: it handles half a chef’s tasks by assembling pizzas, while a worker handles ingredient prep and baking.

I realized our queue displayed expected pizza production but omitted the ingredients needed for that production. We were asking users to manage refill timing like one chef even though the robot—autonomously—took over the task that sets the pace. Pizzas would come out without sauce or cheese, you’d throw them away, curse, and only then reload the ingredients. On top of everything else, this posed a major adoption hurdle.

The Queue with relevant alerting

Reloading tasks can be completed inside the Queue

I introduced ingredient refill flags to the queue attached to the pizzas dependent on them. This significantly squared task sharing with the robot. Pizza production became free from ingredient constriction. Using the robot required less guesswork in general.

Ensuring transparency in the robot's processes became a priority as we refined the interface, with a renewed focus on in-depth, onsite research.

Visual System

We pushed our information hierarchy on the workflow level to meet the communication challenges of a commercial kitchen. We ruthlessly focused the task list the robot showed to workers, aiming to display one clear task at a time.

Then we overhauled the visual design with high visibility colors in a limited palette. We reoriented the screen layout for vertical space and scaled touch controls for easy interaction at arm's length. Inspired by smartwatch UIs, I upped proportional contrast until the most important information on any given screen was legible from a distance of about 16 feet.

Initial language of Expo

Expo after being modified to better suit kitchen conditions

Initial Expo Queue components

Expo's new high-visibility components

For overall clarity, we avoided introducing shorthand and abbreviations. Spending extra time on icon legibility, we also aimed for language-agnostic design. One day, this would help market penetration.

Expo

Product Design / Interface / Research / Testing

Picnic’s Pizza Station is a robotic makeline deployed into overwhelmed commercial kitchens, placing toppings on dough like a 3D printer. To succeed in this operating environment, I designed an iPad interface, codenamed Expo, to serve as our single point of interaction and kitchen display system (KDS).

I led this project under Kyu Han, director of design
In collaboration with Jack Lo, designer
Derek Soike, software engineer
Cole Rogers, controls engineer
Chris Harberle, product manager

01 ▸ Overview

Brave New Robot

When I joined Picnic in 2019, we were struggling to make the leap out of our test kitchen and into our first commercial setting. In testing, expertly trained engineers ran the Pizza Station from huge terminals. In real kitchens, that infrastructure can’t be there, and we couldn’t budget customer training: the median labor seniority in the restaurant sector is less than eight weeks.  

We conceived of Expo for regular kitchen workers to fluently operate a half-ton robot they’ve never seen before. Post launch, as an initiative to minimize adoption friction, Expo’s role increased to full parity with restaurant point-of-sale (POS) and kitchen display systems. I designed the order management feature to fulfill these functions, as well as the robot operation UI.

Fully-featured Expo was critical to Picnic’s expansion¹, allowing us to partner with more kitchens than ever before without dependency on custom integration solutions. Ease-of-use scores were immensely favorable and new user training time fell from a couple of days to a couple of hours. Order management improvements let workers devote uninterrupted time to operating the Pizza Station, realizing throughput gains of over 70%.

I led this project in collaboration with
Kyu Han, director of design
Jack Lo, designer
Derek Soike, software engineer
Cole Rogers, controls engineer
Chris Harberle, product manager

00 ▸ Overview

Brave New Robot

When I joined Picnic, we were struggling to make the leap out of our test kitchen—controlled directly from standing terminals—and into our first commercial setting. We needed users to be able to operate the Pizza Station without help, and we couldn't rely on training: the median labor seniority in the restaurant sector is less than eight weeks.  

I designed the prototypes, behavior, and visual system for our first customer facing interface, and shipped continual improvements after our market debut. Expo fulfilled our business goal of providing regular kitchen workers clear workflows, letting them proficiently operate a half-ton robot they’d never seen before. I oversaw feature additions and refinement on an ongoing basis post-launch as we expanded.

The main order queue is the home screen of the Pizza Station. The Station makes any pizza added to the Queue.

The order entry screen

The cleaning utility screen

01 ▸ First Steps

Designing the general layout

My first explorations for the interface assumed having quick access to ALL available controls would build confidence when dealing with something so complex.

This was the wrong approach. The control panel looked familiar if you're used to industrial hardware, but was overwhelming and categorically alien to our users.

Rearranging these controls around a representation of the Station also didn't help.

These failures proved I'd need to develop an organization scheme that came from a familiar place. I refocused on the workflows of the kitchen instead of the shape of the technology we were building.

From conducting live process studies, I found the intimidating complexity of using the Pizza Station could be broken down to just two existing kitchen workflows: making pizzas and performing maintenance.

I surfaced the backend queue as the main feature of the app as most working hours would be spent making pizzas.

Performing maintenance happened twice a day, and I located these out of the way until needed. This was the wining approach, which we carried throughout the life of the interface.

02 ▸ Visual System

Getting the look right for high readability

Most apps are designed for arms' length use, but workers would need to know what was happening on screen from across a kitchen in full swing.

I experimented with different levels of information density and layouts. Over iterations, I started taking cues from consumer apps, rather than industrial dashboards, leading to the evolution here.

The persistent view of the completed orders was axed almost immediately, as they did not require ongoing management

I moved to a portrait layout to take advantage of vertical space for our Queue

Progress bars were removed in the final version. Since every pizza was processed in under a minute from start, this information was not helpful

I took the opportunity to increase size and vibrancy of the elements here

Streamlining the information onscreen to suit our chaotic operating environment was my highest priority.

I included some extreme scale elements, inspired by the info design of smartwatches, and then added density as this scale would allow. I also increased the color contrast for elements of top-level importance.

03 ▸ The Queue Interactions I

Fine tuning Queue interactions for kitchen use

Users need to be able to rearrange the Queue to respond to any developments in the restaurant. I experimented with degrees of control.

My first iteration of this behavior let users tap and drag any order to any spot in the queue. This was an unnecessarily fine degree of control and required a user be able to both recognize and execute a gesture control with kitchen gloves on

The final version used a simple "send-to-back" and "rush-to-top" set of buttons, which aligned with the general degree of order rearrangement in kitchens while also allowing any degree of stack fine tuning for extreme edge cases.

The next problem to solve was letting users know when to refill toppings, as they would often run out and be surprised.

I realized the pizza queue could be used to closely estimate when a refill would be needed. I removed our passive fill display and instead injected a warning directly ahead of the pizza that would use the last of that topping. This eliminated passive monitoring from a worker's responsibility and allowed them to use their own assessment of the kitchen environment to choose a good time to reload, with plenty of heads up.

04 ▸ Order Entry

Making the Order Entry screen fast and foolproof

Restaurant menus have certain invalid combos. Some pizzas are only offered in large sizes, for example.
‍
Given this and the large amount of menu options offered by some of our partners, I designed a progressive disclosure scheme for the order entry pane that would only display valid options given previous choices.

Selecting a size and dough type will eliminate incompatible options

Selecting a pizza type will display options for toppings

05 ▸ Maintenance and Cleaning

Compromising between freedom and structure with our utility wizards

The workflow with the most involved human-machine interaction is cleaning. The system has four modules that have to be cleaned with a mix of manual tasks and automated functions.

The first concepts I tested locked users into a module-by-module maintenance flow as an attempt to enforce strict procedure. These were not well received

Again performing work studies, we saw that each flow had large time gaps where a user was waiting for an automatic process to finish. We also learned a worker's basic mental model for cleaning the Station did not separate the functional modules: users would clean ALL the components they could of the same type, THEN move on.

I planned out a new scheme with less reliance on the boundaries of the modules that still enforced a step order.

"Lane stacks". One stack for each module. Each stack item = one action to perform

The stacks also had the quality of naturally displaying progress along each workflow. An empty stack was a clear indication of completeness

Users had freedom to use their discretion to complete tasks for ANY module as long as they were performed in stack order. While waiting for an automatic cleaning function to wrap up in one module, they could start cleaning another.
‍
This scheme was very well received and cut down on cleaning time by a factor of 50%.

06 ▸ Impact

Market reception to our interface

Fully-featured Expo was critical to Picnic’s expansion¹, allowing us to partner with more kitchens than ever before without ballooning our integration costs. Ease-of-use scores were immensely favorable and new user training time fell from a multiple days to a couple of hours. Ongoing improvements to our order management feature let workers devote uninterrupted time to operating the Pizza Station, realizing throughput gains of 70%+ over the initial user experience.

1. In 2022, after the launch of Expo V2, we raised a successful Series A funding round of $13m after demonstrating revenue.

07 ▸ Final Words

Working at a robotics startup as a designer

Working in HMI is the most complicated thing I've done in my career. Becoming too familiar with the technical aspect of how a physical system works can make you LESS familiar with the perspective of a user. In every instance of being lost or stuck on an aspect of interface design, a better direction was uncovered by going back to the people we were trying to serve, working with them, and becoming advocates of their perspective.

I also gained a deeper appreciation for working cross-functionally. Here at Picnic, in order to realize a specific vision, the impetus was on me to communicate a shared technical vision for MANY stakeholders: mechanical, software, marketing, data, food science, and supply chain management. A successful, quality design was one that made sense from all our perspectives, with concessions traded in order to focus on our main goal: a quality product for our users, of which we all became designers.