Differentiating in Point to Point Mobility

Renderings of TickTock’s lightweight mobility platform for retail applications. Shown hanging clothes and carrying cardboard boxes, plastic totes, and shopping baskets.

[Note: This is Part 3 in a five-part series on TickTock’s robots: P1, P2, P3, P4, P5]

Robot arms have always garnered attention for being used in manufacturing, but there’s an often overlooked segment regarding robots that simply move from point A to point B. The Robot Report lists over a dozen players in this space, and notes how many took off after Amazon’s acquisition of Kiva Systems for $775M in 2012. It’s a fast growing market and TickTock wanted to break into this space with something new and different.

Our initial focus on the consumer robotics market drove us to create a unique software platform that ran on super low cost components. We also needed robots that worked right out of the box without complex setup or training. The key was finding an application that most benefited from this advantage, and that’s where retail stores really shined.

Back of house vs customer facing

We found a trend that quickly sorted commercial robots into two categories: Ones that only worked around trained employees, and those that dared to work around the public.

The Fetch Freight, Otto, and 6 River Systems robots represent the common tower, flatbed, and cart form factors found in industrial mobile robots today. The OSHbot from Fellow Robots, Knightscope K5, and Relay from Savioke, represent some of the new uses for robots that work around the public.

While warehouse and manufacturing robots performed the same tasks all day in fairly static environments, retail stores are in constant flux. The layouts change, foot traffic is unpredictable, and the tasks being formed differ from hour to hour.

The chaos of the holiday 2017 shopping season was the perfect backdrop to go study this space and we saw tons of wasted labor moving goods around. The logistics clearly differed between grocery, clothing, and big box retailers, but we knew a mobility opportunity was there and set out to built a prototype to prove it.

Our big differentiation was in price and UX

We needed a form factor that could carry physical products and began with a scaled up home butler concept that we called TickTock Pro. It would offer goods transport in a lightweight and easy to use package that was designed to work safely around customers.

Early animation of a TickTock Pro mobility concept robot with adjustable shelving that could be stowed at the base. The three wheel design with off-center drive was later scrapped for a traditional center-mounted differential drive in TickTock One.

Making it cheap was the easy part, thanks to the six months of significant effort we already put into our unique robotic software stack. TickTock’s code ran on Android and used commodity mobile sensors, which dropped the need for an industrial grade x86 Intel PC and LIDAR. This easily cut $2,000 or more from the unit cost compared to anything out there.

The top row of components are low cost and more commonly found in your smartphone. The lower row are more expensive, physically larger, and draw more power. This new architecture led to TickTock creating a custom robotics stack for Android, rather than use the more common Linux and ROS framework.

As a benchmark comparison, you can get an electric scooter for under $200 that will carry a 200lb human for many miles on a single charge. Then add in a few hundred dollars for the components from an AR-capable smartphone to serve as the brains and sensors. Together, these lighter components also reduced overall weight and power consumption, which has a ripple effect in lowering total cost.

Our first prototype cost less than $2,500 in single unit quantities and was made from off the shelf parts sold at mark-up. We felt confident that the BOM + assembly costs would be below $1,000 in volume, and easily manufactured by any contract manufacturer currently making scooters and smartphones.

Testing a tower robot design

We spent several weeks measuring common retail product package sizes and their weight, plus a look at the tight spaces they needed to move through. A clear need emerged for a tall tower that could hang clothing or carry 18"x 24" boxes at a total capacity of around 200lb.

To test this theory, we started with an Arlo mobility kit from Parallax, and framed it with some 80/20 that held cafeteria trays for shelves. Sandbags of various weights tested different payload configurations to ensure it wouldn’t tip and we attached a second Android phone as a screen to display a list of map destinations.

Rapid prototype of a tower robot design using a Parallax wheel and motor kit with 80/20 framing for structure. Shelves were adjustable and sandbags could be placed at different heights to test stability.

TickTock’s software ran on an Android Asus Zenfone AR, and we just needed a USB driver to talk with the Parallax kit’s micro-controller at the base. A little bit of local planner tuning and we were ready to test in a store.

A trial run through Home Depot’s kitchen area where a destination is given by touching the top display. An inset shows the view from the mobile augmented reality ARRviz app.

Our ARRViz app (short for “augmented reality robot visualization”), let us see what the robot was thinking as it navigated the narrow aisles of Home Depot’s kitchen section.

The tropical plant section caught us by surprise since the layout changed entirely between the day we mapped the store and when we brought in the robot. Thanks again to ARRviz and our use of Android, it only took a minute to pull out a phone and remap it.

The TickTock prototype was driven through the tropical plants section to test a diverse set of visual landmarks and obstacles. It then drove out the back door to the outside garden area.

The making of a looks-like / works-like model

Bryan De Leon took inspiration from hydrofoils used on racing ships as they smoothly cut through choppy waters. This tall and open look gave the impression of something fast that wasn’t bulky and intimidating.

As the team validated the overall dimensions, Bryan De Leon was busy working on the industrial design for something we could show off to retailers. We wanted to give it a unique look from a typical cylindrical “trashcan robot”, and took inspiration from hydrofoils and tall sailing ships.

A modular shelving system stowed the decks in the base for quick access when needed. An accessory slot opened on the side where additional sensors such as cameras or RFID readers could be plugged in via USB.

This new designed became known as TickTock One, and had adjustable flat shelves for products and included a hanger rod for clothing retailers. Some lights were added as a conversation starter since the Human Robot Interaction (HRI) design for mobile robots hasn’t become standardized yet. We’ll need this eventually so that everything from sidewalk robots to self-driving cars can let users know which way they’re going.

This call-out diagram for TickTock One shows where the hanger rod, accessory mount, vision system, and mechatronics were located.

I’ll be honest, the blue LEDs just looked cool ;-)

Ten different sensors and systems would combine to provide overlapping safety coverage.

We also needed to pack a lot of sensors in the system for safety. The initial prototype used a combination of color, depth, and ultrasonic sensors to avoid obstacles. We also tested a low cost radar that was terrific at seeing optically challenging materials and things moving quickly.

The Parallax wheel and motor kit fit snug inside the 3D printed base of TickTock One.

The electro-mechanical guts inside TickTock One used the same Parallax kit and Asus Zenfone AR as our proof of concept. The housing was all custom though and 3D printed by the amazing folks at Fathom in Oakland, and we assembled it all in our office in Santa Clara.

John Moretti (left) and Bryan De Leon (right) assembling the split-halves of the TickTock One tower. It was too large to print in one piece and needed to be split apart by Fathom and glued together in our shop.

Time to go for a spin

TickTock’s office was carved off in a corner of the OLogic building in Santa Clara, where Ted Larson’s team makes the electronics for most Bay Area robotics companies. It’s not quite an “incubator”, but was a great place to start our company and very amenable to robots running around.

TickTock One with a sample payload of printer paper making the rounds in the OLogic/TickTock office in Santa Clara, CA. This low angle view shows the negative space in the design and blue lighting effects.

Office testing is great but we needed to learn more from the real world: Is the software capable of navigating a large retail store? Can it hold real products? Do employees or customers freak out with a robot running around them?

No one seems to mind shopping around a mobile autonomous robot as TickTock One cruised Safeway’s aisles to transport a variety of product types across different sections of the store.

TickTock’s mapping and navigation worked fine and gave us confidence that Google’s AR Core + VPS was indeed the right SLAM solution (more on the tech in a future post). We also found that no one seemed to care that a robot was roaming down the aisles. Maybe that’s because we’re in Silicon Valley, or maybe it’s just 2018 and people expect to see robots?

Where we were weakest was in payload size and ergonomics. The tower needed to expand on all sides to more easily accommodate boxes. It also needed slanted shelves to make it easier to pull products from boxes without taking the boxes off the robot. That would have to come in the next iteration of the tower design, but we also wanted to try a cart.

A Rubbermaid bot seems long overdue

You’ve probably noticed the proliferation of Rubbermaid carts in offices, stores, and commercial spaces. There are dozens of different designs for them and plentiful distribution options such as ULINE. So we spent a weekend hacking one into a robot with the same software and Parallax kit down below.

Our friends at Go Engineer let us simulate an outdoor building-to-building transport use case with a Rubbermaid cart turned into a robot. This is a very real example for them as boxes of 3D printing supplies are stored in one building, and later needed in another building nearby where rows of additional printers are.

The cart design is what Canvas Technology and 6 River Systems use, while towers are more common among Fetch Robotics and Locus Robotics. We’ll probably never have “general purpose robots”, and robot companies will always need multiple form factors to match the application. Carts and towers seem like a good start and we’re starting to see “flatbeds” as a replacement for pallets too.

Robotics startups are tough

TickTock’s focus was on our software but we were indeed a hardware startup. That made fund raising a challenge, especially for the retail space that is known for exceptionally long sales cycles.

We’re starting to see new companies offering robots for autonomous floor mops, and others toting camera systems to take inventory. Moving physical products around retail stores was going to be a new concept and that meant the metrics were still unproven. The retailers we spoke too were excited by idea and gave great feedback about TickTock One, but we raised too little capital to keep the effort going.

I’ll share more on the inventory replenishment metrics we observed in a future post and hopefully we’ll see someone else fill the need. We know that retail isn’t going away and the experience will only get better when mundane tasks can be automated so that associates can help customers instead of pushing carts around from A to B.

[This is Part 3 in a five-part series on TickTock’s AR-powered robots. Be sure to check out Amazon’s Echo Show on Wheels, Consumer Robot Concepts, Low Cost Mobile Robots Using Android, Robots for Retailers, Augmented Reality and Robotics Overlap]