[Part 1, Part 2, Part 3, Part 4]

Around 2012, six years after Harvest’s founders left iRobot in search of the next big thing, we reached a perplexing impasse.  We’d discovered an application that aligned wonderfully with robots’ abilities, we had built a robot that worked well even amid challenging circumstances, and we had customers across the nation using our robots every day.  But that wasn’t enough for a venture capital backed company.  To support our 30+ employees and our investors’ expectations, we needed to grow rapidly.  Unfortunately, we were growing quite slowly.  

While it was still financially possible, we needed to try something else.  

Warehouses

In the olden days “pallet-in, pallet-out” was the warehouse rule.  A pallet loaded with a single type of product would arrive from a manufacturer and be stored at a designated spot in the warehouse.  Later, that pallet would be sent out to a retailer.  Forklifts featured prominently in the process.  Logistics were simple and efficient with businesses on both ends.  Then the internet changed everything. 

The power of the new medium enabled shoppers to skip the local brick and mortar retailer.  Now, from home, they could browse a website and have desired items shipped directly to them.  That required revamping operations in the warehouse (from where the products were now shipped).  Pallets had to be unpacked, individual item stored on shelves, and workers had to cruise the aisles to collect and then pack for shipment whatever the customer had ordered.  

Webvan, an internet grocer, was an early mover in this brave new world of e-commerce.  Webvan secured close to $1 billion in investment, built a warehouse to support their business model, and then declared bankruptcy after just three years.  One of the reasons for Webvan’s demise was the high cost and low efficiency of the warehouse part of their business.  

A director at Webvan, Mick Mountz, saw the problem up close.  Following the conventional approach, workers simply couldn’t pick orders fast enough to make the operation profitable.  Motivated by Webvan’s untimely end, Mick sought a better way.  A radical idea occurred to him.  Instead of having the workers walk to the products, what if the products came to the workers—transported by robots?  This would greatly improve the productivity of each worker.  Eliminate the walking and orders could be fulfilled two to four times faster.   

To implement that idea, Mick and two others founded Kiva Systems.  They spent several years developing an order fulfilment system—it relied on a fleet of strong, squat robots.  The robots, called drive units, retrieved movable shelving units called pods, on which the products were stored.  Robots brought pods, one at a time, to a worker who would pick the ordered product from a bin on the pod and place it in a shipping box.  The robot then returned the pod to the storage area.  An all-knowing piece of software, the warehouse control system (WCS), directed and kept track of all the robots, all the pods, and all the products.  

Convincing warehouse owners to try the system proved difficult. It required that they abandon their current practices, remove all their existing shelving, and reconfigure the warehouse.  A dense grid of unique barcode stickers had to be positioned carefully on the floor.  By observing the stickers with downward-pointing cameras the robots could continuously track their precise locations.  Dedicated pick/pack stations around the periphery of the product storage area were the only places staff would interact the system to pick orders and replenish inventory.  

After a slow start, Kiva became a glowing success, attracting much business.  Amazon took notice.  They purchased Kiva Systems for $775 million in 2012 and began to build new and convert old warehouses to use the revolutionary system.  (Simultaneously, they pulled the rug out from under the other vendors Kiva had sold systems to.  Now, a critical tool those vendors needed for their operations was owned by their biggest rival.  Awkward!)

Kiva’s success captivated the robotics world.  Soon, companies attempting to build warehouse robots of various sorts started popping up all over.  Warehouses also appeared an attractive pivot to us at Harvest.  We had developed a robot that picked up and moved items around nursery farms and greenhouses; we were sure we could build a warehouse robot on the same basic platform.

Of course, we couldn’t lug pods around, we’d have to find a technique different from what Kiva had done and perhaps target a different part of the market.  One problematic aspect of Kiva was its all-or-nothing approach—existing warehouses had to be gutted and reconfigured.  Scrapping hundreds or thousands of existing selves, buying more expensive transportable pods to replace them, and exchanging well-known (if inefficient) processes for unfamiliar robot-based ones was daunting for many warehouse owners.  But what if there was a much lower cost, less frightening way to improve worker efficiency?  

We figured out how warehouses could keep their existing shelving and adopt automation at their own pace, without having to leap into the unknown as Kiva demanded.  Our idea was that workers would still pick items from existing shelves, but they would never need to walk back to the packing station.  

The Harvest system initially worked like this:  A customer would place an order for one or more items.  The WCS would locate the first item on the shelves and dispatch a robot carrying a tote (an open-topped plastic box) to the right point in the warehouse.  A worker would get a message directing him or her to rendezvous with the robot.  (The system would ensure that the robot arrived first, so the worker never had to wait for the robot.)  The worker would then place the ordered item in the robot’s tote and then both would move on to their next assignment.

If there were only one item in the order, the robot would next take the tote to the packing station.  But if additional items were needed to complete the order, the robot would carry the tote to the proper location and rendezvous with another worker to collect the next item.

Later we devised a refinement.  Robot and worker didn’t actually need to be at the same place simultaneously.  Instead, the robot could drop off a tote near the spot where the ordered item was located. The worker could then arrive at any time and place the item in the tote. This done, the worker would alert the WCS that that “pick” task had been completed.  Another robot would then pick up the tote and carry it to its next destination.  Such asynchronous rendezvous required fewer robots and kept them in constant, productive use.   (See the system, rebranded as NextShift, in action here.)

It seemed to us that our approach would be attractive to smaller or more risk-averse warehouses.  We’d appeal to any customer that wanted the advantages of automation but balked at the palpitation-inducing hurdle of throwing away everything familiar and starting over.  In contrast to Kiva, our customers could ease into automation by adopting our system a few robots at a time.

WCS

The new warehouse robot we built was designated the TM-100.  It used cameras to recognize barcodes attached to the shelves.  It also used the same type of laser scanner as the HV-100s to detect obstacles and to measure the distance to shelves and other structures.  This combination of barcodes and range measurements enabled each robot always to know where it was.  While constantly reporting its position to the WCS, the WCS would tell the robot where to go next.

The robots needed to do very little of their own thinking.  Only if an unexpected obstacle was detected (if say, a product accidentally fell into the path of a robot or a worker got in the way) would the TM-100 halt or divert from the WCS-commanded path on its own initiative.

Thus we would rely on the new WCS we were developing to do most of the heavy lifting.  It would optimize efficiency by coordinating many moving parts.  To build that clever code we recruited one MaryEllen Sparrow to be our Director of Software Development.  Earlier jobs in the industry had granted MaryEllen crucial experience in that area.

Although I’d helped move things in the warehouse direction, I found the change in emphasis from robots to software less than compelling.  For me, robots are more fun when they are the stars of the show—when they have to deal with real world surprises.  But, by design, surprises were mostly absent in the warehouse setting.  That was another reason I was looking for other options in 2015 (see Hatching Tertill).

Harvest put together what I thought was an impressive system.  Robots scurried about our demo warehouse.  Workers picked products into totes, and the robots quickly delivered them to the right places.  We built the efficient, low-cost alternative we had intended.  

The only thing missing was paying customers. 

Restructuring

A familiar Yogi Berra-ism goes, “If the people don’t want to come out to the ballpark, nobody’s gonna stop them.”  And similarly for us, no one was going to stop customers from not buying our robots.  

That was a problem because we needed the income.  In anticipation of rapid growth, Harvest had earlier taken on about $2 million in venture debt.  When we became unable to pay back that debt on schedule, the lending bank took over ownership of the company.  That wiped out our investors.  We founders shared the same fate as we were also investors.

At that point, the bank could have simply sold off Harvest’s assets to recoup some of what we owed.  But when such a liquidation happens, very little cash is generated.  Who would want to buy the sophisticated robots in our inventory if the company that built them no longer exists to support them?  Our assets would be much more valuable if the company could somehow continue operating. 

That was the enlightened approach Harvest’s bank took.  MaryEllen Sparrow on the warehouse side, and Charlie Grinnell on the nursery and greenhouse side still believed in the products they had develop.  And both wanted to give them more time to succeed.  The bank allowed each to buy the assets the bank now owned.  In that way the bank recouped some of their losses and both parts of the company continued to operate.  

The warehouse portion of Harvest split off to become NextShift Robotics, led by MaryEllen.  And Charlie once again became Harvest’s CEO.  Harvest had shrunk from over 30 people to three but that was enough to continue serving our existing customers.  (Charlie felt very strongly about not leaving them in the lurch.)  Now, with the investors gone, no debt, and a good inventory of parts, Harvest was small but profitable!

No longer a startup with unicorn potential, Harvest settled back to being mostly a service company.   But, with great care, Charlie husbanded the company’s income thus retaining at least a small possibility of developing a new breakthrough product.  Our attempt to do just that will be the subject of a later post.