Kiva Systems
Amazon Robotics

Kiva Sys­tems start­ed with the idea that prod­ucts should come to peo­ple, not the oth­er way around. I built the sys­tem archi­tec­ture and adap­tive motion lay­er that made that vision pos­si­ble — a foun­da­tion that now pow­ers Amazon’s mil­lion-robot net­work worldwide.

Kiva Sys­tems began in 2003, when I was on sab­bat­i­cal from Cor­nell, work­ing with entre­pre­neur Mick Mountz and pro­fes­sor Pete Wur­man (then on sab­bat­i­cal from NC State) to rethink how goods move inside ware­hous­es. Instead of peo­ple walk­ing to prod­ucts, our insight was to have prod­ucts come to peo­ple: fleets of small, mobile robots that lift­ed entire shelv­ing “pods” and brought them to sta­tion­ary pickers.

My role was to shape the core sys­tem archi­tec­ture, devel­op the algo­rithms for nav­i­ga­tion, fleet coor­di­na­tion, and real-time con­trol, and over­see the robot design to ensure it fit seam­less­ly into the over­all sys­tem. I built the entire motion lay­er as one large, cou­pled adap­tive sys­tem. New robots didn’t just ship straight from assem­bly to a customer—they first spent about an hour run­ning in our test facil­i­ty, learn­ing how to oper­ate in a ware­house envi­ron­ment, to han­dle pods that could weigh any­where from just a few pounds to over a thou­sand with wide­ly vary­ing mass dis­tri­b­u­tions, and to adapt to man­u­fac­tur­ing variabilities—which were sig­nif­i­cant because we delib­er­ate­ly kept robot costs low. Much like a fawn find­ing its legs, each robot devel­oped the abil­i­ty to move con­fi­dent­ly and pre­cise­ly under dif­fer­ent con­di­tions. Once deployed, each robot con­tin­ued to adapt in real time, com­pen­sat­ing for things like wear and tear to main­tain peak per­for­mance. Just as impor­tant­ly, they shared what they learned: nav­i­ga­tion was based on phys­i­cal fea­tures on the floor that weren’t known pre­cise­ly and could change over time, so when one robot refined its under­stand­ing of a fea­ture, that infor­ma­tion was imme­di­ate­ly shared with the whole fleet. This con­stant adap­ta­tion and col­lec­tive learn­ing were a key part of the “secret sauce” that made Kiva work.

The results were dra­mat­ic. Kiva deploy­ments col­lapsed walk­ing time, increased stor­age den­si­ty, and enabled ful­fill­ment sites to scale through­put quickly—a com­pelling propo­si­tion as e‑commerce surged. In March 2012, Ama­zon agreed to acquire Kiva Sys­tems for approx­i­mate­ly $775 mil­lion in cash—nearly one per­cent of Amazon’s mar­ket cap­i­tal­iza­tion at the time—making it the company’s sec­ond-largest acqui­si­tion to date, after Zap­pos in 2009 ($1.2 bil­lion). At that point, Amazon’s three largest acqui­si­tions had all been linked to Kiva: Zap­pos and Quid­si (par­ent of Diapers.com) were both major Kiva cus­tomers, and now Kiva itself was join­ing Ama­zon. Ama­zon had seen first­hand how our tech­nol­o­gy could accel­er­ate ful­fill­ment and deliv­er a com­pet­i­tive edge—insight that I believe strength­ened the case for bring­ing the capa­bil­i­ty in-house. Kiva’s head­quar­ters and engi­neer­ing tal­ent became Amazon’s inter­nal engine for mobile ful­fill­ment automa­tion, and the tech­nol­o­gy stopped being sold to out­side cus­tomers as Ama­zon focused on build­ing its own inter­nal advantage.

In 2015, Ama­zon rebrand­ed Kiva Sys­tems as Ama­zon Robot­ics, for­mal­iz­ing the group’s role as the company’s inter­nal robot­ics orga­ni­za­tion. Under that ban­ner, the orig­i­nal Kiva “dri­ve” robots evolved into a grow­ing port­fo­lio deployed across a rapid­ly expand­ing glob­al foot­print of ful­fill­ment and sor­ta­tion cen­ters. The DNA of the Kiva approach remained vis­i­ble: tight­ly inte­grat­ed hard­ware, soft­ware, and fleet-lev­el orches­tra­tion, opti­mized for dense stor­age and human–robot inter­ac­tion on busy floors.

Although I moved to ETH Zürich in 2008 and tran­si­tioned to chief tech­ni­cal advi­sor at Kiva, the dis­trib­uted con­trol philoso­phies and scal­able coor­di­na­tion meth­ods I helped devel­op remain at the heart of the sys­tem. Those ideas are exact­ly the kinds of inno­va­tions that make mil­lion-robot fleets pos­si­ble, and Kiva served as the com­mer­cial prov­ing ground where they reshaped an industry.

The scale tells the sto­ry. A lit­tle over a decade after the acquisition—and two decades after Kiva’s founding—Amazon announced in June 2025 that it had deployed its one-mil­lionth mobile robot across more than 300 facil­i­ties. While Ama­zon now fields sev­er­al mobile robot fam­i­lies, the vast major­i­ty remain Kiva-style “dri­ve” sys­tems, includ­ing new­er vari­ants like Pro­teus. Based on Amazon’s own 2016 esti­mate of rough­ly $22 mil­lion in annu­al oper­at­ing sav­ings per Kiva-equipped site, scal­ing to today’s fleet sug­gests $6–7 bil­lion in year­ly savings—about 8–9% of the company’s 2024 oper­at­ing expenses.