About Us
Pickrook Robotics
We build the perception and grasp intelligence layer that makes fixed-station arms commercially viable at 3PL scale without robotics engineers on staff.
Our Story
How Pickrook started
Zoe Marchetti spent two weeks benchmarking pick performance at a Pittsburgh-area 3PL in fall 2021 as part of a Carnegie Mellon Robotics Institute field study, watching a highly skilled human picker consistently outpace a first-generation vision robot because the robot stalled for 3–8 seconds on any SKU it had not seen before in training — which happened on roughly 1 in 12 picks.
The bottleneck was not the arm hardware or the compute — it was that every deployed warehouse vision system in 2021 was a product-specific recognizer, not a general-purpose picker. The moment a SKU rotated or a new client onboarded, the system degraded until someone relabeled a dataset and retrained the model, which took days the 3PL could not afford.
Pickrook’s first attempt was a generalized grasp model trained on synthetic and real mixed-object datasets, deployed on a single KUKA arm at the same Pittsburgh 3PL under a paid pilot agreement in Q2 2022. The model handled novel SKUs with under 1% error rate on first encounter, which the facility manager described as a first in their experience with any vision system.
Zoe Marchetti (ex-Carnegie Mellon Robotics Institute research engineer) co-founded Pickrook with Felix Nduka (ex-Symbotic WMS integration) and Ingrid Sorenson (ex-KUKA Robotics application engineer). Two additional engineers from CMU and RPI complete the founding team, all with direct 3PL floor experience.
Mission
Make autonomous piece-picking reliable enough for 3PLs without a robotics team
Make autonomous piece-picking reliable enough that mid-size 3PLs can run it without robotics engineers on staff. Every design decision at Pickrook — from the edge-compute architecture to the operator-facing dashboard — is made to ensure the robotics expertise lives in the software, not the customer’s org chart.
The 3PL operators Pickrook serves compete on throughput speed and pick accuracy SLAs but face chronic labor shortages and high turnover in pick roles. Pickrook gives them the automation economics of a large enterprise distribution center without requiring the engineering team those economics traditionally demanded.
Focus
Piece-picking for mid-size 3PLs — that’s all we do
Pickrook is not a general robotics platform. Not a mobile robot. Not an enterprise warehousing suite. Pickrook is focused entirely on the perception and grasp intelligence layer that makes fixed-station robot arms commercially viable at mid-size 3PL scale without requiring robotics engineers on staff.
The company targets 3PLs running 3–12 fulfillment facilities at 50,000–500,000 sq ft per site, with $30M–$300M annual 3PL revenue. Pickrook is not designed for single-site owner-operator micro-fulfillment or for enterprise national 3PLs with in-house robotics engineering teams running proprietary hardware.
This narrow focus is deliberate. The piece-picking problem at this scale is hard enough to require a dedicated company. Pickrook will not widen its focus until the mid-size 3PL problem is solved completely.
Values
How we work
Generalization before specialization
We build systems that work on things they have never seen before. Product-specific recognizers are not a solution at 3PL scale; general-purpose pick models are.
Measured in picks per hour, not demos
Every pilot agreement includes throughput and accuracy benchmarks. If the system doesn’t hit them in production conditions, the contract doesn’t renew.
Operator-first deployment
Facility operators — not robotics engineers — should be able to install, configure, and maintain the system. The expertise lives in the software.
Error recovery as a first-class feature
Every pick system fails sometimes. The difference between a useful system and a liability is how gracefully and automatically it recovers without human intervention.
Honest about what the system cannot pick
No vision system picks everything. Pickrook surfaces its confidence thresholds transparently and flags items it cannot handle reliably rather than attempting and failing silently.
Headquartered in Pittsburgh, PA — close to the Carnegie Mellon Robotics Institute and the 3PL operations where the original problem was found.
Meet the founding team
Five engineers who have worked across CMU, Symbotic, and KUKA — and who know the 3PL floor firsthand.
View the Team