Labor is the bottleneck
of logistics.

We're building the robots that remove it.

Oli Robotics develops bimanual AI robots for 3PL warehouses — sorting parcels, picking orders, processing returns. Trained on proprietary teleoperation data. Deployed as a service.

Based inSan Francisco
StageSeries Seed, closing Q2 2026
Fleet24 robots, 6 sites
HiringEngineering, Ops, BD

$1.5 trillion moves through 3PL every year.
70% of the cost is human hands.

Third-party logistics is the backbone of global commerce — and its margins are being crushed by labor it can't hire fast enough.

The industry doesn't need another conveyor belt. It needs hands that think.

100%+
Annual turnover · US fulfillment
+35%
Wage growth since 2020
Flat
Productivity · same period
$1.5T
Annual 3PL spend · 70% labor
// CHAPTER 03

Bimanual robots, built for the messy 80% of warehouse work.

// THE SYSTEM

The hardware stack.

Built for the messy 80% of warehouse work — mixed SKUs, poly bags, returns, irregular shapes — that conveyor automation can't touch.

01
Two arms
Because one hand can't open a bag while the other picks from it. Bimanual coordination at human-scale speed.
BIMANUAL
02
Force-feedback teleoperation
Operators train the robot through demonstration, not code. Haptic gloves capture intent, not just trajectory.
HAPTIC
03
On-device inference
Jetson Thor runs vision-language-action models locally. No cloud round-trip. Sub-50ms decisions.
EDGE
04
Stereo perception
ZED depth cameras for sub-millimeter grasp planning across reflective, deformable, and irregular geometry.
STEREO
ZED·STEREOJETSON·THORVLA · π0 / GR00T2000·TOPS·128GBTELEOP·GLVFORCE·FEEDBACK6·DOF · 200Hz
// CAPABILITIES — DEPLOYED TODAY
T·01
Parcel sorting
by destination, weight, fragility
T·02
Pick-and-place
across mixed SKUs and pack types
T·03
Returns processing
unpack · inspect · re-shelve
FLEETPOLICY·v3.4↻ 6h cadenceSTAGE 01TELEOPSTAGE 02TRAINSTAGE 03AUTOPKT

Every shift makes the fleet better.

Traditional robots ship with fixed capabilities. Ours ship with a learning loop — and the moat compounds with every grasp.

01

Teleoperate

Human operators handle edge cases through force-feedback gloves. Every motion becomes labeled training data — joint trajectories, contact forces, intent.

T+0
02

Train

T+12h
03

Autonomize

T+24h
// COMMERCIAL TERMS

Pay for output, not hardware.

No capex. No integration teams. No six-month deployments. You staff your warehouse the way you'd scale a SaaS license.

RAAS · TERM SHEETv3.4 · 2026
01
Monthly subscription
per robot · per shift
02
Performance SLA
throughput · accuracy guaranteed
03
All-inclusive
hardware · software · support
04
Elastic capacity
scale up for peak · down after
Break-even vs. laborMonth 4 · most US metros

Three curves just crossed.

The window to build the category leader is open. It won't stay open long.

CURVE · 01
MODELS
20182026

Foundation models for robotics work.

VLA architectures (π0, GR00T, RT-2) cleared the generalization barrier in 2024–2025. Robots can finally learn tasks instead of being programmed for them.

CURVE · 02
COMPUTE
20182026

Compute fits on the robot.

Jetson Thor delivers data-center-class inference at the edge. No cloud round-trip, no latency tax — sub-50ms decisions on-device.

CURVE · 03
LABOR
20182026

Labor economics broke.

3PLs are paying signing bonuses for warehouse pickers. The unit economics of RaaS now beat human labor in most US metros — for the first time ever.

// THE TEAM

We've shipped robots that worked in the wild.

Before Oli, our team built and deployed an autonomous robotic café in San Francisco — a fully bimanual system serving real customers, viewed by 4M+ people on social media.

We learned what physical-AI startups usually learn the hard way:

  • L·01Hardware reliability is the product.
  • L·02Data quality beats data quantity.
  • L·03The last 5% of autonomy takes 95% of the work.

We're applying those lessons to a market a thousand times bigger.

Logistics is about to be rebuilt.