D-01
Agriculture
Selective harvesting, scouting, and precision intervention across row-crop and orchard environments.
Live · Telemetry Stream
Lat 47.6062° NLon 122.3321° WFrame 0000
◖ Strang Research / v0.9 Foundation
Autonomous intelligence
for the working world.
We build general-purpose foundation models that let industrial robots see, reason, and act in unstructured environments — from greenhouses to gigafactories.
VISION-LANGUAGE-ACTION◆6-DOF MANIPULATION◆DEXTROUS GRASPING◆MULTI-AGENT COORDINATION◆WORLD MODELS◆IMITATION LEARNING◆SIM-TO-REAL TRANSFER◆WHOLE-BODY CONTROL◆DIFFUSION POLICIES◆ON-DEVICE INFERENCE◆VISION-LANGUAGE-ACTION◆6-DOF MANIPULATION◆DEXTROUS GRASPING◆MULTI-AGENT COORDINATION◆WORLD MODELS◆IMITATION LEARNING◆SIM-TO-REAL TRANSFER◆WHOLE-BODY CONTROL◆DIFFUSION POLICIES◆ON-DEVICE INFERENCE◆
01 / Mission
A foundationfor matter.
Software learned to reason. Now matter must. Strang is building the underlying intelligence that lets a single robotic platform pick fruit at dawn, assemble a drivetrain by noon, and palletize crates by dusk — without rewriting a line of code.
Founded
2026 · San Francisco & Bengaluru
02 / Capabilities
Six primitives. One stack. Every embodiment.
C-01●
Perception
Multi-modal fusion across RGB, depth, force-torque, and proprioception running at 240Hz on-device.
C-02●
Reasoning
Hierarchical world models that plan over long horizons and recover gracefully from physical perturbations.
C-03●
Manipulation
Sub-millimeter dextrous control learned from 11M hours of teleoperation and synthetic rollouts.
C-04●
Mobility
Whole-body locomotion policies transferred zero-shot across wheeled, tracked, and bipedal morphologies.
C-05●
Coordination
Decentralized fleet learning — robots share experience without sharing data.
C-06●
Safety
Constraint-aware policies with formal verification of joint-space envelopes and human-proximity guards.
STRANG-FM · 18.4B params◖ inferring
03 / The Model
A single brain trained on the physics of work.
Strang-FM is a vision-language-action transformer trained on the largest corpus of robotic interaction ever assembled — over 4 petabytes of teleoperation, kinesthetic demonstration, and high-fidelity simulation.
- 18.4B
- parameters
- 4.2 PB
- training corpus
- 240 Hz
- control loop
- 27
- robot morphologies
04 / Domains
Trained in simulation. Hardened in the field.
D-02
Assembly
Sub-millimeter electronics assembly with adaptive force control and online quality inspection.
D-03
Manufacturing
Multi-robot lines that re-program themselves around new SKUs, defects, and material substitutions.
D-04
Logistics
Mixed-SKU palletizing, depalletizing, and case picking at warehouse-scale throughput.
◖ Limited Pilot Access · 2026
Bring Strang to your factory floor.
We partner with a small number of industrial operators each quarter to co-develop and field-validate new capabilities. Tell us what your robots need to learn.