INTRODUCING THE TROSSEN PHYSICAL AI LINEUP.
Human demonstration, robot data, training and evaluation, real embodiment — one architecture across all four. The same arms, the same software, and the same data path from the first demonstration to the robot that does the work. We built a room in San Francisco to show the whole loop running, not a product table.
THE RESIDENCY
EXPERIENCE THE WHOLE PLATFORM FOR THE NEXT TWO WEEKS IN SF.
Teleoperation feel, station geometry, how a dataset actually gets recorded — those are questions to be answered in person. So we took a space in San Francisco and set the platform up end to end: an operator demonstrating on Glide, episodes recording through the SDK, and the same arms running on a bench and on a mobile base.
Come work a task with your own hands, bring a problem from your lab or line, or take the whole thing virtually if you can't get to the room.
THE WORKFLOW
FROM DEMONSTRATION TO DEPLOYMENT
Capture, training, deployment — our platforms cover all three, so none of them has to be the part you build yourself.
DEMONSTRATE
HUMAN DEMO
Capture operator intent cleanly — leader-follower teleoperation at the station, or handheld capture away from the robot entirely. Servo stiction and low-speed judder never make it into the recording.
COLLECT
ROBOT DATA
Synchronized, structured episodes — joint states, camera streams, and task events recorded together through one configuration-driven data layer rather than a per-rig script.
TRAIN & EVALUATE
TRAINING & EVALUATION
Inspect the dataset, convert it into the framework your team already uses, train, then evaluate the policy back on the hardware that recorded it — not on a proxy.
EMBODY
REAL EMBODIMENT
Move the same arms and the same software from a fixed bench to a mobile base. Changing the embodiment shouldn't mean rebuilding the data pipeline underneath it.
THE LINEUP
SIX PRODUCTS. ONE PHYSICAL AI ARCHITECTURE.
Rivet moves, Workbench fixes it in place — both on the WidowX arm, in three configurations: Core, Pro, and Pro Heavy. Glide captures the operator's intent; Cockpit supports the operator. Pricing below is proposed and pending final approval.
.png)
SOFTWARE, CLOUD & ECOSYSTEM
THE HARDWARE IS JUST THE BEGINNING
The reason the same task doesn't have to be re-plumbed at every stage is the software layer underneath it. It ships with the hardware, it's open, and it's built to hand your data to whatever you already train with.
TROSSEN SDK
The open-source data layer. Configuration-driven episode capture that records joint states, camera streams, and task events together, then converts them into the format your training stack expects.
-
Synchronized multimodal episode recording
-
MCAP episodes, Foxglove inspection
-
Supported LeRobot conversion path
-
Hardware- and model-agnostic by design
-
Preconfigured workflows and training examples
TROSSEN CLOUD
The service layer for teams running more than one system: centralized robot-data storage plus the fleet, remote, and lifecycle workflows that multi-site programs keep having to build for themselves.
-
Centralized storage for collected datasets
-
Device identity, configuration, and health direction
-
Remote diagnostics and support direction
-
A path from collection station to training workflow
-
Built across the portfolio, not bolted to one robot
ECOSYSTEM
We'd rather be the documented hardware entry point into the tools you already use than ask you to commit your hardware budget to one policy architecture.
-
Versioned docs, API reference, and tutorials
-
Public code repositories and sample datasets
-
Named integrations with published test status
-
Support tickets, downloads, and knowledge base
-
No software feature walls between arm tiers
WHY TROSSEN
YOUR PHYSICAL AI PARTNER, FOREVER AND ALWAYS
We believe you're building the next wave of human automation, and we're here to support the future you're working toward.
SPEED
One development environment from research through deployment. Arms arrive with the gripper, power, cables, software, and preconfigured workflows — the integration work you'd otherwise do before your first useful dataset is already done.
TRUST
We publish what was tested, not what might be possible. Every integration is listed with the exact configuration and test date behind it, and we state the boundaries plainly — these are research and supervised-deployment systems, not safety-rated cobots.
SERVICE
Designed for access and replacement, with documented service procedures rather than a return-to-factory policy. Docs, tutorials, downloads, and a support path are part of the product, not an upsell.
ECOSYSTEM
Open-source SDK, ROS 2, LeRobot, public repositories, versioned documentation, and sample datasets. Your data belongs to your training stack, not to ours.
DEPLOYABILITY
The same arms and the same software from a fixed bench to a mobile base, with volume pricing and a supported path from evaluation to pilot. Scaling up shouldn't mean starting over.
ONE ARCHITECTURE
Core, Pro, and Pro Heavy share a controller, a software direction, and a data path. Choose the task envelope you need without resetting your toolchain or retraining your team.
REQUEST ACCESS
REQUEST A PRIVATE DEMO.
Tell us what you are building and what you want to see. We will route you to the right demo block, product workflow, or partner event.
BEST FIT FOR PRIVATE DEMOS
Robotics startups, data collection companies, research labs, enterprise R&D teams, investors, existing customers, and strategic partners.