Runs on the robot
No cloud round-trip — deterministic, sub-cycle latency at the edge.

On-robot edge runtime for low-latency closed-loop manipulation.
An optimized on-robot runtime that loads a compiled GripSim policy and world model and runs the perception→prediction→actuation loop at control-loop latency, with fleet updates and safety limits.
Built to be GPU-essential, NVIDIA-aligned, and production-ready.
No cloud round-trip — deterministic, sub-cycle latency at the edge.
Models compiled for the target Jetson for speed and efficiency.
Push new policies across a fleet with NVIDIA Fleet Command.
Built-in guards on force, speed, and workspace bounds.
A simple interface for the OEM's controller to call.
Handles real-time vision and tactile streams on-device.
| Target HW | NVIDIA Jetson Thor / Orin |
|---|---|
| Optimization | TensorRT + CUDA |
| Latency | Control-loop rate (target sub-cycle) |
| Fleet | NVIDIA Fleet Command OTA |
| Streams | Holoscan / DeepStream |
| Delivery | NVIDIA AI Enterprise |
Start free. Scale to a fleet. Pricing shown is indicative, pre-launch.
Cross-embodiment VLA manipulation foundation model + fine-tuning toolkit.
Explore Dextra →
Tactile-and-vision contact world model that predicts grasp outcomes.
Explore Tactyl →
Domain-randomized manipulation simulation + data-generation engine.
Explore Simdex →The manipulation layer teams ship with, not a research demo.
One policy runs across many grippers and hands — no per-robot rewrite.
Tactyl predicts slip and force so grasps react before they fail.
Reflex executes on Jetson at control-loop latency, with safety limits.
Every deployment feeds the shared model through the data flywheel.
Benchmarks, sim-to-real notes, and product updates. No hype, no spam.
Tell us your robot and task — we'll scope a pilot.