Integrations

Architecture

Fits into robotics stacks you already use

LimbBot sits between task-level planning and robot hardware, with simulation and cloud tools for continuous adaptation.

ROS 2 nodes and SDK interfaces
Jetson-class edge inference
Isaac-based training and validation
Cloud fine-tuning with deployment review

NVIDIA

NVIDIA ecosystem integrations

The training and deployment workflow is built around physical AI infrastructure.

Isaac Lab

GPU-parallel RL environments for locomotion and contact-rich tasks.

Isaac Sim

Physics validation and scenario generation before robot tests.

Jetson Thor

Edge inference target for low-latency control loops.

CUDA and TensorRT

Optimized policy execution on supported NVIDIA hardware.

Omniverse/OpenUSD

Digital twins, assets, and scene interchange.

Cosmos

Synthetic world and scenario generation input for broader training distributions.

GR00T tooling

Humanoid workflow alignment for embodied foundation models.

Robot software

Interfaces for robot teams

LimbBot API is designed for teams using modern robotics middleware and deployment systems.

ROS 2

Control node, topics, actions, and telemetry streams.

REST

Cloud job orchestration and model management.

Telemetry

Structured logs for policies, guard events, and calibration.

Containers

Deployment packaging for robot compute and lab servers.

Cloud

Data and adaptation services

Cloud components support tuning, review, and training rather than real-time control.

Studio

Fine-tune policies for new dynamics and tasks.

Sim

Run Isaac-based RL and regression suites.

Model registry

Track approved policies and deployment envelopes.

Audit logs

Record who approved policy movement into a robot fleet.

How-to

Integration path

Connect model

Import URDF or USD, actuator limits, sensors, and compute profile.

Wire interfaces

Install ROS 2 node or SDK hooks for commands and telemetry.

Calibrate

Run a guided routine to fit dynamics and state estimation.

Deploy guarded

Run Core with Guard limits and observe telemetry before scale.

Reference

Reference architecture

A typical deployment uses cloud simulation and fine-tuning, a policy registry, and on-robot runtime with strict guard policies.

Cloud jobs are not in the real-time loop
Approved policies are versioned and signed by release process
Robot runtime exposes health, confidence, and guard events
Customer planners keep task ownership

Compatibility

Integration matrix

Area	Current direction	Notes
Simulation	Isaac Lab, Isaac Sim	primary training path
Robot middleware	ROS 2, SDK	API details vary by program
Edge compute	Jetson Thor class	reviewed per control loop and sensors
Scene format	OpenUSD	used for training worlds and digital twins
Planning	LimbBot Planner or customer planner	goal interface contract

Workflow

Developer workflow

Teams can treat LimbBot as a versioned control component with test gates.

Local test

Run interface checks against a simulated body.

Simulation regression

Replay terrain, perturbations, and payload cases.

Hardware gate

Deploy to a lab robot under limited speed and force.

Fleet release

Promote approved policy versions by robot group.

Security

Security and deployment controls

Robotics programs need controlled model movement and auditability.

Least privilege

Separate cloud training roles from robot deployment roles.

Approval gates

Human approval before policy promotion.

Telemetry scope

Collect only agreed runtime and safety signals.

Incident review

Use logs to trace guard interventions and recovery events.

Questions

Integration FAQ

ROS 2 is the default integration path, but SDK and REST interfaces support other stacks.

No. Real-time control runs on-robot at the edge; cloud is for training and review.

LimbBot centers on Isaac, with import paths discussed per program.

Yes. LimbBot can consume goals from customer planners or LimbBot Planner.

Integrate LimbBot with your robotics stack

Send your middleware, compute, and robot model details for an integration review.

Request access