Skip to navigation

Mock, simulation and real modes

View as Markdown

Developing a controller is slow if every run needs hardware. Modes let an application run against synthetic data, a simulator or the real robot, usually with the same controller. This page explains what each mode connects to and what it tells you. Each application reference documents its modes; CLI mode discovery lists their names.

The diagram below compares the I/O patterns around a shared controller. The branches are alternatives, not a single application with every mode enabled.

Isaac OS process State Commands State Commands State / sensors Commands Real: physical robot Simulation: robot protocol sim: Isaac Sim Mock: synthetic inputs Robot state input UCX receive bridge Controller Validation / discard sink Robot command output UCX send bridge

Reusing the controller pipeline across modes makes comparisons more useful, but mode names do not guarantee the same graph, model, configuration or timing. For example, GR00T/G1 uses different model profiles in mock and Isaac Sim modes, while SO-101 sinus uses a separate simulation graph that bypasses the motion planner used in mock and real modes. Check the application reference before comparing results across modes.

Mock

Isaac OS process Synthetic inputs Controller Validation / discard sink

Mock replaces robot or simulator I/O with synthetic inputs and a sink inside the Isaac OS process. The open loop shown here checks execution, output contracts and pipeline timing without modeling the robot’s response to commands.

For example, the SO-101 reach mock publishes a constant pose without integrating commands, and the GR00T/G1 mock uses deterministic inputs. These checks do not establish physical behavior or closed-loop performance.

Synthetic workloads can also include feedback: the 1 kHz control benchmark uses a simulated robot plant, rather than the open loop depicted above.

Real

Isaac OS process State Commands Real robot Robot state input Controller Robot command output

Real mode closes the loop through physical hardware. Input and output operators use the robot’s driver or SDK to receive state and send commands. For example, SO-101 reach uses a Feetech serial driver, while G1 WBC uses the Unitree SDK over DDS.

Simulation

Simulation closes the loop through a simulated robot and environment. Depending on the simulator, this can include physics and rendered sensor data. The reference applications illustrate two I/O patterns; each application’s available modes determine which it supports.

Vendor-specific simulator

Isaac OS process Robot protocol: state Robot protocol: commands Compatible robot simulator Robot state input Controller Robot command output

A simulator that implements the robot’s protocol can reuse its input and output operators. For example, G1 WBC uses the same robot-facing executable for real and unitree_sim. Unitree MuJoCo implements the robot protocol over DDS; unitree_sim selects domain 1 and the simulation configuration.

Isaac Sim (sim)

Isaac OS process State / sensors Commands Isaac Sim UCX receive bridge Controller UCX send bridge

The sim mode exchanges state, sensor data and commands with Isaac Sim over UCX. Each application defines its message schemas, scene and control timing. Simulator timestamps can affect controller pacing and motion. Start the simulator and application separately; see Isaac Sim setup.