Your robotics native agent, grounded in physical world intuition, to accelerate engineering with precision.

Drift is an end-to-end robotics simulation engineering agent. It turns natural-language prompts into production-grade simulation workspaces for ROS2, Gazebo, MuJoCo, and Isaac Sim, generating URDF, SDF, MJCF, and USD files, configuring plugins and controllers, and scaffolding colcon workspaces. Beyond setup, Drift runs scenario variations, captures synthetic data for training, and (coming soon) trains policies on managed cloud compute, collapsing the full sim engineering loop into a single tool.

Drift supports ROS2 and three major simulators: Gazebo, MuJoCo, and Isaac Sim (coming soon). Drift generates the matching world and robot description files: SDF for Gazebo, MJCF for MuJoCo, and USD for Isaac Sim. It also produces the launch files, controllers, and plugin configuration needed to run them end to end.

General-purpose coding agents don't understand ROS2 plumbing, URDF and SDF semantics, simulator physics, or colcon build conventions. Drift is purpose-built for robotics. It reasons about TF trees, joint controllers, plugin compatibility, mass and inertia validity, and Gazebo, MuJoCo, and Isaac Sim configuration as first-class concepts. The result is working simulations on the first run, not code that compiles but won't launch.

Yes. Drift is an end-to-end simulation engineering agent, not just a setup tool. Tell Drift the axes you want to sweep, such as lighting, sensor noise, object placement, friction, obstacle density, payload mass, or starting pose, and it generates parameterized world variants, runs them, and reports where your robot succeeds or fails. This turns simulation from a one-off setup into a repeatable test loop you can iterate on like a CI pipeline.

Yes. Drift can scaffold synthetic data pipelines on top of your Gazebo, MuJoCo, or Isaac Sim worlds, capturing RGB, depth, segmentation masks, point clouds, joint states, and force/torque streams in standardized formats like ROS bags and parquet. This makes it straightforward to generate training datasets for perception models, behavior cloning, and policy learning directly from Drift-generated simulations, without writing recording infrastructure yourself.

Coming soon. Drift is rolling out training-in-simulation support backed by managed cloud compute, letting you spin up parallel sim instances on GPU servers, run reinforcement learning or behavior-cloning loops, and pull trained policies back to your local workspace, all from the Drift CLI. This closes the full loop inside one tool: world generation → scenario testing → data collection → policy training.

Drift combines frontier LLMs with custom-trained agents for robotics tooling and physics-aware models grounded in real-world dynamics. This combination lets Drift catch physically invalid configurations like bad inertia tensors, broken joint limits, or unstable collision geometry before launch, instead of debugging them in simulation afterward.

Not yet. Drift today generates simulation workspaces, including URDFs, controllers, launch files, and Gazebo, MuJoCo, and Isaac Sim worlds. The generated ROS2 packages can be ported to real hardware manually, but a first-class sim-to-real deployment flow is on the roadmap, not in the current beta.

Yes. Drift accepts existing URDF and SDF files referenced in your prompt (for example, "use my robot.urdf and add a stereo camera") and extends them in place. Drift performs best when existing files follow standard ROS2 conventions; highly customized or non-standard structures may need manual adjustment.

Yes. Drift configures standard Gazebo Harmonic plugins automatically, including cameras, LiDAR, IMUs, depth sensors, and force/torque sensors. For third-party or custom .so plugins, mention them by name in your prompt and Drift will add the plugin block to the SDF; you may still need to verify build paths and plugin parameters manually.

No. Drift is ROS2-only. ROS1 support is not on the roadmap, since the ROS ecosystem itself is migrating off ROS1 (Noetic reached end of life in May 2025).

Drift runs locally on your machine. Your workspace, URDF and SDF files, mesh assets, and ROS code never leave your environment. Only the text of your prompts and the diffs Drift proposes are sent to our API. This makes Drift safe to use on internal robot designs and IP-sensitive workspaces. An air-gapped or offline mode is not available in the current beta.

Drift runs on Ubuntu. macOS and Windows are not officially supported today; engineers on those platforms typically run Drift inside an Ubuntu VM, a container, or a remote dev box.