srb agent teleop — Teleoperate Agent
Test environments and agent behavior by controlling robots with supported devices.
Usage
srb agent teleop --env ENV_ID [options]
Options | Shared Agent Options
| Option | Description | Default |
|---|---|---|
--teleop_device [DEV …] | Interface (keyboard, spacemouse, ros, gamepad, haptic) | env default, else [keyboard] |
--pos_sensitivity VAL | Translation sensitivity | 1.0 |
--rot_sensitivity VAL | Rotation sensitivity (rad/s) | π (≈3.14) |
--invert_controls/--invert | In manipulation mode, flip the first two translational controls. Boolean optional forms --no-invert_controls and --no-invert disable inversion. | True |
| Teleoperation via Policy | ||
--algo ALGO | Algorithm of the policy (RL or IL) | |
--model MODEL | Path to the policy checkpoint or an SRB model artifact directory; this exact policy is loaded | |
--obs {state,visual} | Observation modality of the policy: selects the agent-config variant and the log tree searched for its latest checkpoint | state |
Teleoperation Modes
Action space selects direct or policy control:
Direct Teleoperation
Use direct teleoperation when actions map to high-level commands such as desired end-effector pose or mobile-robot acceleration.
Examples
In _manipulation, use ur10:
srb agent teleop -e _manipulation env.robot=ur10
In _orbital, use spacemouse:
srb agent teleop -e _orbital --teleop_device spacemouse
Teleoperation via Policy
Use a trained RL or imitation-learning policy for complex or high-dimensional actions, such as humanoid joint control.
Any algorithm supported by srb agent eval works. The
device supplies command-like observations. The policy generates actions.
Examples
Reference: Reinforcement Learning Workflow
Reference:srb agent train— Train Agent
-
Train Dreamer on
locomotion_velocity_tracking:srb agent train --headless --algo dreamer -e locomotion_velocity_tracking env.num_envs=256 -
Teleoperate with it:
srb agent teleop --algo dreamer -e locomotion_velocity_tracking -
To load another checkpoint instead of the latest, set
--model:srb agent teleop --algo dreamer -e locomotion_velocity_tracking --model logs/locomotion_velocity_tracking/dreamer/...
SpaceMouse input
Failed or None reads discard buffered six-axis motion. No successful read for
0.5 s also discards it. Valid reads restore input, while lost reports during
continuing reads retain the last motion.
Reset discards in-flight reads. Failed reads preserve the gripper toggle. This is not a hardware stop or inactivity timeout.
Native teleoperation
Native srb real_agent teleop requires generated COMMAND_MAPPING metadata and
direct ROS, SpaceMouse, or haptic input. ROS is the default. Keyboard, gamepad,
and policy-command arguments are unavailable. See the
native direct-input contract.
Public Demo Launch
For public events, excavation_demo defaults to one environment, manual reset,
long horizon, all HUD lines visible, and the high-quality RTX preset.
srb agent teleop --env excavation_demo
Add --hide_ui for fullscreen without the Isaac Sim UI:
srb agent teleop --env excavation_demo --hide_ui
Task config defaults to spacemouse, haptic, and keyboard; thus
--teleop_device is optional. Set venue or visual tier with Hydra overrides
such as env.scenery=lunalab or env.demo_regolith_quality=visual.