GitHub  •  LinkedIn  •  X  •  Discord

Telekinesis Isaac Sim

Telekinesis Isaac Sim is a Python client for the Telekinesis Isaac Sim bridge, built on a C++ core.

Installation

It's highly recommended to install into a dedicated Conda environment:

conda create -n isaacsimclient python=3.12
conda activate isaacsimclient

Then install the package:

pip install telekinesis-isaacsim-client

Example Usage

from telekinesis.isaacsim_client import IsaacSimClient

client = IsaacSimClient(
    api_key="",
    base_url="http://127.0.0.1:8766",
    websocket_base_url="ws://127.0.0.1:8766",
)

created = client.articulation.create("/World/ur10e")
print(created["articulation_id"])

client.articulation.set_joint_positions(created["articulation_id"], [0.0] * created["num_dof"])

Suction grippers have their own group. close() blocks until the grip settles, so the status it returns is the settled one:

gripper = client.surface_gripper.create("/World/suction_gripper")
result = client.surface_gripper.close(gripper["surface_gripper_id"])
print(result["status"], result["gripped_objects"])

Cameras are registered the same way. capture() blocks until the frame is current, so the outputs it returns are that frame's own, and it activates any supported render output the camera isn't producing yet. Each output is a numpy array, because the bridge sends pixels as raw bytes rather than as JSON numbers:

camera = client.camera.create("/World/camera", resolution=[320, 240])
print(camera["active_data_types"], camera["supported_data_types"])

# depth isn't active on a new camera, so this first capture activates it.
frame = client.camera.capture(camera["camera_id"], ["rgb", "depth"])
print(frame["rgb"].shape, frame["depth"].dtype)

# A captured pointcloud is relative to the camera; ask for stage coordinates
# with world_frame=True.
frame = client.camera.capture(camera["camera_id"], ["pointcloud"])

Conveyors and lightbeam sensors are registered the same way. A belt runs at a signed speed along the travel direction its scene authored, so reversing one means starting it with a negative velocity rather than sending a direction:

belt = client.conveyor.create("/World/ConveyorBelt_A08", cargo_root="/World")
print(belt["direction"], belt["nominal_speed"])

client.conveyor.start(belt["conveyor_id"])          # its own authored speed
client.conveyor.start(belt["conveyor_id"], -0.5)    # reversed, 0.5 m/s
client.conveyor.stop(belt["conveyor_id"])

A lightbeam sensor is sampled rather than queried: a reading is whatever the last physics step left behind, so read() needs the timeline to be playing. broken is true when any beam in the curtain is broken:

sensor = client.lightbeam.create("/World/LightBeam_Sensor")
client.lightbeam.set_configuration(sensor["lightbeam_id"], num_rays=3, curtain_length=0.2)

reading = client.lightbeam.read(sensor["lightbeam_id"])
print(reading["broken"], reading["linear_depth"])

More examples will soon be available on Telekinesis IsaacSim Examples

Resources

• Documentation: full API reference and guides.

Support

• Open an issue on GitHub.
• Email: support@telekinesis.ai
• Discord

License

Proprietary. © Telekinesis. All rights reserved.