DreamLake
Browse docs

Robot motion

An MCAP recording may contain link poses in /tf, joint angles in /joint_states, or both. Link poses can be drawn directly. Joint angles need a robot description to determine where each link goes.

Inspect the frame tree

For a recording with tf messages, add these views to its .dreamrc:

yaml
views:
  - view: recon3d
    up: z
    tracks: [{ field: '/tf::*', as: transform3d }]
  - view: timeline

The scene shows axes at tracked frames and connections between them. The reader composes parent-relative transforms across /tf and /tf_static. Scrub the timeline to check the motion; drag the scene to change the viewpoint.

Add the robot geometry

Add models to the same recon3d entry. For a G1 recording:

yaml
models:
  - https://huggingface.co/datasets/live9080/dreamlake-robots/resolve/main/g1/g1_29dof.urdf

URDF link names match the recorded frame names. Mesh paths resolve relative to the URDF; supported assets include STL, OBJ, DAE, and glTF. Expand a Xacro file before loading it. If the meshes appear but do not move, compare the link names with the fields listed for your recording.

Use joint states

For a recording of joint angles, bind the topic to joints alongside models:

yaml
joints: [/joint_states]
tracks: [{ field: '/tf::pelvis', as: transform3d }]

The model supplies the joint axes and link structure for forward kinematics. Values are matched by joint name, mimic joints follow their source, and bounded joints use the URDF limits. The optional pelvis track places this robot's base; without a base pose, it stays at the origin. A link's recorded pose takes precedence over a computed pose.

Resolve package paths

For package:// mesh references, provide the package URL when the URDF's own path does not identify it:

yaml
models:
  - url: https://example.com/robot/panda.urdf
    packages:
      franka_description: https://example.com/franka_description

The generated MCAP layout can choose a preset from matching frame or joint names. Inspect its models entry to see which description was selected, then replace it if your recording uses a different model. Supported message and container encodings are listed in the reader reference.