fix(sim): set ImageFormat.RGB for MuJoCo video frames

[Nabla7]

MuJoCo's renderer outputs RGB, but Image.from_numpy() defaults to BGR. This caused red/blue channel swap in Rerun camera visualization when running in simulation mode.

[greptile-apps]

Greptile Overview

Greptile Summary

This PR fixes a color channel swap bug in MuJoCo video frames by explicitly specifying ImageFormat.RGB when converting frames to Image messages. MuJoCo's renderer outputs RGB data, but Image.from_numpy() defaults to BGR format, causing red/blue channels to be swapped in Rerun visualizations.

Key changes:

• Added ImageFormat import to mujoco_connection.py
• Specified format=ImageFormat.RGB parameter in Image.from_numpy() call
• Added explanatory comment about MuJoCo renderer format and Image.from_numpy() default

Related issue discovered:
While reviewing, I found a similar bug in manipulation_module.py line 908 where RGB data is passed to Image.from_numpy() without specifying the format parameter.

Confidence Score: 5/5

• This PR is safe to merge with no risk
• The fix is minimal, well-documented, and directly addresses the root cause. The change only affects MuJoCo video frame processing and correctly specifies the RGB format that MuJoCo's renderer outputs. The fix follows the same pattern used in other parts of the codebase (webcam.py, dji_video_stream.py, unitree connection.py) where RGB data is explicitly tagged with ImageFormat.RGB.
• manipulation_module.py has a similar bug (not in this PR) that should be fixed separately

Important Files Changed

File Analysis

Filename	Score	Overview
dimos/robot/unitree_webrtc/mujoco_connection.py	5/5	Correctly fixes RGB/BGR channel swap by specifying ImageFormat.RGB when creating Image from MuJoCo renderer output

Sequence Diagram

sequenceDiagram
    participant MP as MuJoCo Process
    participant Renderer as mujoco.Renderer
    participant SHM as Shared Memory
    participant MC as MujocoConnection
    participant Image as Image.from_numpy
    participant Rerun as Rerun Visualization

    MP->>Renderer: render() camera frame
    Renderer-->>MP: RGB uint8 pixels (H×W×3)
    MP->>SHM: write_video(pixels)
    
    MC->>SHM: read_video()
    SHM-->>MC: RGB frame array
    MC->>Image: from_numpy(frame, format=ImageFormat.RGB)
    Note over Image: Correctly tagged as RGB<br/>(previously defaulted to BGR)
    Image-->>MC: Image object with RGB format
    MC->>Rerun: log image
    Rerun->>Rerun: Display with correct colors

Loading

[greptile-apps]

Additional Comments (1)

dimos/manipulation/visual_servoing/manipulation_module.py
Image.from_numpy() defaults to BGR format, but viz_rgb is RGB. Specify format explicitly:

            msg = Image.from_numpy(viz_rgb, format=ImageFormat.RGB)