Removed Depth and Segmentation Code

Remove reaming viewer code

remove viewer

Remove visualization flags

Mouse-Based Camera Control:

Introduced a run_counter in app.py to track the number of simulation runs, ensuring each run has a distinct identifier.

Using LLMPhy paper's definition of stability to implement classify orientation stability

Explicitly Reset the Simulation State

Comment out code for joints etc

Added options to see joint frames and transparent geometries

Remove function

Change to mjv_addGeoms

Correct function call placement of draw_world_axes

Visually render the coordinate axes (X, Y, Z) in the MuJoCo simulation

LLM prompt formatted as a natural language

Comment out print statement

Return a list of dictionaries (preferred for LLM-friendly JSON) python Copy Edit

Return gif_path and stability_flags

Comment out code

Stability Classification at End of Simulation

The MuJoCo Python binding requires positional arguments in a specific order and format (NumPy arrays with shape (3,) or (3,1)), and crucially, you must pass the force and torque as NumPy arrays, not lists.

Apply a Force to the Tray

Enable contact visualization in the renderer to debug collisions

Move variable definition and initialization up.

This will make the black disc fly in toward the center from the tray’s corner as intended, with clean and controllable motion.

Fix error

Can can dynamically adjust the camera view of a MuJoCo simulation programmatically

mujoco.Renderer(model, width=640, height=640) is asking for an image size larger than what the MuJoCo model framebuffer allows by default.

Multiple randomized dynamic objects. A pusher that activates mid-simulation. Full support for parameterized object count and rendering.

Updated run_tray_simulation() (safe for HuggingFace Spaces)

Change mj_render to mjr_render

TraySim module