Add docs for gear assembly validation

[ashwinvkNV]

Description

Doc change to to show the users how to use play.py to verify the trained policy succeeds

[greptile-apps]

Greptile Summary

This PR enhances the gear assembly policy documentation by adding two new validation steps between training and real robot deployment. The changes guide users to first validate their trained policy in Isaac Lab using play.py, then test the complete ROS integration pipeline with Isaac Sim + Isaac ROS before deploying to real hardware.

Key Changes:

• Added Step 3 showing how to run play.py with a checkpoint for Isaac Lab validation
• Added Step 4 directing users to the Isaac Sim + Isaac ROS tutorial for full system validation
• Included a 2.2MB GIF demonstrating successful gear insertion validation in Isaac Lab
• Renumbered existing "Step 3: Deploy on Real Robot" to "Step 5" to accommodate the new validation steps

The documentation improvements follow a logical progression from training → Isaac Lab validation → ROS integration testing → real robot deployment, helping users catch issues earlier in the pipeline.

Confidence Score: 5/5

• This PR is safe to merge with no concerns - it only adds helpful documentation and a visual asset
• This is a pure documentation change that adds valuable validation steps to help users verify their trained policies before deploying to real hardware. The changes are well-structured, accurate (verified script paths and parameters), and the GIF is properly added via Git LFS. No code changes or functional modifications were made.
• No files require special attention

Important Files Changed

Filename	Overview
docs/source/policy_deployment/02_gear_assembly/gear_assembly_policy.rst	Added Step 3 for Isaac Lab validation and Step 4 for Isaac Sim + Isaac ROS validation before real robot deployment
docs/source/_static/policy_deployment/02_gear_assembly/isaaclab_gear_insertion_validation.gif	Added validation GIF (2.2MB) via Git LFS showing gear insertion policy execution

Sequence Diagram

sequenceDiagram
    participant User
    participant TrainScript as train.py
    participant IsaacLab as Isaac Lab
    participant PlayScript as play.py
    participant IsaacSimROS as Isaac Sim + Isaac ROS
    participant RealRobot as Real Robot

    User->>TrainScript: Start training with domain randomization
    TrainScript->>IsaacLab: Train policy in simulation
    IsaacLab-->>TrainScript: Save checkpoint (model_1499.pt)
    TrainScript-->>User: Training complete
    
    Note over User,PlayScript: Step 3: Validation in Isaac Lab
    User->>PlayScript: Run play.py with checkpoint
    PlayScript->>IsaacLab: Load policy and run inference
    IsaacLab-->>PlayScript: Execute gear insertion
    PlayScript-->>User: Visual validation in viewer
    
    Note over User,IsaacSimROS: Step 4: Validation with Isaac Sim + Isaac ROS
    User->>IsaacSimROS: Test ROS integration pipeline
    IsaacSimROS->>IsaacSimROS: Test policy inference with ROS
    IsaacSimROS->>IsaacSimROS: Test robot control through ROS actions
    IsaacSimROS-->>User: Full system validation
    
    Note over User,RealRobot: Step 5: Real Robot Deployment
    User->>RealRobot: Deploy policy to hardware
    RealRobot-->>User: Execute gear assembly task

Loading

[Mayankm96]

Could we avoid using gifs? Webpm is better for files.

NIT: Would be great to only get the video of the viewport and not the entire GUI (unless needed)