[release/0.2.1] Change the prompt for the finetuning

[pulkitg01]

Summary

Align static apple prompt

Detailed description

• Reason: align the simulated static apple workflow with the same prompt used for the real setup.
• Changed the GR00T LeRobot training config prompt to move the apple to the plate.
• Updated the closed-loop evaluation config to use the same prompt.
• Updated the static apple policy-training docs snippet so users finetune with the matching prompt.
• Impact: training and evaluation now use consistent language conditioning across sim and real workflows.

[isaaclab-review-bot]

Review Summary

This PR simplifies the language instruction prompt from a detailed description to a more concise task statement across 3 configuration files and documentation.

Changes Reviewed

File	Change
docs/pages/example_workflows/static_apple/step_3_policy_training.rst	Updated documentation prompt
isaaclab_arena_gr00t/lerobot/config/g1_static_apple_config.yaml	Updated training config prompt
isaaclab_arena_gr00t/policy/config/g1_static_apple_gr00t_closedloop_config.yaml	Updated inference config prompt

Observations

✅ Consistency: The prompt change is applied consistently across all relevant files (documentation, LeRobot training config, and closed-loop policy config).

✅ Commit hygiene: Good to see the follow-up commit removing the trailing period for cleaner prompt formatting.

Clarification Request

The prompt is changing from:

"Pick up the apple from the shelf and place it onto the plate on the same shelf next to it."

to:

"move the apple to the plate"

Question: Could you confirm this matches the prompt used during the finetuning of checkpoint-20000? Language-conditioned policies can be sensitive to prompt wording, so ensuring the inference prompt matches the training prompt is important for optimal performance.

---

This is an automated review. Please address any questions before merging.

[shanereetz]

LGTM 🚀