Robosuite Tutorials

This repository contains tutorials and examples for working with the robosuite robot simulation framework and reinforcement learning.

Setup

Ensure your Conda environment is active. For example:

conda activate hf

Tutorials

1. Getting Started (src/01_getting_started/)

• 01_hello_robosuite.py: A basic hello-world style tutorial using the interactive PyTorch/TorchRL wrapper to control a Panda robot using random actions in a PickPlace environment.
• 02_visualize_camera.py: Demonstrates offscreen rendering from multiple cameras (agentview, eye-in-hand) and saving the recorded frames to a video file (camera_observations.mp4).
• 03_depth_and_semantics.py: Demonstrates advanced rendering features, including:
  • Retrieving depth maps and converting raw normalized depth to actual metric distance (meters) using get_real_depth_map.
  • Retrieving instance-level and class-level semantic segmentation maps.
  • Resolving segmentations back to human-readable object class and instance names (e.g., Milk, Panda, bin1).
  • Compiling all four modalities (RGB, Depth, Instance, and Class) into a 2x2 grid video (depth_and_semantics.mp4).

To run the depth and semantics tutorial:

# Run from the src/ directory
python 01_getting_started/03_depth_and_semantics.py