Betting for Sim-to-Real Performance Evaluation

This code provides the implementation to facilitate the paper titled "Betting for Sim-to-Real Performance Evaluation" accepted for publication at Robotics: Science and Systems (RSS) 2026.

Repository Structure

• synthetic_examples/ - Synthetic distribution experiments (Section III.A)
• pickandplace_acc/ - Pick-and-place accuracy experiments (Section III.B)
• locomotive_tracking/ - Locomotive tracking experiments (Section III.C)

Reproducibility

For each experiment, navigate to the corresponding folder and follow the reproducibility instructions in the respective README file:

• Synthetic Examples: cd synthetic_examples and follow synthetic_examples/README.md
• Pick-and-Place Accuracy: cd pickandplace_acc and follow pickandplace_acc/README.md
• Locomotive Tracking: cd locomotive_tracking and follow locomotive_tracking/README.md

Requirements

Python Version: Python 3.8 or above

Major Dependencies: Most required packages are listed in requirements.txt. Install them with:

pip install -r requirements.txt

Note on Experiment-Specific Requirements:

• synthetic_examples/ and pickandplace_acc/ have minimal requirements (no MuJoCo or PyTorch needed)
• locomotive_tracking/ requires MuJoCo and PyTorch for robot simulation

One More Thing

For readers of interest, we have created a toy example using the D120, the dice with the most faces in the world (120 sides, and also the largest mathematically fair die possible), to illustrate the core ideas of our approach with animated visualizations.

Open the interactive demo here: dice/demo.html (clone the repo and double-click to open locally for proper HTML rendering)