This is the project page for the 2025 HRI best technical paper nominee "Contrastive Learning from Exploratory Actions: Leveraging Natural Interactions for Preference Elicitation".
https://interaction-lab.github.io/CLEA/
This repository contains the code for re-running the experiments conducted in that paper, the website that provides and overview of the paper, and the various artifacts (dataset, user responses, etc.) that are necessary for training representations.
The code currently in this repository was tested on the following system configuration:
| Component | Spec |
|---|---|
| OS | Ubuntu 24.04.4 LTS (Linux 7.0.0-28-generic, x86_64) |
| CPU | Intel Core Ultra 9 285K (24 cores) |
| RAM | 62GB |
| GPU | NVIDIA GeForce RTX 5090 (32GB VRAM), driver 580.173.02 |
| CUDA | 12.0 |
| Python | 3.13.11 |
| PyTorch | 2.9.1+cu128 |
| Package Management | Conda |
This setup is generally quite excessive for this project. Many of the original experiments were conducted on a 2018 laptop with a 4GB GPU and a 2022 Macbook Air. The general requirements are:
Software
- Python 3.8+ (see
src/README.mdfor setup instructions andsrc/requirements.txtfor the full dependency list) - Conda recommended for environment management, though you are welcome to port this to uv if you prefer!
- PyTorch; a CUDA-capable GPU is recommended for training but the training scripts fall back to CPU automatically if none is available.
Hardware
- GPU: 8GB+ recommended. This is especially for training the visual/auditory models (CPU training works but is much slower)
- RAM: 16GB+ recommended. The dataloaders preload the full raw stimulus set (images/spectrograms) into memory per training run
- Disk: ~1.5GB for the full repository, including the dataset and example results.
There are two top-level folders in this repository. static contains the files served on the website, and src contains the code. Please refer to the README.md in the src folder for step-by-step instructions for running the paper's code.
Parts of this project page were adopted from the Nerfies page and eliahuhorwitz/Academic-project-page-template.
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
