README.md

metadata

title: Representation Over Routing Demo
emoji: 🚀
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.14.0
python_version: '3.13'
app_file: app.py
pinned: false
license: cc-by-4.0
short_description: Interactive LunarLander demo for Representation over Routing
models:
  - ben-dlwlrma/Representation-Over-Routing

Representation over Routing Demo

Interactive demo for the preprint "Representation over Routing: Overcoming Surrogate Hacking in Multi-Timescale PPO".

This Space visualizes the four pretrained ablation-stage agents from the associated model repository. The default selection shows Stage 4: Target Decoupling, the proposed method, performing a LunarLander rollout with a fixed deterministic demo seed.

Links

• Paper: https://arxiv.org/abs/2604.13517
• Model Weights: https://huggingface.co/ben-dlwlrma/Representation-Over-Routing
• GitHub Repository: https://github.com/ben-dlwlrma/Representation-Over-Routing

Notes

The paper experiments were conducted on LunarLander-v2. This hosted demo uses LunarLander-v3 for compatibility with current Gymnasium releases, while keeping the same actor architecture and pretrained weights.