UR7e-Policy-Lab / README.md
devkumarhf's picture
Publish browser-local SAC inference and DLS comparison
590673f verified
|
Raw History Blame Contribute Delete
1.53 kB
---
title: UR7e Policy Lab
emoji: 🦾
colorFrom: green
colorTo: gray
sdk: static
app_file: index.html
license: mit
short_description: Run a frozen SAC robot policy locally and compare DLS.
tags:
- robotics
- reinforcement-learning
- control
- simulation
---
# UR7e Policy Lab
Run the real exported deterministic SAC actor inside your browser, or compare a fixed damped-least-squares controller. Orbit and zoom the 3D kinematic skeleton, choose among 100 prespecified target trajectories, and perturb sensor observations. No inference API, paid GPU or third-party script is used.
The reference evaluation is the separate 600-episode Python CPU experiment linked in the app. Browser preview uses an independently implemented JavaScript runtime and seeded noise generator; it is not the source of the published Python metrics. The runtime is checked against exported observations, actions and forward kinematics.
[Download model and reproduce](https://huggingface.co/devkumarhf/ur7e-sac-line-following) · [Engineering source](https://github.com/devkumar-projects/UR7e-Reinforcement-Learning-Suite) · [Experimental analysis](https://www.kaggle.com/code/devkumarkaggle/ur7e-sac-vs-dls-reproducible-policy-benchmark)
Default kinematics only: no dynamics, collision, contact-force or hardware validation. Robot is displayed as a kinematic skeleton; no proprietary meshes are included. Original project and weights: Dev Kumar. Upstream robot description terms are preserved in THIRD_PARTY_NOTICES.md and UR-DESCRIPTION-LICENSE.