KO
|
EN
gitlite — search
Search
#python
#java
#python3
#arduino
#golang
#machine-learning
#rust
#html
#flask
#javascript
#seismology
#nodejs
AutoSERL
★ 14
Open GitHub ↗
No description available.
Download README (.md)
Explore Similar Repositories
cpa-grok
:
CLIProxyAPI 反代 Grok 免费号，免费体验 Grok-4.5 模型
skills
:
shanjian skills
Demo_repository_4_test
:
TEST_repository_описание
awesome-generative-ui
:
Developer resources for building with and learning about generative UI
TrafficAnalyzer
:
应急响应流量分析工具，支持pcap、excel、log等多种数据源，同时结合加解密、反编译等多个常用工作流。
// repository documentation
Was this content helpful?
★ 0
(0 ratings)
Select Rating:
★
★
★
★
★
Submit Feedback
Recent Feedback
×
Download README
Do you want to download the
README.md
file for
AutoSERL
?
Download (.md)
# One Demonstration Is Enough for Real-World Robotic Reinforcement Learning [](https://opensource.org/licenses/Apache-2.0) [](https://autoserl.github.io/) **Webpage: [https://autoserl.github.io/](https://autoserl.github.io/)** We provide the code and usage instructions for AutoSERL, which is built upon the [HIL-SERL](https://github.com/rail-berkeley/hil-serl) codebase. ## News - **[2026-06]** AutoSERL was accepted to ECCV 2026. ## Table of Contents - [One Demonstration Is Enough for Real-World Robotic Reinforcement Learning](#one-demonstration-is-enough-for-real-world-robotic-reinforcement-learning) - [Installation](#installation) - [Code and Usage Instructions](#code-and-usage-instructions) - [Contact](#contact) - [Acknowledgement](#acknowledgement) <!-- - [Citation](#citation) --> ## Installation Please follow the same environment setup instructions as [HIL-SERL](https://github.com/rail-berkeley/hil-serl). ## Code and Usage Instructions The implementations of the ***Sliding Window Intervention***, ***Safety Recovery***, and ***Intervention Termination*** mechanisms are located in the `auto_intervention_wrapper` class within the `serl_robot_infra/franka_env/envs/wrappers.py` file. During training, the `auto_intervention_wrapper` class must be imported and configured with the appropriate parameters in `examples/experiments/task/config.py`, as shown in the table below. | Parameter | Description | | :--- | :--- | | `demo_path` | The single demonstration trajectory used for reference by the automatic intervention mechanism. | | `th1` | Intervention termination threshold (in meters). | | `th2` | Intervention start threshold (in meters). | | `l_term` | Intervention termination threshold used by the Intervention Termination Criterion (in timesteps). | | `l_stag` | Window length used by the Safety Recovery Mechanism to detect stagnation (in timesteps). | | `recover_point0` | The index of the safe recovery target to which the robot can be guided when a safety risk occurs. | | `recover_point1` | The index of a trajectory point at which the robot is in stable contact with the interaction object. | ## Contact If you have any questions, please contact yuwanliu06@gmail.com. ## Acknowledgement Our codebase is developed based on [HIL-SERL](https://github.com/rail-berkeley/hil-serl). We sincerely thank the HIL-SERL team for their excellent work. ## Citation If our work is helpful to your research, we would appreciate your consideration of citing our paper: ```bibtex @article{liu2026autoserl, title={One Demonstration Is Enough for Real-World Robotic Reinforcement Learning}, author={Liu, Yuwan and Yu, Hongze and Liu, Song and Wang, Yuhan and Zhang, Junge and Yang, Yaodong and Chen, Yuanpei and Zhang, Ceyao}, journal={arXiv preprint arXiv:2607.01651}, year={2026} } ```
