|
--- |
|
title: GenSim |
|
emoji: 📈 |
|
colorFrom: purple |
|
colorTo: indigo |
|
sdk: gradio |
|
sdk_version: 3.39.0 |
|
app_file: app.py |
|
pinned: false |
|
license: apache-2.0 |
|
--- |
|
|
|
# Generative Simulation Interactive Demo |
|
|
|
This demo is from the paper: |
|
|
|
<!-- [Code as Policies: Language Model Programs for Embodied Control](https://code-as-policies.github.io/) |
|
--> |
|
Below is an interactive demo for the simulated tabletop manipulation domain, seen in the paper section IV.D |
|
|
|
## Preparations |
|
1. Obtain an [OpenAI API Key](https://openai.com/blog/openai-api/) |
|
|
|
## Usage |
|
1. Type in desired task name in the box. Then GenSim will try to run through the pipeline |
|
2. The task name has the form word separated by dash. For instance, 'place-blue-in-yellow' and 'align-rainbow-along-line'. |
|
|
|
## Guideline |
|
|
|
## Known Limitations |
|
1. The code generation can fail or generate infeasible tasks. |
|
2. The low-level pick place primitive does not do collision checking and cannot pick up certain objects. |
|
3. Top-down generation is typically more challenging if the task name is too vague or too distant from motions such as stacking. |
|
|
|
|
|
## Note |
|
For GPT-4 model, each inference costs about $0.3. For GPT-3.5 model, each inference costs about $0.03. |
|
|
|
|
|
## Acknowledgement |
|
Thanks to Jacky's [code-as-policies](https://huggingface.co/spaces/jackyliang42/code-as-policies/tree/main) demo. |
