DeepClaw-Logo

The DeepClaw Benchmark¶

The DeepClaw is a benchmarking model zoo that functions as a Reconfigurable Robotic Manipulation System for Robot Learning. The main homepage can be found at here. This is the GitHub repository of DeepClaw source code, including instructions for installing and using DeepClaw.

Resources¶

Documentation: https://bionicdl-sustech.github.io/DeepClaw/
Paper explaining DeepClaw: arXiv:2005.02588 [cs.RO]
Papers using DeepClaw:

Code Organization¶

The DeepClaw code is organized as follows:

configs/                    configuration for robotic station for manipulation tasks.
data/                       Common dataset
deepclaw/drivers/           drivers for various robotic hardware, i.e. ur, franka, aubo.
deepclaw/models/            model zoo for segmentation, classification, pick planning, and motion planning.
deepclaw/utils/             server setup with dockers and client setup for laptops (x86) and jetson (arm).
projects/proj_TrashSorting  a sample project to run deepclaw for sorting trash.
docs/                       description of this document as a manual.

Bibliography¶

@misc{wan2020deepclaw,
    title={DeepClaw: A Robotic Hardware Benchmarking Platform for Learning Object Manipulation},
    author={Fang Wan and Haokun Wang and Xiaobo Liu and Linhan Yang and Chaoyang Song},
    year={2020},
    eprint={2005.02588},
    archivePrefix={arXiv},
    primaryClass={cs.RO}
}