TCG

<palign="center"> System Overview

Dependencies

- Ubuntu 20.04
- Python 3.6

The file of the conda environment is environment.yml. We use [V-REP 3.5.0] as the simulation environment.

Code

We do experiments on a NVIDIA 1080 Ti GPU. It requires at least 8GB of GPU memory to run the code.

To train a regular TCG policy from scratch in simulation, first start the simulation environment by running V-REP (navigate to your V-REP directory and run ./vrep.sh). From the main menu, select File > Open scene..., and open the file simulation/simulation.ttt. Then navigate to this repository in another terminal window and run the following:

Training

To train from scratch, run

python main.py

Testing

python main.py
--is_testing --test_preset_cases --test_target_seeking
--load_ckpt --critic_ckpt CRITIC-MODEL-PATH --coordinator_ckpt COORDINATOR-MODEL-PATH
--config_file TEST-CASE-PATH

We use different colors to label the target object in simulation/random. To design your own challenging test case in simulation/preset.

Acknowledgments

We reference the following code in our project

• Visual Pushing and Grasping Toolbox

• A Deep Learning Method to Grasp the Invisible

• Light-Weight RefineNet for Real-Time Semantic Segmentation

• TransCG: A Large-Scale Real-World Dataset for Transparent Object Depth Completion and A Grasping Baseline