Optimized-tactile-sensor-model-for-shadow-hand-in-Gym-simulation

This project optimized the gym-0.20.0 default tactile sensor configuration for hand tasks.

*The project also supply an adding install version in another branch, if you already install the gym, you could directly use that branch.

Requirement

python = 3.6
Mujoco = 150
mujoco-py = 1.50.1.0
gcc&g++ = 4.8
Cython = 0.29.21
opencv-python = 4.3.0.38
HER (need for running demo)

Installation

1. To install this optimized gym version, use
   git clone https://github.com/WilliamAlexanda/Optimized-tactile-sensor-model-for-shadow-hand-in-Gym-simulation.git

2. To install mjpro150, use this link: https://roboti.us/download.html Please follow this link to install the license: https://roboti.us/license.html

3. To install mojoco-py, use
   pip install mujoco-py==1.50.1.0

4. (Optional) To install HER, please refer to this link: https://github.com/TianhongDai/hindsight-experience-replay

Test

1. After complete step 1~3, you could use python default action.py to see if the environment is correctly installed. If correct, the new window will show the dexterous hand with new tactile sensor configuration.
2. After complete the step 4, you could put the "HandManipulateBlockTouchSensors-v0" into the saved_model floder of HER and use python demo.py --env-name=HandManipulateBlockTouchSensors to run the demo.

Reference

https://github.com/openai/gym/tree/v0.20.0