Nikhil-Paleti/sawyer_analysis_reinforcement_learning

Analysis of rethink robotics sawyer. Along with reinforcement learning

Introduction

Sawyer_analysis.ipynb contains:

• Forward Kinematics
• Inverse Kinematics
• Manipulator Jacobian calculation
• Null Space motion
• yoshikawa's manipulability measure
• Trajectory planning

Sawyer_RL.ipynb contains:

• Function to train a ddpg model to lift a cube using sawyer robot
• Function to visualize the trained model
  Demonstartion of the learnt model: https://www.youtube.com/watch?v=UYUPgIz7v30

Steps to reproduce the environment:

1. Install Cuda 11.3 and the corresponding cudnn version
2. Insiall mujoco 2.1.0
3. Install Conda
4. Clone the base environment
5. Install mujoco-py
6. clone the robosuite repository and follow their installation steps
7. Install pytorch

Model for simulation

The model for simulation is taken from: https://github.com/vikashplus/sawyer_sim

Known issues:

1. Continuing training by loading models is not complete. We are not saving the buffer so loading only networks and training is not providing good results.

citations

@inproceedings{todorov2012mujoco, title={Mujoco: A physics engine for model-based control}, author={Todorov, Emanuel and Erez, Tom and Tassa, Yuval}, booktitle={2012 IEEE/RSJ International Conference on Intelligent Robots and Systems}, pages={5026--5033}, year={2012}, organization={IEEE} }

@inproceedings{robosuite2020, title={robosuite: A Modular Simulation Framework and Benchmark for Robot Learning}, author={Yuke Zhu and Josiah Wong and Ajay Mandlekar and Roberto Mart'{i}n-Mart'{i}n}, booktitle={arXiv preprint arXiv:2009.12293}, year={2020}

}