Learning to graps with primitive-shaped objects

Proposed Framework

Training Baxter to grasp objects based on vision feedback and joint-torque control.

Reinforcement Learning algorithm: Guided Policy Search.

The environment is simulated using Gazebo/ROS Kinetic.

Training session

gif

Learned policy

gif

Cristian Camilo Beltran-Hernandez
Cristian Camilo Beltran-Hernandez
PhD Student Robotics & Artificial Intelligence

My research interests include reinforcement learning, robot manipulation and software development.

Related