Panda Manipulator Control with Obstacle Avoidance Through Reinforcement Learning in a Simulated Environment

Abstract

Robotic systems with learning capabilities have many powerful applications in unstructured environments. Through reinforcement learning, robots can quickly adapt to new situations and learn from direct environmental interaction. This work proposes a simulation environment based on Robotics Toolbox for Python to solve a classic problem of the inverse kinematics of manipulators, ensuring that the robot reaches the desired position without colliding with the obstacles present in the scene. The potential of this reinforcement learning method is illustrated through simulation using the Franka-Emika Panda manipulator trained by the Deep Deterministic Policy Gradient algorithm.