This site is open source. Improve this page »
Gradient optimization for inverse kinematics
Inverse kinematics is one of the hardest problems in robotics. Given desired goal position and orientation for an end-effector, inverse kinematics seeks to find a suitable series of joint configurations to meet that goal. In this paper, we develop a forward kinematics model in Python and apply gradient-based methods using automatic differentiation to develop an inverse kinematics model of an arbitrary $n$-joint robot arm. We evaluate the performance of standard gradient descent against Nesterov Momentum gradient descent. We conclude with a discussion of the limitations of our 2D model and areas for extension to gain more fidelity with real-world robots.
May 31st, 2021