Neural-Network-Based pose estimation for cable-driven parallel robots

Brief Project Description

This project deals with machine learning algorithms for the pose estimation of Cable-Driven Parallel Robots. Such large scale systems are intrinsically complex to model, as time-varying nonlinear phenomena govern their mechanics. Therefore, conventional direct-kinematics based pose estimation is inherently limited. The potential of neural networks for addressing highly accurate pose estimation (mm) on large-scale robots (several m) is yet to be addressed, and this work aims at reviewing the state of the art of neural-network based pose estimators in robotics, conventional sensor-fusion schemes for robotics, and implement them on a cable-driven prototype, comparing and contrasting their costs and benefits.

This thesis is conducted in collaboration with: DIN UNIBO

Apply Now