Positional Encoding for Robot Neural Self-Collision Checking
[ 1 ] Instytut Robotyki i Inteligencji Maszynowej, Wydział Automatyki, Robotyki i Elektrotechniki, Politechnika Poznańska | [ P ] employee
[2.2] Automation, electronics, electrical engineering and space technologies
2024
chapter in monograph / paper
english
- multilayer perceptron
- positional encoding
- collision checking
EN Multiple problems in robotics that are solved using relatively simple multilayer perceptron are defined in low-dimensional feature space. In this paper, we use a multilayer perceptron for self-collision detection of a mobile-manipulating robot. The input vector to the neural network is the configuration of the robot that consists of only six values. We show that enhancing the input vector by features obtained from positional encoding widely used in computer graphics improves classification accuracy. We show the results that suggest that positional encoding improves learning highfrequency functions and better represents higher-frequency details of the trained relation.
360 - 365
publisher's website
final published version
at the time of publication
20