Push-to-See

Learning Non-Prehensile Manipulation to Enhance Instance Segmentation via Deep Q-Learning

This is the main repository of our Push-to-See model. For technical details, the paper is accessible via this link.

Model weights

Trained model weigths can be download through the following links:

• Mask-RG (trained on Inria objects for 40 epochs): link
• Push-DQN (trained for 41k episodes): link

Demos

A short description of the model and video demonstrations can be seen on this video.

(Instructions for installation, training and running will be included in this readme file. For now, please contact the corresponding author if you require any help via baris.serhan@manchester.ac.uk)