Autonomous systems for robotic product sortation

Model fine-tuning and grasp optimization for high-accuracy and high-throughput robotic piece-picking.

• Designed a custom Convolutional Neural Network (CNN) for state estimation of robotic arm and surroundings in simulation

• Set-up 3 cameras in simulation, pre-processed RGBD images (cropping, rotating, converting to 3-channels) using PIL library, Torchvision

• Integrated it with a Deep Reinforcement Learning network (DDPG+HER) to learn control policies for push, pick-place, slide task

TA for Machine Learning and Artificial Intelligence for Engineers (24-787)