Neural Networks in Robot Planning

Learn about neural networks in robotic planning through a Hebrew-language lecture that explores the integration of neural networks within the "sense-plan-act" paradigm of autonomous robotics. Delve into advanced concepts of world models for robotic manipulation, including the Causal InfoGAN model for rope manipulation and Deep Latent Particles (DLP) model. Explore how neural networks can accelerate planning algorithms through Bayesian Online Planning, understanding their role as approximate posteriors in tree search formulations. Associate Professor Aviv Tamar from Technion's Electrical and Computer Engineering department shares expertise in reinforcement learning and robot learning, demonstrating how neural network uncertainty can be leveraged for more efficient search processes in autonomous robotic systems.