Posts

We introduce Grounded Curriculum Learning (GCL), a novel framework that improves real-world reinforcement learning in robotics by aligning simulated task distributions with real-world tasks while considering task sequences and robot performance history.

In this assignment, we implemented two types of GANs - a Deep Convolutional GAN (DCGAN) and a CycleGAN. The DCGAN was trained to generate grumpy cats from random noise, while the CycleGAN was trained to convert between two types of cats (Grumpy and Russian Blue) and between apples and oranges. Both GANs were implemented with data augmentation and differentiable augmentation techniques.