Ansh Shah - Reading List

Efficient Dynamics Modeling In Interactive Environments With Koopman Theory

by Arnab Kumar Mondal, Siba Smarak Panigrahi ... and Siamak Ravanbakhsh

Leveraging Koopman Theory to efficiently learn a dynamics model which is further used to learn a policy. From Fluid Dynamics to Robotics RL environments.

DiffTORI: Differentiable Trajectory Optimization for Deep Reinforcement and Imitation Learning

by Weikang Wan, Ziyu Wang, Yufei Wang, Zachory Erickson and David Held

Using differentiable MPC to provide an additional gradient flow, supplementing the policy gradient, for policy updates.

Training Diffusion Models With Reinforcement Learning

by Kevin Black, Michael Janner, Yilun Du, Ilya Kostrikov and Sergie Levine

Using predefined metrics as sparse rewards to finetune diffusion models

ALOHA Unleashed: A Simple Recipe for Robot Dexterity

2024, Google DeepMind

Demonstration Imitation learning with a Multi View Transformer with proprioception can be used to learn dual arm fine manipulation skills.

Learning Task-Driven Control Policies via Information Bottlenecks

Year: 2016

An elegant application of information theory principles to enhance robustness in deep learning.

Hypernetworks

by David Ha, Andrew Dai, and Quoc V. Le, Google Brain

Examines how soft weight sharing in a multitask setup can help extract abstract concepts when sufficient tasks and data points are available.