Reward Conditioned

Reward-conditioned reinforcement learning (RCRL) focuses on training agents to perform actions based on predicted rewards, simplifying policy learning compared to traditional methods. Current research emphasizes improving the efficiency and robustness of RCRL, particularly addressing challenges like slow convergence in multi-armed bandit problems and generalization to unseen reward levels, with approaches including Bayesian methods and normalized weight functions to enhance performance. These advancements are significant for various applications, including vision-and-language navigation and natural language generation, where RCRL offers a flexible and data-efficient framework for training agents from suboptimal or limited data.

Papers

October 17, 2024

Reward-free World Models for Online Imitation Learning
Shangzhe Li, Zhiao Huang, Hao Su
Reinforcement Learning Imitation Learning Offline Imitation Reward Free Soft Q Learning Reward Conditioned

June 16, 2024

Improving Reward-Conditioned Policies for Multi-Armed Bandits using Normalized Weight Functions
Kai Xu, Farid Tajaddodianfar, Ben Allison
Reinforcement Learning Multi Armed Bandit Policy Gradient Policy Learning Weight Normalization Reward Conditioned

March 27, 2024

Scaling Vision-and-Language Navigation With Offline RL
Valay Bundele, Mahesh Bhupati, Biplab Banerjee, Aditya Grover
Offline Reinforcement Learning Vision and Language Navigation Reward Conditioned

February 4, 2024

BRAIn: Bayesian Reward-conditioned Amortized Inference for natural language generation from feedback
Gaurav Pandey, Yatin Nandwani, Tahira Naseem, Mayank Mishra, Guangxuan Xu, Dinesh Raghu, Sachindra Joshi, Asim Munawar, Ramón Fernandez Astudillo
Reinforcement Learning Language Generation Human Feedback Bayesian Perspective Brain Function Amortized Inference Distributional Policy Gradient Amortised Inference Reward Conditioned

December 10, 2023

Synergizing Quality-Diversity with Descriptor-Conditioned Reinforcement Learning
Maxence Faldor, Félix Chalumeau, Manon Flageat, Antoine Cully
Policy Gradient Evolutionary Algorithm Quality Diversity Evolutionary Robotics MAP Elite Reward Conditioned

May 18, 2023

Bayesian Reparameterization of Reward-Conditioned Reinforcement Learning with Energy-based Models
Wenhao Ding, Tong Che, Ding Zhao, Marco Pavone
Energy Based Model Reparameterization Method Adaptive Inference Reward Conditioned

Reward Conditioned

Papers

Reward-free World Models for Online Imitation Learning

Improving Reward-Conditioned Policies for Multi-Armed Bandits using Normalized Weight Functions

Scaling Vision-and-Language Navigation With Offline RL

BRAIn: Bayesian Reward-conditioned Amortized Inference for natural language generation from feedback

Synergizing Quality-Diversity with Descriptor-Conditioned Reinforcement Learning

Bayesian Reparameterization of Reward-Conditioned Reinforcement Learning with Energy-based Models