Abstract:Learning a control policy capable of adapting to time-varying and potentially evolving system dynamics has been a great challenge to the mainstream reinforcement learning (RL). Mainly, the ever-changing system properties would continuously affect how the RL agent interacts with the state space through its actions, which effectively (re-)introduces concept drifts to the underlying policy learning process. We postulated that higher adaptability for the control policy can be achieved by characterizing and representing actions with extra "degrees of freedom" and thereby, with greater flexibility, adjusts to variations from the action's "behavioral" outcomes, including how these actions get carried out in real time and the shift in the action set itself. This paper proposes a Bayesian-flavored generalized RL framework by first establishing the notion of parametric action model to better cope with uncertainty and fluid action behaviors, followed by introducing the notion of reinforcement field as a physics-inspired construct established through "polarized experience particles" maintained in the RL agent's working memory. These particles effectively encode the agent's dynamic learning experience that evolves over time in a self-organizing way. Using the reinforcement field as a substrate, we will further generalize the policy search to incorporate high-level decision concepts by viewing the past memory as an implicit graph structure, in which the memory instances, or particles, are interconnected with their degrees of associability/similarity defined and quantified such that the "associative memory" principle can be consistently applied to establish and augment the learning agent's evolving world model.

Grey Reinforcement Learning for Incomplete Information Processing.

Mastering Robot Control Through Point-based Reinforcement Learning with Pre-training.

Deep Reinforcement Learning for NLP.

Guiding Reinforcement Learning with Incomplete System Dynamics

Robustness and Performance of Deep Reinforcement Learning.

Autonomous Learning and Navigation of Mobile Robots Based on Deep Reinforcement Learning

How to Train Your Robot with Deep Reinforcement Learning; Lessons We've Learned

Grounded Reinforcement Learning: Learning to Win the Game under Human Commands

Reinforcement learning for mobile robot: From reaction to deliberation

Decentralized Multi-Agent Reinforcement Learning with Global State Prediction

A Goal-Conditioned Reinforcement Learning Algorithm with Environment Modeling

Direct and indirect reinforcement learning

Towards Generalization and Data Efficient Learning of Deep Robotic Grasping

Robot Representation and Reasoning with Knowledge from Reinforcement Learning

GRI: General Reinforced Imitation and its Application to Vision-Based Autonomous Driving

Overview of Reinforcement Learning from Knowledge Expression and Handling

Deep Reinforcement Learning: An Overview

How to train your robot with deep reinforcement learning: lessons we have learned

Improving Generalization in Reinforcement Learning Training Regimes for Social Robot Navigation

Generalized Reinforcement Learning: Experience Particles, Action Operator, Reinforcement Field, Memory Association, and Decision Concepts