Wield: Systematic Reinforcement Learning With Progressive Randomization

TL;DR

Wield is a novel system that streamlines task design in reinforcement learning by decoupling interfaces from task representations and introducing a staged randomization protocol for systematic evaluation.

Contribution

It introduces Wield, a system that facilitates task design and evaluation in reinforcement learning through modular primitives and a new staged randomization protocol.

Findings

Enables decoupling of system interfaces from task representations.

Provides a structured protocol for incremental model evaluation.

Supports practical reinforcement learning with flexible task design.

Abstract

Reinforcement learning frameworks have introduced abstractions to implement and execute algorithms at scale. They assume standardized simulator interfaces but are not concerned with identifying suitable task representations. We present Wield, a first-of-its kind system to facilitate task design for practical reinforcement learning. Through software primitives, Wield enables practitioners to decouple system-interface and deployment-specific configuration from state and action design. To guide experimentation, Wield further introduces a novel task design protocol and classification scheme centred around staged randomization to incrementally evaluate model capabilities.