Multi-Goal Reinforcement Learning: Challenging Robotics Environments and Request for Research

TL;DR

This paper introduces a new suite of challenging robotics control tasks within OpenAI Gym, designed for Multi-Goal Reinforcement Learning, and proposes research ideas to enhance RL algorithms, especially in multi-goal settings.

Contribution

It presents a set of complex robotics environments for Multi-Goal RL and suggests research directions to improve RL algorithms using techniques like Hindsight Experience Replay.

Findings

New robotics environments with sparse rewards for RL research

Framework for Multi-Goal RL with goal-conditioned inputs

Proposed research ideas for algorithm improvements

Abstract

The purpose of this technical report is two-fold. First of all, it introduces a suite of challenging continuous control tasks (integrated with OpenAI Gym) based on currently existing robotics hardware. The tasks include pushing, sliding and pick & place with a Fetch robotic arm as well as in-hand object manipulation with a Shadow Dexterous Hand. All tasks have sparse binary rewards and follow a Multi-Goal Reinforcement Learning (RL) framework in which an agent is told what to do using an additional input. The second part of the paper presents a set of concrete research ideas for improving RL algorithms, most of which are related to Multi-Goal RL and Hindsight Experience Replay.