Revisiting Sparse Rewards for Goal-Reaching Reinforcement Learning

TL;DR

This paper demonstrates that sparse, minimum-time reward formulations in goal-reaching reinforcement learning can outperform dense rewards, with effective early indicators and rapid learning on real robots from pixel inputs.

Contribution

It reveals the advantages of sparse reward schemes for goal-reaching tasks, introduces goal-hit rate as an early success indicator, and shows rapid learning on real robots from scratch.

Findings

Sparse rewards lead to higher-quality policies.

Goal-hit rate predicts learning success.

Robots learn pixel-based policies in 2-3 hours.

Abstract

Many real-world robot learning problems, such as pick-and-place or arriving at a destination, can be seen as a problem of reaching a goal state as soon as possible. These problems, when formulated as episodic reinforcement learning tasks, can easily be specified to align well with our intended goal: -1 reward every time step with termination upon reaching the goal state, called minimum-time tasks. Despite this simplicity, such formulations are often overlooked in favor of dense rewards due to their perceived difficulty and lack of informativeness. Our studies contrast the two reward paradigms, revealing that the minimum-time task specification not only facilitates learning higher-quality policies but can also surpass dense-reward-based policies on their own performance metrics. Crucially, we also identify the goal-hit rate of the initial policy as a robust early indicator for learning success in such sparse feedback settings. Finally, using four distinct real-robotic platforms, we show that it is possible to learn pixel-based policies from scratch within two to three hours using constant negative rewards.