Learning of Long-Horizon Sparse-Reward Robotic Manipulator Tasks with Base Controllers

TL;DR

This paper introduces a reinforcement learning method that integrates traditional base controllers to efficiently learn long-horizon, sparse-reward robotic manipulation tasks, significantly improving performance and sample efficiency.

Contribution

It proposes a novel approach combining base controllers with deep RL to enhance exploration and learning in complex robotic tasks, including synthesizing multiple controllers.

Findings

Outperforms base controllers in stacking and cup tasks

Achieves orders of magnitude better sample efficiency than learning from demonstrations

Demonstrates potential for industrial robot manipulation systems

Abstract

Deep Reinforcement Learning (DRL) enables robots to perform some intelligent tasks end-to-end. However, there are still many challenges for long-horizon sparse-reward robotic manipulator tasks. On the one hand, a sparse-reward setting causes exploration inefficient. On the other hand, exploration using physical robots is of high cost and unsafe. In this paper, we propose a method of learning long-horizon sparse-reward tasks utilizing one or more existing traditional controllers named base controllers in this paper. Built upon Deep Deterministic Policy Gradients (DDPG), our algorithm incorporates the existing base controllers into stages of exploration, value learning, and policy update. Furthermore, we present a straightforward way of synthesizing different base controllers to integrate their strengths. Through experiments ranging from stacking blocks to cups, it is demonstrated that the learned state-based or image-based policies steadily outperform base controllers. Compared to previous works of learning from demonstrations, our method improves sample efficiency by orders of magnitude and improves the performance. Overall, our method bears the potential of leveraging existing industrial robot manipulation systems to build more flexible and intelligent controllers.