Learning to Push by Grasping: Using multiple tasks for effective learning

TL;DR

This paper demonstrates that multi-task learning in end-to-end robot control models improves performance and data efficiency, enabling better grasping and pushing skills with fewer training examples.

Contribution

It introduces a multi-task learning approach for robot control that outperforms task-specific models trained on the same data amounts.

Findings

Multi-task models outperform task-specific models with equal data.

Shared learning reduces data requirements for effective robot control.

Joint training on grasping and pushing improves overall performance.

Abstract

Recently, end-to-end learning frameworks are gaining prevalence in the field of robot control. These frameworks input states/images and directly predict the torques or the action parameters. However, these approaches are often critiqued due to their huge data requirements for learning a task. The argument of the difficulty in scalability to multiple tasks is well founded, since training these tasks often require hundreds or thousands of examples. But do end-to-end approaches need to learn a unique model for every task? Intuitively, it seems that sharing across tasks should help since all tasks require some common understanding of the environment. In this paper, we attempt to take the next step in data-driven end-to-end learning frameworks: move from the realm of task-specific models to joint learning of multiple robot tasks. In an astonishing result we show that models with multi-task learning tend to perform better than task-specific models trained with same amounts of data. For example, a deep-network learned with 2.5K grasp and 2.5K push examples performs better on grasping than a network trained on 5K grasp examples.