Lots of learning algorithms have been created to solve robot manipulation problems (such as robotic stacking) with a large and diverse set of objects. However, most of them focused on tasks like grasping that do not typically require complex inter-object contact dynamics.
An industrial robot. Machines like this are used to perform robotic stacking operations in a variety of industrial settings. Image credit: Inst. of Robotics, JKU via Flickr, CC BY-NC-ND 2.0
Therefore, a recent paper on arXiv.org investigates the stacking of diverse objects in the real world via a learned policy, using only information from RGB cameras and proprioception.
Proposed solution for improvement of robotic stacking
Two tasks were proposed: the first required to master stacking for a set of 5 specific combinations of objects. The second challenged general stacking strategies from a large set of training objects. The study confirms the possibility of learning a vision-based policy that can stack multiple combinations of objects.
It also demonstrates a variety of stacking strategies for non-cuboid objects. The results can be obtained without the need for human demonstrations.
We study the problem of robotic stacking with objects of complex geometry. We propose a challenging and diverse set of such objects that was carefully designed to require strategies beyond a simple “pick-and-place” solution. Our method is a reinforcement learning (RL) approach combined with vision-based interactive policy distillation and simulation-to-reality transfer. Our learned policies can efficiently handle multiple object combinations in the real world and exhibit a large variety of stacking skills. In a large experimental study, we investigate what choices matter for learning such general vision-based agents in simulation, and what affects optimal transfer to the real robot. We then leverage data collected by such policies and improve upon them with offline RL. A video and a blog post of our work are provided as supplementary material.
Research paper: Lee, A. X., “Beyond Pick-and-Place: Tackling Robotic Stacking of Diverse Shapes”, 2021. Link: https://arxiv.org/abs/2110.06192