Links

Abstract

In this work we propose algorithms and methods that enable learning dexterous object manipulation using simulated one- or two-armed robots equipped with multi-fingered hand end-effectors. Using a parallel GPU-accelerated physics simulator (Isaac Gym), we implement challenging tasks for these robots including regrasping, grasp-and-throw, and object reorientation. To solve these problems we introduce a decentralized Population-Based Training (PBT) algorithm that allows us to massively amplify the exploration capabilities of deep reinforcement learning. We find that this method significantly outperforms regular end-to-end learning and is able to discover robust control policies even for the most challenging tasks.

Additional Agent Demonstrations

Single-Arm Tasks

Dual-Arm Tasks

Reorientation - Alternative Behavior

This policy, trained with Population-Based Training with 16 agents discovers an alternative strategy. It avoids some in-hand rotations and prefers to put the object back on the table to rotate it there without any risk of dropping the object.