Bi-Manual Manipulation and Attachment via Sim-to-Real Reinforcement Learning

TL;DR

This work targets bi-manual robotic manipulation and sim-to-real transfer by learning policies in simulation that operate in real-time joint space at $4$ Hz with no observation filtering. It introduces two coordination-focused tasks, Pickup and Connect magnet blocks, and demonstrates robust real-world transfer on two xArm6 robots with magnetic blocks, achieving $100\%$ pickup success and $65\%$ Connect success under certain conditions. The contributions include a sim-enhanced training pipeline with joint-space control, large-scale PPO training (billions of steps), frame-stacking, and a minimal yet challenging Connect Task designed to stress coordination and collision avoidance. The results show that modest simulator adjustments and direct joint-space policies can transfer effectively to physical bi-manual systems, enabling scalable development of multi-arm manipulation with reduced perception and planning overhead.

Abstract

Most successes in robotic manipulation have been restricted to single-arm robots, which limits the range of solvable tasks to pick-and-place, insertion, and objects rearrangement. In contrast, dual and multi arm robot platforms unlock a rich diversity of problems that can be tackled, such as laundry folding and executing cooking skills. However, developing controllers for multi-arm robots is complexified by a number of unique challenges, such as the need for coordinated bimanual behaviors, and collision avoidance amongst robots. Given these challenges, in this work we study how to solve bi-manual tasks using reinforcement learning (RL) trained in simulation, such that the resulting policies can be executed on real robotic platforms. Our RL approach results in significant simplifications due to using real-time (4Hz) joint-space control and directly passing unfiltered observations to neural networks policies. We also extensively discuss modifications to our simulated environment which lead to effective training of RL policies. In addition to designing control algorithms, a key challenge is how to design fair evaluation tasks for bi-manual robots that stress bimanual coordination, while removing orthogonal complicating factors such as high-level perception. In this work, we design a Connect Task, where the aim is for two robot arms to pick up and attach two blocks with magnetic connection points. We validate our approach with two xArm6 robots and 3D printed blocks with magnetic attachments, and find that our system has 100% success rate at picking up blocks, and 65% success rate at the Connect Task.

Figures (14)

• Figure 1: Visualization of simulated and real-world block Connect Task.
• Figure 2: Dual xArm6 configuration.
• Figure 3: Magnetic blocks and their components.
• Figure 4: Camera and lighting configuration.
• Figure 5: Overall robot system architecture diagram.