N$^{2}$M$^{2}$: Learning Navigation for Arbitrary Mobile Manipulation Motions in Unseen and Dynamic Environments

Despite its importance in both industrial and service robotics, mobile manipulation remains a significant challenge as it requires seamless integration of end-effector trajectory generation with navigation skills as well as reasoning over long-horizons. Existing methods struggle to control the large configuration space and to navigate dynamic and unknown environments. In the previous work, we proposed to decompose mobile manipulation tasks into a simplified motion generator for the end-effector in task space and a trained reinforcement learning agent for the mobile base to account for the kinematic feasibility of the motion. In this work, we introduce Neural Navigation for Mobile Manipulation (N$^{2}$M$^{2}$), which extends this decomposition to complex obstacle environments, extends the agent's control to the torso joint and the norm of the end-effector motion velocities, uses a more general reward function and, thereby, enables robots to tackle a much broader range of tasks in real-world settings. The resulting approach can perform unseen, long-horizon tasks in unexplored environments while instantly reacting to dynamic obstacles and environmental changes. At the same time, it provides a simple way to define new mobile manipulation tasks. We demonstrate the capabilities of our proposed approach in extensive simulation and real-world experiments on multiple kinematically diverse mobile manipulators.

I. Introduction

While recent progress in control and perception has propelled the capabilities of robotic platforms to autonomously operate in unknown and unstructured environments [1], [2], [3], [4], this has largely focused on pure navigation tasks [5], [6]. In this work, we focus on autonomous mobile manipulation, which combines the difficulties of navigating unstructured, human-centered environments with the complexity of jointly controlling the base and arm. Mobile Manipulation is commonly reduced to sequential base navigation followed by static arm manipulation at the goal location. This simplification is restrictive as many tasks such as door opening require the joint use of the arm and base and is inefficient as it dismisses simultaneous movement and requires frequent repositioning.