Design and Motion Planning for a Reconfigurable Robotic Base

Johannes Pankert, Giorgio Valsecchi, Davide Baret, Jon Zehnder, Lukasz L. Pietrasik, Marko Bjelonic, Marco Hutter

Introduction

Mobile manipulators in the construction industry face a fundamental design paradox: they must be compact enough to navigate cluttered indoor environments and pass through standard doorways, yet stable enough to support heavy robotic arms performing high-precision tasks like drilling, milling, or plastering. Traditional four-wheeled platforms often struggle with "wobble"—a phenomenon where the robot's weight shifts between different sets of wheels on uneven ground, leading to significant errors at the tool tip.

Researchers from ETH Zurich have addressed this trade-off by developing a reconfigurable mobile base that utilizes a novel "A-configuration" footprint and a unified motion planning framework. This system allows for both high maneuverability in tight spaces and rock-solid stability during manipulation.

The Challenge: Stability vs. Precision

While a robot is considered "stable" if it does not tip over, construction tasks require a higher standard of "precision stability." Most four-wheeled robots suffer from support switching. Because four points rarely lie perfectly on a single plane (due to uneven floors or mechanical play), the robot's weight typically rests on only three wheels at any given time.

As a robotic arm moves, the center of pressure—or Zero Moment Point (ZMP)—shifts. When the ZMP crosses the boundary between one triangle of wheels and another, the base tilts slightly. For a long-reach robotic arm, a millimeter of tilt at the base can translate into centimeters of deviation at the end-effector, rendering tasks like precise drilling impossible.

Hardware and Configuration Design

The proposed robot features four legs attached to a central body via passive joints. These joints can be locked using magnetic brakes, providing a rigid structure during operation without the weight and complexity of high-torque active motors. Each leg is equipped with an individually steerable wheel (swerve drive).

The robot can transition between three primary footprints:

  1. H-Configuration: A narrow, elongated footprint designed for navigating through doors and narrow corridors.
  2. X-Configuration: A wide, symmetric footprint that provides maximum resistance against tipping in all directions.
  3. A-Configuration: The key innovation of this research. By adopting an asymmetric layout—where two wheels are positioned close together and two are far apart—the researchers created overlapping "Support Triangles." This design ensures the ZMP remains within a single stable zone throughout the arm's range of motion, effectively eliminating the wobble caused by support switching.

Unified Motion Planning

To manage the complexity of simultaneous base movement, leg reconfiguration, and arm manipulation, the team implemented a Model Predictive Control (MPC) strategy using the OCS2 framework. Unlike traditional systems that treat navigation and reconfiguration as distinct operational modes, this unified MPC solves for all degrees of freedom simultaneously over a 10-second receding horizon. This allows the robot to dynamically adjust its footprint while moving, ensuring it is always in the optimal configuration for the task at hand.

Experimental Results

The system was rigorously evaluated through stability benchmarks and real-world construction scenarios:

  • Wobble Suppression: Comparative tests showed that the A-configuration almost entirely eliminated spikes in base angular velocity (vibrations) during arm rotation, whereas H and X configurations experienced significant tilting.
  • Tracking Accuracy: Using high-precision surveying tools for ground-truth data, the robot achieved a Mean Average Error (MAE) of 1.67 cm while following a 20.53m sinusoidal path. This level of precision is sufficient for demanding construction applications.
  • Field Demonstration: The robot successfully performed an automated plastering task on a 3.5m x 0.75m wall. It demonstrated the ability to navigate over obstacles like cables and maintain stability against the forceful interaction of a plaster spray gun.

Limitations and Future Directions

While the current system represents a significant step forward, several areas for improvement remain. The MPC currently utilizes a kinematic model; incorporating a dynamic model would allow for faster, more aggressive movements by accounting for the inertia of the heavy robotic arm.

Additionally, the robot currently relies on wheel odometry, which exhibited a 16.1 cm drift over 16 meters in tests. Integrating LIDAR or camera-based SLAM directly into the MPC will be essential for long-term autonomous operation. Finally, the researchers aim to automate the reconfiguration triggers, allowing the robot to autonomously "shrink" to avoid obstacles or "widen" to prepare for heavy lifting.

The authors have released their MPC and swerve-steering control modules as open-source software, providing a valuable resource for the broader robotics and ROS community.

Summary updated

BibTeX
@article{pankert2022design,
  author = {Johannes Pankert and Giorgio Valsecchi and Davide Baret and Jon Zehnder and Lukasz L. Pietrasik and Marko Bjelonic and Marco Hutter},
  title = {{Design and Motion Planning for a Reconfigurable Robotic Base}},
  journal = {IEEE Robotics Autom. Lett.},
  year = {2022},
  doi = {10.1109/lra.2022.3189166},
}
Figure