TL;DR
This paper presents a novel control framework for a bimanual mobile manipulator (the CMU ballbot) to dynamically maneuver a wheelchair. The system achieves smooth, compliant wheelchair motion by integrating whole-body control with a pushing pose optimizer and steering controller, enabling safe human-robot interaction in crowded environments.
What Problem Does This Paper Address?
Maneuvering nonholonomic systems like wheelchairs with mobile manipulators (MMs) is challenging. These systems have complex dynamics and are often used in human-centered environments, demanding safe and smooth operation. Existing approaches often struggle with the inherent nonholonomic constraints, varying inertial parameters, and the need for compliance, especially when dealing with heavy loads like a person in a wheelchair. The authors aim to bridge this gap by developing a controller that can effectively push a wheelchair while ensuring passenger comfort and safety for those around.
Methods
The core of the proposed framework is a pushing pose optimizer that plans the MM's whole-body motion based on a quasi-static analysis. This optimizer treats the MM's arms as passive spring-like elements, leveraging body leaning as the primary mechanism for velocity control.
Key components include:
- Dynamic Model: The paper details the dynamic model of the CMU ballbot, including its 7-DoF arms and balancing capabilities, and a simplified planar model for the wheelchair.
- Online Parameter Identification: An Extended Kalman Filter (EKF) is used to estimate unknown wheelchair parameters (mass, CoM, friction) online, adapting to varying loads.
- Task-Space Impedance Control: This is used for the MM's arms to ensure stable physical interaction and track desired end-effector motions.
- Balancing Controller: A cascaded PID and PD controller manages the ballbot's balance and tracks desired CoM leaning angles.
- Pushing Pose Optimizer: This module calculates optimal body leaning angles (Φ) to achieve desired wheelchair velocities while satisfying equilibrium equations. It formulates this as an unconstrained quadratic programming problem.
- Steering Controller: This heuristic aims to minimize the lateral body leaning angle (by) by planning desirable end-effector positions. It leverages the wheelchair's steering angle (β) to control the z-axis torque, enabling turns and minimizing aggressive body movements.
Datasets
Not discussed in the paper. The experiments were conducted using real-world hardware (CMU ballbot and wheelchair) with varying loads (empty wheelchair and wheelchair with a human subject).
Key Insights and Ideas
- Whole-Body Motion for Nonholonomic Cart Pushing: The paper demonstrates that by intelligently coordinating the MM's body leaning and arm poses, it's possible to achieve precise control of nonholonomic carts like wheelchairs.
- Quasi-Static Analysis for Pushing: Treating the arms as passive springs and relying on body leaning for control simplifies the complex dynamics and allows for human-like motion smoothness.
- Steering Heuristic for Reduced Lean: The steering controller is crucial for minimizing the ballbot's lateral body movement, which is important for passenger comfort and avoiding collisions in tight spaces.
- Integrated Compliance: The framework emphasizes overall system compliance, not just in the arms but also in the base, which is vital for safe physical human-robot interaction.
- Online Parameter Estimation: The ability to estimate wheelchair parameters online allows the system to adapt to different loads, from an empty cart to a person, without prior knowledge.
Key Results
- Velocity Tracking: The proposed framework successfully tracked desired linear velocities up to 0.45 m/s and angular velocities up to 0.3 rad/s with loads varying from 11.8 kg (empty) to 79.4 kg (human subject).
- Response Time: Average system response times were around 1.6-1.7 seconds for empty wheelchairs and 1.1-1.8 seconds for wheelchairs with a person.
- Motion Smoothness: The ballbot's maneuvering resulted in smoother wheelchair motion compared to human maneuvering, with lower acceleration norms (e.g., 25.3% higher acceleration norm on the x-axis compared to human).
- Physical Compliance: The system demonstrated good compliance, requiring only 22.7 N (x-axis) and 12.3 N (y-axis) of force to move the wheelchair when disturbed, indicating safety for accidental collisions.
- Navigation: The controller was integrated with the ROS navigation framework and successfully navigated the wheelchair through a lab with predefined obstacles, demonstrating real-time replanning and velocity command tracking.
Limitations and Future Work
- Limitations:
- The steering angle is limited to β ∈ [-35°,35°] due to arm workspace constraints.
- The paper acknowledges that the current framework is evaluated in a controlled lab environment.
- Future Work:
- Autonomous affordance detection for grasping point identification.
- Evaluation of the ballbot pushing wheelchairs on ADA-compliant ramps.
- Application of the framework to other non-holonomic cart systems like wheelbarrows and shopping carts.
Summary updated