Trajectory Control for Differential Drive Mobile Manipulators

Harish Karunakaran, Gopeshh Raaj SubbarajarXiv 2023 · DOI · arXiv

TL;DR

This paper presents a comprehensive approach for controlling differential drive mobile manipulators, specifically the Fetch robot, to follow complex end-effector and platform trajectories. They successfully integrate dynamics modeling, trajectory planning, and inverse kinematics to enable precise pick-and-place tasks in a simulated warehouse environment, addressing challenges like nonholonomic constraints and system coupling.

What Problem Does This Paper Address?

Mobile manipulators, combining a mobile base with a robotic arm, offer enhanced workspace and dexterity for applications like warehouse automation. However, controlling these systems presents significant challenges. The primary difficulties lie in decomposing tasks into coordinated motions for both the mobile base and the manipulator, managing the nonholonomic constraints of the mobile platform, and handling the dynamic coupling between the base and the arm. These issues can lead to degraded performance, errors, and instability, especially in unstructured environments where sensing might be limited. Differential drive bases, in particular, introduce complexities due to their inherent kinematic singularities and sensitivity to wheel velocity variations. This paper aims to provide a robust control framework to overcome these challenges for a specific mobile manipulator system.

Methods

The authors developed a multi-module approach for controlling the Fetch robot:

  • System Dynamics: They derived the robot's dynamic model using Denavit-Hartenberg (DH) parameters and Lagrange's method. This involved calculating the inertia (M), Coriolis (C), and gravity (G) matrices to represent the system's equations of motion.
  • Manipulator Trajectory Planning: For point-to-point motion of the manipulator joints, they employed a quintic polynomial trajectory planner. This method was chosen to ensure smooth motion by constraining position, velocity, and acceleration, thus avoiding jerky movements.
  • Mobile Base Trajectory Planning: The A* path planning algorithm was used for the mobile base. They explored both 4-connected and 8-connected grid spaces with Manhattan and Euclidean heuristics, finding the 8-connected Euclidean heuristic to be superior for their simulated environment.
  • Inverse Kinematics (IK): They utilized the Trac_ik package, an open-source ROS implementation that enhances the KDL pseudoinverse Jacobian solver. Trac_ik was chosen for its improved performance, ability to handle joint limits, and higher success rate compared to other solvers.
  • Finite State Machine (FSM): A state machine was implemented to orchestrate the pick-and-place task, managing transitions between different states like planning, grasping, and moving based on robot status and task completion.
  • PID Control: A Proportional-Integral-Derivative (PID) controller was used for both the manipulator joints and the mobile base to ensure they accurately follow the planned trajectories.

Datasets

Not discussed in the paper. The authors used a simulated environment (Gazebo) and the Fetch robot's URDF model.

Key Insights and Ideas

  • Quintic Polynomials for Smoothness: Using quintic polynomials for manipulator trajectory planning is crucial for avoiding jerky movements (impulsive jerk) that can arise from discontinuities in acceleration, leading to better system stability and performance.
  • A* for Mobile Base Navigation: The A* algorithm, particularly with an 8-connected Euclidean heuristic, provides an effective way to find optimal paths for the mobile base in a grid-based environment.
  • Trac_ik's Superiority: The Trac_ik solver demonstrated significantly better performance (faster solution times and higher success rates) compared to classical IK solvers like Orocos KDL for the Fetch robot's kinematics.
  • Integrated Control: The successful integration of separate planning modules for the manipulator and mobile base, orchestrated by a state machine, is key to achieving complex manipulation tasks.
  • Simulation as a Viable Testbed: Gazebo provided a robust platform for developing and testing the entire control system before potential hardware deployment.

Key Results

The paper reports successful implementation of trajectory control for both the mobile base and the manipulator of the Fetch robot in a simulated environment. They demonstrated the ability to perform a pick-and-place task. Specific quantitative results mentioned include:

  • Trac_ik solver took an average of 0.44ms to find an IK solution over 100 trials, compared to Orocos KDL's 0.72ms.
  • Trac_ik had a higher success rate than Orocos KDL.

Qualitative results include the successful execution of the pick-and-place task, with figures showing generated joint trajectories, velocity, and acceleration profiles, as well as the planned path for the mobile base.

Limitations and Future Work

The authors acknowledge several limitations and propose future directions:

  • Lack of Global Object Localization: The current system requires pre-defined object locations. Future work aims to develop a global planner that can autonomously locate objects and determine base and end-effector movements.
  • Separate Planning: The mobile base and manipulator are currently planned separately. The goal is to develop a unified planner for both.
  • Hardware Implementation: The current work is limited to simulation. Future plans include implementing the system on an actual Fetch robot in an indoor warehouse environment.
  • Computer Vision Integration: Incorporating computer vision techniques is proposed to enhance the power and robustness of the state machine.
  • Modularity: Making the system modular would allow for easier porting to different robotic platforms.
  • Open-Source Contribution: The authors express a desire to make their trajectory control package open-source, noting the lack of robust, generic packages currently available.

Summary updated

BibTeX
@misc{karunakaran2023trajectory,
  author = {Harish Karunakaran and Gopeshh Raaj Subbaraj},
  title = {{Trajectory Control for Differential Drive Mobile Manipulators}},
  year = {2023},
  eprint = {2304.00122},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2304.00122},
}
Figure