Manipulability optimization for multi-arm teleoperation

Florian Kennel-Maushart, Roi Poranne, Stelian CorosICRA 2021 · DOI

TL;DR

This paper introduces a novel VR-based teleoperation system for multi-arm robots that improves manipulability by allowing a single rotational degree of freedom to be optimized locally for each arm. This approach helps avoid kinematic singularities and workspace limitations, leading to smoother and more reliable control, especially in complex manipulation tasks.

What Problem Does This Paper Address?

Teleoperation, the remote control of robots, is a powerful tool for tasks where full autonomy is challenging or human intervention is crucial. With the rise of collaborative robots and VR technology, teleoperation is becoming more accessible. However, controlling multiple robot arms simultaneously, especially when they have different kinematic structures or are manipulating a shared payload, presents significant challenges. A key issue is avoiding "singular configurations" – robot arm poses where movement becomes unpredictable or impossible. Naively mapping human motion to robot arms can easily lead to these configurations, causing undesirable behavior and hindering task completion. This paper specifically addresses the problem of maintaining robot manipulability (the ability to move freely in all directions) during multi-arm payload manipulation, particularly in real-time teleoperation scenarios where human operators might not be aware of the robot's kinematic limitations.

Methods

The core of the proposed method lies in an optimization-based Inverse Kinematics (IK) solver. Instead of strictly enforcing all end-effector pose constraints, the system relaxes one rotational degree of freedom for each robot arm. This relaxation allows for a local optimization step that aims to maximize the manipulability index of each individual arm.

Here's a breakdown of the key techniques:

  • Optimization-based IK: The initial IK problem is formulated to minimize end-effector position error and joint regularization, while respecting joint limits.
  • Local Manipulability Optimization: After the primary IK solution is found, a secondary optimization is performed. This step uses the generalized inverse of the velocity Jacobian to compute small adjustments to joint angles. These adjustments are specifically designed to rotate around a chosen axis (e.g., the handle's local x-axis) to improve the manipulability index.
  • Thresholding and Prior Configuration: To prevent oscillations and ensure stability, the manipulability improvement is only accepted if it exceeds a certain threshold. The previously computed optimal joint angles are used as the initial guess for the next iteration, promoting smooth transitions.
  • VR Interface: A user-friendly VR interface is developed, allowing operators to intuitively grasp and move the payload in a virtual environment. This interface translates the operator's movements into commands for the robot system.

Datasets

Not discussed in the paper. The experiments were conducted using real robot setups and predefined or recorded teleoperation trajectories.

Key Insights and Ideas

  • Relaxing Constraints for Better Performance: The idea of relaxing a single DOF to improve manipulability is a clever way to balance precise control with robustness. It acknowledges that for many tasks, perfect end-effector pose is less critical than smooth, predictable motion.
  • Local Optimization for Multi-Arm Systems: Applying manipulability optimization locally to each arm, rather than globally for the entire multi-arm system, simplifies the problem and makes it computationally tractable for real-time applications.
  • VR as an Intuitive Control Interface: The use of VR significantly enhances the operator's ability to control complex multi-arm systems, making it easier to manage the payload and implicitly guide the robots.
  • Graceful Degradation: The method aims for "graceful degradation" – when approaching limitations, the system doesn't abruptly fail but rather shows a predictable decline in performance, allowing the operator to adapt.

Key Results

The paper demonstrates the effectiveness of their method across various scenarios and robot platforms (UR5 and ABB YuMi).

  • Improved Manipulability: In all tested scenarios (circle, square, and teleoperation trajectories), the proposed method consistently improved the manipulability index compared to a fully constrained IK approach. For example, in the circle trajectory, mean manipulability was significantly higher with the proposed method.
  • Reduced Position Error: For the UR5 robots, the method led to a reduction in mean position error, particularly noticeable in the circle and teleoperation trajectories. While YuMi robots showed higher positional errors overall, the proposed method still offered improvements.
  • Smoother Motion: The method generally resulted in lower velocity, acceleration, and jerk values, especially for the YuMi robots, indicating smoother and more controlled movements.
  • Computational Efficiency: The method adds only a small overhead (around 0.58ms) to the computation time per frame, allowing for near real-time performance (around 144 fps).

Limitations and Future Work

  • Collision Avoidance and Re-grasping: The current work focuses on manipulability optimization and does not explicitly address collision avoidance or dynamic re-grasping strategies, which are important for complete robotic systems.
  • Task Allocation and Dynamics: The paper mentions that future work could explore how to leverage the redundancy of multi-arm systems for dynamic re-grasping or task allocation, and how to incorporate robot dynamics.
  • Simple Payload Attachment: The payload was attached with simple handles. More complex payloads or attachment mechanisms might require further investigation.
  • Operator Adaptation: While the method improves robot behavior, further research could explore how to better adapt the system to operator behavior and preferences.
  • Mobile Bases: The authors note that their method can be extended to robots on mobile bases by incorporating the base's DOFs into the optimization.

Summary updated

BibTeX
@inproceedings{kennelmaushart2021manipulability,
  author = {Florian Kennel-Maushart and Roi Poranne and Stelian Coros},
  title = {{Manipulability optimization for multi-arm teleoperation}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2021},
  doi = {10.1109/icra48506.2021.9561105},
}
Figure