Human-Robot Co-Transportation using Disturbance-Aware MPC with Pose Optimization

Al Jaber Mahmud, Amir Hossain Raj, Duc Minh Nguyen, Xuesu Xiao, Xuan Wang, Wang, XuanarXiv 2024 · arXiv

TL;DR

This paper introduces a novel control strategy for human-robot co-transportation that uses Model Predictive Control (MPC) enhanced with a pose optimization mechanism. This approach effectively mitigates disturbances, like human unpredictability, by dynamically adjusting the robot's arm configuration to minimize predicted costs, leading to more stable and efficient collaboration.

What Problem Does This Paper Address?

Collaborative tasks between humans and robots, such as co-transportation, are crucial for reducing human workload. However, these tasks are inherently challenging due to unpredictable human behavior and potential robot actuation noise. Existing control methods often struggle to adapt to these disturbances, leading to reduced efficiency and safety. Furthermore, the complex whole-body dynamics of mobile manipulators (combining a mobile base and a robotic arm) add another layer of difficulty. This paper aims to address these challenges by developing a control system that can robustly track desired trajectories while actively compensating for disturbances and optimizing the robot's own configuration.

Methods

The core of the proposed solution is a Disturbance-Aware Model Predictive Control (MPC) with Pose Optimization. This approach builds upon standard MPC by incorporating two key innovations:

  • Disturbance Awareness: The MPC formulation explicitly accounts for disturbances, modeled as Gaussian noise, which can represent human uncertainty or robot actuation noise. This is achieved by using a disturbance-aware Discrete Algebraic Riccati Equation (DARE) to estimate the optimal control cost.
  • Pose Optimization: Recognizing that a robot arm often has redundant joint configurations for a given end-effector pose, the system dynamically selects the "best" joint angle configuration (pose) from a candidate set. This selection is driven by the goal of minimizing the estimated future control cost, effectively choosing a pose that is more robust to disturbances or leads to lower energy consumption.

The overall control strategy operates in a two-step iterative design:

  1. Cost Estimation: For a given planning horizon, the algorithm first estimates the optimal control inputs and the associated cost-to-go for various candidate robot arm poses, considering the predicted disturbances.
  2. Pose Selection: The algorithm then selects the joint angle configuration (pose) that yields the lowest estimated cost. The optimal control inputs are then computed based on this selected pose.

The paper also details the whole-body kinematics of the Fetch robot, combining the mobile base and the 7-DOF robotic arm into an 8-DOF system for control.

Datasets

The paper does not explicitly mention training on specific datasets. Instead, it relies on:

  • Simulated Experiments: The algorithm is evaluated in Gazebo simulations.
  • Hardware Demonstrations: The approach is validated on a Fetch robot.

The "datasets" used for evaluation are generated trajectories with simulated human uncertainties, where the uncertainty is modeled as Gaussian noise with a specified covariance matrix.

Key Insights and Ideas

  • Redundancy as an Advantage: The paper leverages the inherent redundancy in robotic arms not just for reaching targets but as a strategic tool to mitigate disturbances and reduce control costs.
  • Proactive Disturbance Mitigation: Instead of just reacting to disturbances, the system proactively chooses a robot configuration that is expected to lead to better performance in the presence of these disturbances.
  • Dual-Phase Optimization: The separation of control input optimization and pose optimization into a two-phase process provides a tractable way to solve a complex problem that would be computationally prohibitive if optimized simultaneously.
  • Theoretical Foundation for Cost: The use of a disturbance-aware DARE provides a theoretical basis for estimating the cost-to-go, which is crucial for making informed pose selection decisions.
  • Hardware Feasibility: The successful demonstration on a Fetch robot validates the practical applicability of the proposed method.

Key Results

  • Improved Tracking Performance: The proposed PO-HU (Pose Optimization - Human Uncertainty) algorithm consistently outperformed baseline algorithms (NPO-HU: No Pose Optimization - Human Uncertainty, and PO-NHU: Pose Optimization - No Human Uncertainty) in terms of total cost (Ctotal) across various trajectories and disturbance levels.
  • Significant Cost Reduction: In complex scenarios, especially with predicted human future poses being less accurate, the PO-HU method showed significantly lower accumulated costs compared to baselines. For instance, in trajectory Bp, the PO-NHU algorithm's cost could "explode" due to a bad pose, while PO-HU remained stable.
  • Pose Optimization's Impact: The results suggest that pose optimization has a greater impact on reducing total cost than simply characterizing human uncertainty.
  • Periodic Optimization Viability: A periodic pose optimization (pPO-HU) strategy, performed every 5 time steps, achieved performance close to optimizing at every step, offering a trade-off between computational load and performance.
  • Computational Efficiency: The execution time analysis showed that the proposed algorithm's overhead for pose optimization is manageable, especially when steps can be parallelized, and it does not introduce significant extra computation compared to baselines.

Limitations and Future Work

  • Candidate Set Generation: The paper uses a sample-based approach to generate candidate poses for optimization. The effectiveness of this approach depends on the size and quality of the candidate set.
  • Human Uncertainty Modeling: While the paper models human uncertainty as Gaussian noise, real-world human behavior can be more complex.
  • Generalization: The current work focuses on co-transportation.

Future Work: * Generalization to Multi-Human/Multi-Robot Systems: Extending the algorithm to handle more complex collaborative scenarios involving multiple agents. * Leveraging Deep Learning: Exploring deep learning methods to dynamically determine when pose optimization is needed and how to efficiently select candidate poses. * More Sophisticated Human Models: Incorporating more advanced models of human behavior and uncertainty.

Summary updated

Abstract

This paper proposes a new control algorithm for human-robot co-transportation using a robot manipulator equipped with a mobile base and a robotic arm. We integrate the regular Model Predictive Control (MPC) with a novel pose optimization mechanism to more efficiently mitigate disturbances (such as human behavioral uncertainties or robot actuation noise) during the task. The core of our methodology involves a two-step iterative design: At each planning horizon, we determine the optimal pose of the robotic arm (joint angle configuration) from a candidate set, aiming to achieve the lowest estimated control cost. This selection is based on solving a disturbance-aware Discrete Algebraic Ricatti Equation (DARE), which also determines the optimal inputs for the robot's whole body control (including both the mobile base and the robotic arm). To validate the effectiveness of the proposed approach, we provide theoretical derivation for the disturbance-aware DARE and perform simulated experiments and hardware demos using a Fetch robot under varying conditions, including different trajectories and different levels of disturbances. The results reveal that our proposed approach outperforms baseline algorithms.
BibTeX
@misc{mahmud2024human,
  author = {Al Jaber Mahmud and Amir Hossain Raj and Duc Minh Nguyen and Xuesu Xiao and Xuan Wang and Wang, Xuan},
  title = {{Human-Robot Co-Transportation using Disturbance-Aware MPC with Pose Optimization}},
  year = {2024},
  eprint = {2404.00514},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2404.00514},
}
Figure