ICRA 2025: Multi-Nonholonomic Robot Object Transportation with Obstacle Crossing using a Deformable Sheet

Weijian Zhang, Charlie Street, Masoumeh Mansouri

University of Birmingham

Overview

We address multi-robot formation planning where nonholonomic robots collaboratively transport objects using a deformable sheet in unstructured, cluttered environments. Our paper has been selected as the Best Conference Paper Award Finalist & Best Paper Award Finalist on Multi-Robot Systems at ICRA 2025!

Features

• A heuristic path exploration method that efficiently evaluates a set of homotopically distinct solution spaces for the formation.

• A two-stage iterative motion planning framework for finding locally time-optimal collision-free formation trajectories using a deformable sheet.

Requirements

• ROS Noetic or later
• Ubuntu 20.04 or later
• yaml-cpp 0.8.0 or later
• You'll also need a license for the Mosek optimization toolbox https://www.mosek.com/ (this package includes a downloader for the Mosek code, but you have to get your own license). Mosek has free licenses available for academic use.

Installation

1. Create a new workspace:

$ mkdir -p ~/CPDOT/src
$ cd ~/CPDOT/src
$ catkin_init_workspace

2. Clone the package into the workspace:

$ git clone git@github.com:HyPAIR/CPDOT.git

3. Install dependencies:

$ cd ~/CPDOT
$ rosdep install --from-paths src --ignore-src -r -y

4. Build the workspace (Set MOSEK_DIR to the root path of your MOSEK license (e.g., /home/yourname/mosek/7):

$ catkin_make --cmake-args -DMOSEK_DIR=/home/yourname/mosek/7

Parameter values in the simulation

Parameter	Value	Description
$L$	$0.65m$	Car-like robot wheelbase
$v_{m}^{car}$	$1.0m/s$	Linear velocity limit
$a_{m}^{car}$	$1.0m/s^2$	Linear acceleration limit
$\phi_{m}^{car}$	$0.68rad$	Steering angle limit
$\omega_{m}^{car}$	$0.2rad/s$	Angular velocity limit
$\Omega_{m}^{car}$	$2.5rad/s^2$	Angular acceleration limit
$-$	$0.2m$	IRIS grid size
$\Delta{t}$	$0.15s$	Time between control inputs
$\mathbf{W}$	$[2, 0; 0, 1]$	Weights for cost function
$l^{max}_i$	$2.0m$	Original sheet's side length
$z_r$	$1.5m$	Height of each contact point
$d_0$	$1.5m$	Default inter-robot distance
$\delta$	$0.2m$	Delta value on inter-robot distance
$\lambda_1$	$1.0$	Weight parameter
$\lambda_2$	$1.0$	Weight parameter
$\lambda_3$	$1.0$	Weight parameter
$\lambda_4$	$1.0$	Weight parameter

Test in Rviz

Launch the simulation to the trajectory optimisation result (4 robots in a simple scenario):

$ roslaunch formation_planner topological_test.launch

Test in Gazebo

Create a world file with 5 car-like robots in 100 random obstacle environments.

$ roslaunch formation_planner write_obs_to_world.launch

Launch a multi-robot transportation simulation, with 5 car-like robots in 100 random obstacle environments.

$ roslaunch formation_planner heterogeneous_triangle.launch

Launch the control node:

$ roslaunch formation_planner control_triangular_formation.launch

Video

A simulation and real-world experiments video demonstrating our proposed framework can be found at bilibili/youtube.

Citation

If you find this work useful, please cite (paper):

@INPROCEEDINGS{11128313,
  author={Zhang, Weijian and Street, Charlie and Mansouri, Masoumeh},
  booktitle={2025 IEEE International Conference on Robotics and Automation (ICRA)}, 
  title={Multi-Nonholonomic Robot Object Transportation with Obstacle Crossing Using a Deformable Sheet}, 
  year={2025},
  volume={},
  number={},
  pages={7349-7355},
  keywords={Limiting;Navigation;Transportation;Probabilistic logic;Hardware;Planning;Iterative methods;Robots;Trajectory optimization;Contracts},
  doi={10.1109/ICRA55743.2025.11128313}}