KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#dsh #ai-agents #dsh-plugin #codex #developer-tools #claude-code #windows #desktop-app #typescript #rust #ai-agent #cli

arm-safety

★ 10 Open GitHub ↗

Code for "Multi-layered Safety of Redundant Robot Manipulators via Task-oriented Planning and Control" (ICRA 2025)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Multi-layered Safety of Redundant Robot Manipulators via Task-oriented Planning and Control

This repo contains the source code for implementing *multi-layered safety of robot manipulators*. This work has been accepted for presentation at the 2025 IEEE International Conference on Robotics and Automation (ICRA). You can find the full text [here](https://arxiv.org/abs/2410.17742).

* Static Obstacles (S1+S2)
<div align="center">
<img width="350" src="docs/sim_static.gif">
<img width="350" src="docs/static.gif">
</div>

* Dynamic Obtacles (S1+S2)
<div align="center">
<img width="350" src="docs/sim_dynamic.gif">
<img width="350" src="docs/dynamic.gif">
</div>

* Unknown Contacts (S1+S2+S3)
<div align="center">
<img width="350" src="docs/sim_unknown.gif">
<img width="350" src="docs/unknown.gif">
</div>

# Dependencies
The software has been tested with ROS Noetic under Ubuntu 20.04.

* ROS Noetic

* Eigen3 for linear algebra

* [Panda Simulator](https://github.com/justagist/panda_simulator) for Gazebo simulation

* [Pinocchio](https://github.com/stack-of-tasks/pinocchio) v3.4.0 with [HPP-FCL](https://github.com/humanoid-path-planner/hpp-fcl) v3.0.0 support for robot and environment modeling

* [CasADi](https://web.casadi.org/) with [IPOPT](https://github.com/coin-or/Ipopt) or [qpOASES](https://github.com/coin-or/qpOASES/tags) support for nonlinear/quadratic programming problems

* [Crocoddyl](https://github.com/loco-3d/crocoddyl) v3.0.0 for differential dynamic programming problems (optional)

* [ViSP](https://visp.inria.fr/) and a RealSense D435i camera for visual feedback in the real world (optional)

* Franka Panda robot for real-hardware validation (optional)

Recommend to install the dependencies from source according to the official instructions. For your convenience, we provide a brief [installation guide](https://github.com/jia-xinyu/arm-safety/tree/master/docs/README.md) for CasADi, Pinocchio, and Crocoddyl.

# How to Run

1. Clone the repo and build:
```
cd ${your catkin workspace}/src
git clone https://github.com/jia-xinyu/arm-safety.git
cd ..
catkin_make
```

2. Launch the Gazebo simulator:
```
source devel/setup.bash
roslaunch arm-safety demo_gazebo.launch
```
For hardware experiments, add the tags `hw:=true` and `robot_ip:=${your robot ip}` to the launch line.

3. Open a new terminal and launch the controller:
```
source devel/setup.bash
roslaunch arm-safety demo_mpc.launch
```
The launch file runs a simple obstacle avoidance case `0_evaluation` by default. We provide four other cases: `1_cabinet`, `2_moving_ball`, `3_contact`, `4_balls`, corresponding to **static obstacles**, **dynamic obstacles**, **unknown contacts**, and **multiple obstacles** respectively. Run them with `traj:=x_xxxxxx` tag.

4. (Optional) To simulate unknown contacts, run the following command in a new terminal:
```
source devel/setup.bash
roslaunch arm-safety demo_contact.launch
```
Collision reaction capability (S3) is activated by `if_safe_contact :1` flag in `hw_sim_mpc.yaml`.

# More Controllers
The repo provides several optimization-based control schemes, including 5 model predictive controllers (MPC) and 2 inverse kinematics (IK) solvers, as listed in the table below.

| File | Model | Obstacle | Space | Solvers | Features |
| ---- | ----- | -------- | ----- | ------ | ------------------ |
| `admmMPC.cpp`     | dyn | ✔ | task | **Crocoddyl, qpOASES** | Alternating Direction Method of Multipliers (ADMM) |
| `fddpMPC.cpp`     | dyn | ✔ | task | **Crocoddyl** | Differential Dynamic Programming (DDP) |
| `dynMPC.cpp`      | dyn |   | joint/task | **IPOPT** | Nonlinear Programming (NLP)  |
| `safeTaskMPC.cpp` | kin/dyn | ✔ | task   | **IPOPT** | NLP, shooting, task-oriented |
| `kinMPC.cpp`      | kin |   | joint/task | **IPOPT** | NLP   |
| `safeIK.cpp`      | kin | ✔ | task | **qpOASES** | Quadratic Programming (QP)       |
| `safeQP.cpp`      | kin | ✔ | task | **qpOASES** | QP, no CasADi |

* All control commands are finally converted into **torque** and sent to motors.

* `admmMPC.cpp` is a partial implementation of the ADMM-based MPC in the [RA-L paper](https://ieeexplore.ieee.org/abstract/document/11031218).

* `safeTaskMPC.cpp` is an implementation of the task-oriented MPC planner in the [ICRA paper](https://arxiv.org/abs/2410.17742).

* We give minimal test examples for these controllers. Run the command:
```
rosrun arm-safety test_common
```

# Citation
If you find this repo useful, please cite:
```bibtex
@inproceedings{jia2025safety,
  author={Jia, Xinyu and Wang, Wenxin and Yang, Jun and Pan, Yongping and Yu, Haoyong},
  booktitle={2025 IEEE International Conference on Robotics and Automation (ICRA)}, 
  title={Multi-Layered Safety of Redundant Robot Manipulators Via Task-Oriented Planning and Control}, 
  address={Atlanta, GA, USA},
  year={2025},
  volume={},
  number={},
  pages={14338-14344}}
```
or
```bibtex
@article{jia2025admm,
  author={Jia, Xinyu and Wang, Wenxin and Yang, Jun and Pan, Yongping and Yu, Haoyong},
  journal={IEEE Robotics and Automation Letters}, 
  title={ToMPC: Task-Oriented Model Predictive Control via ADMM for Safe Robotic Manipulation}, 
  year={2025},
  volume={10},
  number={8},
  pages={7939-7946}}
```

arm-safety

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README